For more than a decade, Uber built one of the most valuable transportation platforms in the world by redefining labor as software-mediated flexibility. Drivers supplied the vehicles, absorbed the depreciation, paid for fuel, maintenance, insurance, financing, downtime, and risk, while the tech platform did the coordination. But now, as autonomous vehicles begin to mature commercially, even as it continues to do business with many AV companies, Uber has repositioned itself as a defender of drivers and equitable mobility, with stated concerns about the impact of automation on workers. This narrative deserves scrutiny.

Uber’s recent white paper, “Unlocking the Promise of Autonomy,” presents the company as a pragmatic middle ground between rapid automation and social responsibility. The report argues for a “hybrid” transportation future where human drivers and autonomous vehicles coexist and warns against a future in which robotaxis create unequal systems of mobility. But the argument is difficult to separate from Uber’s own strategic vulnerability and efforts to slow Waymo and Tesla, as signaled by their recent divorce from the company in Phoenix. Research showing that 55% of AV rides come directly from people who otherwise would have taken a rideshare trip indicates that efforts to slow down automation may be more about platform preservation than social good1.

And more importantly, the irony is hard to ignore: the same company that normalized gig work is now not only glamorizing the quality of that work, but framing the preservation of it as a form of worker protection.

The contradiction becomes even more egregious when viewed alongside Uber’s broader strategy to launch its own autonomous vehicles and avoid liability from things like assaults and sexual harassment from its non-employee driver-partners. The company has aggressively expanded partnerships and investments in autonomous mobility while simultaneously signaling caution about displacement and over-automation. It has done things like launching “Uber Autonomous Solutions” (a division specifically designed to accelerate commercialization of autonomous fleets through financing, fleet operations, support infrastructure, and platform integration), and putting more than $10 billion toward autonomous partnerships with Rivian, Nuro and Lucid.

Andrew Macdonald, Uber’s President and COO, recently stated that the technical hurdles for AVs have “largely been solved” and that commercialization is now the central challenge. At the same time, Uber executives have also been publicly criticizing AV-only operators like Waymo as inequitable, less scalable, or disconnected from broader transportation needs.

Given all of these signals, it starts to look like Uber is using the same old strategy they have employed for years to avoid regulation and beat out competition. And in this case they want to occupy both sides of the argument. They say they want to scale automation, but also protect society from it. They say they want to protect labor while exploiting it.

And this labor contradiction is the core issue. The uncomfortable reality is that Uber’s rideshare model is built on a deeply asymmetrical labor structure. Research going back to 2018 suggested that Uber and Lyft drivers only take home $3.37 per hour after taking into account vehicle costs. While these numbers were revised to a median of $8.55 after receiving pressure from the company and a Medium post from Uber’s Chief Economist, the fact remains—the majority of drivers are never compensated for the full operating cost of their vehicles.

Harry Campbell, Founder of The Rideshare Guy, validated this for me in one of our recent discussions, saying:

Uber shifts a lot of the burden of vehicle maintenance and insurance to unsuspecting drivers. For example, drivers can sign up to drive with personal insurance but there’s a coverage gap during Period 1 (driver online, but waiting for a request). If they get into an accident, many are surprised to find out that there’s a $2,500 deductible which is much higher than the average personal insurance deductible.

Rideshare platforms externalize large portions of the operational expense of vehicles; placing these costs onto workers themselves. This includes maintenance, insurance, idle time, fuel costs, financing, depreciation, and more… while Uber extracts monetary value from its app, at a fair “take rate” of more than 50%—meaning that the driver often gets less than 50% of what the rider pays. While these figures continue to be disputed by Uber, when factored alongside increases in costs related to vehicle operations in recent years, it is likely that raw take-home wages for drivers is lower than the $30/hour median that Uber claims–and this may be much higher that actual driver earnings because it does not account for down time when drivers are online but waiting for a ride.

Comparing this to Waymo and other vertically integrated AV operators, there are key structural differences. Most importantly, the operational costs are internalized into the service itself. The vehicle, software stack, fleet maintenance, sensors, remote operations, and liability structure are all folded into the economics of the ride. The fares increasingly reflect the real or actual cost of delivering a ride to a customer.

That distinction is important. Rideshare platforms have historically obscured the true economics of transport because the drivers themselves absorbed the substantial share (if not all of) the operational cost and financial burden for the vehicle. The drivers, in essence, subsidize the ride. The low apparent price of many trips is a product of drivers using their own vehicles (assets Uber or Lyft do not own or maintain) and agreement to precarious labor standards. In 2025 Human Rights Watch provided research that showed that 67% and 75% of rideshare drivers struggled to afford food and housing respectively. Many were subject to accidents and exposure to violent crime. In the most simple terms autonomous vehicles threaten to expose the economic exploitation and poor workplace conditions. They reveal the resource that rideshare companies have long exploited—drivers.

For years the growth of the rideshare industry has been subsidized by privately owned vehicles and contingent labor. [Note that in most cases rideshare drivers are independent contractors that receive limited benefits.] I have argued before that the concept of “human dignity” in labor matters, and that gig work is a toxic labor system that perpetuates a “Dickens Economy” that does not value labor fairness, workplace safety or health. The same Human Rights Watch report in 2025 found that 50% of drivers experienced wage instability week-to-week and a shocking 70% were worried about being deactivated or suspended by app algorithms. This is confirmed by medical journal articles identifying occupational health issues related to gig driving and other peer reviewed work that underscores the “lack of social protections, algorithmic control, intense competition, and downward pressure on wages.” This is important in the current context as Uber begins to reposition itself as a defender of drivers, reframing gig work as socially beneficial employment worth preserving against automation and autonomous vehicles.

In evaluating this position (that we should preserve the value of gig labor) however, we should be asking honest questions: what exactly are we preserving? Gig drivers lack long-term wage stability, healthcare, retirement benefits, predictable scheduling, labor protections, or meaningful economic mobility. And Uber has made clear that drivers are independent contractors and not employees so that they could avoid awarding corporate benefits to their large “driver partner” labor pool. Yes, it may be an easy job. Yes, it may be a job that is flexible and easy to access. But that doesn’t mean that gig driving jobs are somehow worth retaining or that the ride-hailing labor model is some kind of social compact we made to preserve low-skilled driving. And this is what Uber is trying to position; that gig jobs are dignified labor that should be protected. But consumers should not be deceived; this is not worker advocacy—it’s self-serving as the company is aggressively pursuing its own self-driving car initiatives.

Given that it’s not clear that companies like Uber have a legitimate claim to push back against automated vehicles when they are ending their partnership-with and looking to compete-with Waymo, and after more than a decade of aggressively restructuring transport jobs into a gig economy that supports precarious and unstable work conditions for drivers. This is not being a principled guardian of jobs or transportation equity; it is simply a ploy to slow down other autonomous vehicle companies, while they scale their own technology, reducing their reliance on gig labor. Put simply, to redefine low-quality, contingent, “gig” labor as socially valuable employment at precisely the moment you are trying to reduce reliance on this form of labor, and at the exact time when automation promises to unlock economic advantages, allowing for upskilling and retraining, is disingenuous.

And all this is not to argue that job displacement concerns in an autonomous vehicle future are unimportant. Technological transitions driven by AI and automation will undoubtedly produce disruptions for workers and communities. History suggests as much. But history also suggests these transitions rarely unfold as simple stories of mass unemployment. Most credible research now points toward widespread job transformation, gradual displacement, and the creation of entirely new categories of work alongside productivity gains. For example, The World Economic Forum projects roughly 92 million jobs displaced but 170 million created, for a net gain of about 78 million jobs by 2030. The Chamber of Progress estimates that, “for every 1,000 AVs produced and deployed annually, approximately 190 workers will be needed for manufacturing and servicing these vehicles.”

Yes, some of the concerns Uber highlights are very real government questions that need honest dialogue and decisive policy: from curb management and accessibility, to public transit integration, and equitable deployment. Uber is right that companies and policymakers absolutely should think seriously about workforce transition, retraining, and economic adaptation as well as the build-out. But after years of normalizing the gig economy, arguments about labor feel more self-serving than social responsibility. And the conversation about worker transitions in light of automation deserves more attention than to romanticize today’s gig economy.

If Uber wants to lead the next chapter of mobility, it should do so honestly: not by romanticizing the labor conditions that made its original business model possible, but by acknowledging the economic contradictions that autonomy is beginning to expose. They should be focused on retraining and upskilling, not keeping the same old systems that hold workers back. The future transportation debate should not be framed as humans versus robots. It should be framed around labor transition and how automation helps reshape the dignity of work in mobility.

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, I’m speaking with Don Burnette, founder and CEO of Kodiak AI. Don shares what has kept him motivated after more than a decade in the autonomous vehicle industry, why he remains confident in the future of AVs, and how the space has evolved from years of R&D into real-world commercialization.

We also dive into what Kodiak AI is building and its vehicle-agnostic physical AI approach, which spans long-haul trucking, industrial applications, defense, and more. Don explains why long-haul trucking remains such a large opportunity, what it will take to reach fully driverless operations at scale, and why highway-only autonomy isn’t enough to make the product viable.

Finally, we discuss the differences between highway and city-street autonomy, how Kodiak approaches proving safety, and what it has learned from earlier deployments. Don also shares his perspective on scaling KPIs, the competitive landscape, Kodiak’s experience as a public company, and whether autonomous trucking could ultimately be a larger market than robotaxis.

Chapters

• (00:00) Introduction to Don Burnette

• (01:30) What’s kept Don going in the AV space for over a decade

• (02:15) Why Don is confident in the future of AVs, and some of the challenges the industry faces today

• (04:09) How AVs have moved from R&D to commercialization

• (06:30) What Kodiak AI does

• (07:55) Kodiak AI’s generalized physical AI and how it’s vehicle agnostic

• (10:07) Kodiak AI’s verticals: including long-haul trucking, industrial trucking and defense

• (18:03) The role of robotaxis in propagating autonomy to the general public

• (19:50) Why long-haul trucking has a bigger total addressable market than industrial trucking

• (21:30) Steps to reaching full driverless operations in long-haul trucking, and how Kodiak AI is approaching it

• (23:44) Why highway-only autonomy isn’t enough

• (25:10) Highway vs. city-street autonomous driving

• (26:59) Public perception of accidents and proving the safety case in AV trucking vs. robotaxis

• (30:20) Learning from the experiences of earlier, broader AV deployments

• (33:10) Experience operating as a publicly listed company

• (34:50) KPIs for autonomous truck deployments and which ones Kodiak AI focuses on

• (37:42) Competition in the AV trucking space and Kodiak AI’s position

• (39:25) How individual incidents affect the general perception of autonomy

• (41:50) Is autonomous trucking’s total addressable market bigger than robotaxi’s?

• (43:41) Conclusion and final thoughts

Notes/Links:

• You can find Don Burnette on LinkedIn and X.

• Kodiak AI’s website (link).

-Harry

Top Story of the Week

Uber Exits Serve Robotics Stake as Delivery Alliance Unravels (link, no paywall). I was surprised by this one, since you don’t see many Uber partners walk away, especially smaller ones, and Serve was also born inside the Uber/Postmates ecosystem. Now Uber has sold its stake and Serve doesn’t plan to renew their agreement when it expires next year. The disagreement seems to be less about the robots themselves and more about how to actually operate and scale them. Interestingly, Serve says its experience with DoorDash has been much better, which shows that Uber’s aggregator model isn’t as simple as plugging autonomous supply into the platform. Uber does still have several other delivery robot partners, but with both Waymo and Serve moving away from Uber, it will be interesting to see if this becomes a broader trend.

“Our extensive discussions with Uber since the emergence of this trend in Q2 have clarified that we really have differing views about the operating model to scale our shared autonomous fleet,” Kashani said on the call, adding that those differences span areas like fleet coordination and merchant integration.

Other Stuff

Uber COO Andrew Macdonald: We’re teaming up with Hinomaru Kotsu as our fleet partner to launch our robotaxi pilot in Tokyo later this year with Wayve and Nissan (link). Not sure how I feel about this car.

Waymo expands in Phoenix area with new East Valley territory (link), they also expanded in Orlando to include more of Taft (link). Congrats JJ Ricks!

The company announced Thursday that it's adding 55 square miles of the East Valley to its service area. That includes Gilbert and more of Chandler.

Avride CEO Dmitry Polishchuk: Just crossed 100,000 autonomous rides with Uber in Dallas! (link). *With a safety driver but nice to see them progressing.

Waymo enters second phase of testing in Denver (link).

Waymo has moved to the next phase of testing in Denver by removing the backup drivers. Denver is an important testing ground for Waymo. It’s the company’s only test city that regularly sees snow and ice. The company hopes to eventually expand into the metro area and Denver International Airport.

Uber partners with China’s Pony.ai for 2,000 robotaxis in Europe (link). Europe still hasn't approved driverless robotaxis so while this is interesting, it doesn't mean much yet.

The companies plan to deploy 2,000 of Pony.ai’s self-driving taxis across Europe, and expand the robotaxi partnership to the Middle East, they announced Friday.

Designing Energy Infrastructure for Autonomous and Electric Fleets

MN8 Energy is one of the largest renewable energy and infrastructure owners in the U.S., supporting fleet operators as they scale electrification and autonomy. With expertise in delivering custom, turnkey solutions tailored to each fleet’s duty cycle and operational needs, bringing reliability, uptime guarantees, and long-term operational ownership, MN8 is the choice partner for fleets as they transition to electric and autonomous operations.

To learn more about MN8’s work with fleet and infrastructure partners, reach out to the Distributed Energy Solutions team at DES@mn8.com

Waymo registers Swiss entity in Zurich (link). Interesting how many new entities Waymo has been forming around the globe lately. Sure seems like they’re keeping their options open for a much broader international expansion.

Take a peek inside a Tesla Cybercab (link).

Waymo is proud to partner with the Las Vegas Raiders and Allegiant Stadium to bring autonomous rides to Las Vegas (link). Paint that Ojai black!

‘Former Waymo employee thoughts on ride-sharing’ (link). I don’t agree with the Uber analysis here at all. Air travel is a poor analogy since it’s a large, infrequent purchase. Ridehail is all about increasing frequency and building loyalty. Waymo still doesn’t have a great answer for peak demand, which is exactly where a broader marketplace like Uber can be valuable.

Imagry and UNVI Announce Partnership to Offer Fully Integrated Autonomous Driving Buses (link).

Together, the companies have integrated Imagry’s field-proven AI-based autonomous driving technology into UNVI’s M3-category electric buses to address the growing market demand for safe, reliable, and sustainable public transportation solutions.

‘Waymo soon to be available via the Lyft app in Nashville’ (link).

‘Nearly one year ago, Kodiak rang the Nasdaq Opening Bell as a newly public company. Since then, we’ve grown our customer-owned driverless fleet from 8 to 35 trucks, increased paid driverless operations from 3,000 cumulative hours to more than 40,000, and introduced the seventh generation of the Kodiak Driver’ (link). Had a great conversation with Kodiak CEO Don Burnette about this milestone and a lot more. Listen in on Monday when the episode goes live!

Zoox Safety Case Framework (link).

‘Lovely to see Waymo’s partnership with TechForce out in the wild’ (link). This is a neat partnership, and I’d love to see Waymo team up with Lyft on something similar in Nashville, where over 50% of the full-time roles Lyft has filled at its AV depot have gone to current or former Lyft drivers. Programs like this may only appeal to a relatively small share of drivers, but for the people who do participate, they can be hugely impactful by creating a real pathway into new AV-related jobs. Grab has done a good job here too, including training existing driver-partners to become safety operators through programs like GrabAcademy.

Teamsters sue California DMV over driverless truck rules (link). This seems like a weak argument considering the DMV spent years developing these rules, held multiple public comment periods and a hearing, and the Teamsters themselves participated in the process.

Filed Wednesday in Alameda County Superior Court, the lawsuit alleges the DMV failed to conduct a legally required economic review and did not give the public enough information about the rules' potential effects on jobs and safety.

Waymo back on freeways! My first ride (link).

The Fifth Layer: Why Infrastructure Is AV Rideshare’s Missing $200B Opportunity (link).

Waymo Factory August Update - 953 Ojai, 72% finished. + 12 IONIQ 5s (link).

What else we’re reading/listening to

Uber Says It’s Gaining Market Share from Waymo in San Francisco. I’m Not So Sure (link). My latest piece where I take a closer look at Uber’s claim that it’s gaining market share against Waymo in San Francisco, why Waymo’s expanding operating domain complicates the Yipit data, and what CPUC trip data tells us about whether Uber is actually taking share back from Waymo. It’s been generating some buzz on Reddit too (link).

Wayve’s GAIA-4 and the Future of AV Simulation with Brandon Basso (link). Our latest podcast where TDD Editor sits down with Wayve’s VP of Engineering to discuss how world models are being used to train and validate autonomous vehicles, what GAIA-4’s closed-loop simulation capabilities unlock, and the challenges of validating a generalizable AI driving system, among other things.

Autonomy Insiders by : The Hidden Human Layer Making Autonomous Vehicles Scalable | Nick Allen, TaskUs (link).

TITV by The Information, featuring Lyft CEO David Risher: Why Lyft is Betting Big on Waymo (link).

Related: Lyft’s Hybrid Approach to Autonomy with CEO David Risher (link).

30 Minutes by ADSN: Self-Driving Cars Are Here And What It Means for You ft. Harry Campbell (link).

The Curbivore by : Let Them Linger? The Surprising Case for Robotaxis at the Curb (link). Interesting study that looks at AV’s impact on VMT when they park after a ride vs cruising around. Also echoes some of the discussion I had with on a past podcast (link).

AVs/Humans behaving badly

Dallas pedestrian dies after being struck by SUV, thrown into Waymo (link). I saw some criticism of Waymo over this, but they were literally driving the other direction. The pedestrian was first hit by a human-driven SUV and thrown into the opposite lane before colliding with the Waymo. Hard to see how this is really a Waymo safety story.

‘Autonomous delivery van in China drives straight into ditch’ (link). Who left that gate open?

AV behaving “goodly” :)

‘Not all road users follow a predictable path’ (link).

Shout-outs

Big thanks to TDD readers at Moove, and to Jennifer P, Jake F, and Lars B for referring new subscribers. If there’s someone you think would enjoy TDD, feel free to forward this email or use the referral button below, and we’ll make sure to shout out your company.

Refer a friend

Neat Jobs

Product Manager, Driving Performance Expansion at Waymo (link) via Jacopo Sannazzaro.

Software Engineer - Driving Behavior at Zoox (link) via Mingfeng Zhang.

Sensing Systems Engineer at Wayve (link) via Irene Perali.

Director, Autonomy Perception at May Mobility (link) via May Mobility.

Job Moves

Congrats to friend of TDD Offer Grembek on the move from Kinetic Logic -> Atoms (link). Atoms sure is making a lot of moves lately!

Travis Kalanick’s robotics startup Atoms taps former Uber finance chief as CFO (link).

Omar Mahboob: General Motors -> Motional (link).

Scott Sheridan: Humanoid -> Wayve (link).

An Li: Meta -> NVIDIA (link).

Brianna Swanson: Ocado Technology -> Waabi (link).

We’re now at 11,000+ subscribers! Interested in advertising with TDD? Check out our media kit here and reach out to us. Newsletter spots are currently sold out until September but we do have podcast advertising opportunities available.

Reach out

Until next week :)

-Harry

It’s been a while since I’ve seen updated Yipit market share data for San Francisco, so it was nice of Uber’s CFO, Balaji Krishnamurthy, to share the latest numbers on Twitter recently, even if he was doing it to roast me.

Since Waymo, Uber and Lyft don’t share their city-by-city breakdown of rides publicly, Yipit data gives us one of the clearest looks at how the three are actually stacking up against each other within a given operating area.

Balaji’s reason for sharing the chart was to bolster Uber’s claim on their recent earnings call, that “in more mature AV markets including Los Angeles, San Francisco, and Phoenix, Uber’s category share, both citywide and in AV operating zones, is higher today than it was a year ago.”

Initially, I was skeptical of this claim since Waymo famously gained a ton of market share in San Francisco when they launched. But Uber is arguing that while Waymo may have gained market share quickly after launching in these three markets, it hasn’t continued taking share from Uber over the past year.

The citywide part of that claim isn’t especially surprising. Uber’s operating domain is much larger than Waymo’s, it can serve airports, and it has fewer geographic restrictions overall. And since Uber is still growing its ride volume every quarter, it’scplausible that its growth across the broader market could outpace Waymo’s growth within a much smaller operating area.

But the second part of Uber’s claim, that its market share has also increased within Waymo’s operating zones, is more interesting and deserves a closer look.

Zooming In

We covered Waymo’s rise from 0% market share in SF in August of 2023 to equaling Lyft’s market share of 20% in November of 2024.

Waymo continued on that growth trajectory and things were looking good in April of 2025. If you were to extrapolate their growth rate, Waymo was even ‘on track’ to pass Uber in 12 months.

But that obviously never happened and this is the crux of Balaji’s argument.

Yipit Data Caveats

From the looks of the latest Yipit chart, Waymo’s market share actually went down from 20% in June 2025 to 15% as of June 2026, while Uber’s ticked slightly up. Great news for Uber right? Not so fast.

As Twitter user m0xt pointed out, there is an important caveat with the Yipit data. It only measures trips that start and end within Waymo’s operating domain, and that domain expanded substantially over the past year, first farther down the Peninsula in June 2025 and then to a 260+ square-mile Bay Area service area stretching from San Francisco to San Jose in November.

That matters because every time Waymo expands, Yipit adds a bunch of new territory where Uber is already well established and Waymo is starting from scratch. So you would naturally expect Uber’s share within the newly expanded operating zone to jump and Waymo’s to fall, even if Waymo is still gaining share across the Bay Area overall.

You can see this clearly around Waymo’s June and November 2025 expansions, when Uber’s share jumps and Waymo’s falls. In other words, some of what looks like Uber regaining market share may simply be an artifact of Waymo expanding its service area.

CPUC Data Also Bolsters This Narrative

Although Yipit data is nice to have, California Public Utilities Commission (CPUC) data gives us another important piece of the puzzle since it represents the actual number of paid trips Waymo is doing across California (Los Angeles and San Francisco).

And if we look carefully at the chart above, we see Waymo at 200,000 trips per week as of May 31, 20251 and by November 30, 2025, they were at 300,000 trips per week, a 50% increase.

Yet over roughly the same period, the Yipit chart shows Waymo’s market share within its operating zone declining.

Those two things aren’t necessarily contradictory. Waymo expanded its operating domain twice during this period, meaning Yipit was measuring Waymo against Uber and Lyft across an increasingly large geography. Waymo could therefore be doing significantly more rides overall while holding or even losing share within the much larger area it now serves.

It would be interesting to see Yipit run the same analysis using a fixed geography, for example, Waymo’s original San Francisco operating zone. That would give us a much better apples-to-apples comparison of whether Waymo has actually gained or lost share against Uber over the past year.

So while Uber’s statement appears to be technically correct, it leaves out some pretty important context: Waymo increased its California ride volume by 50% during this period while also significantly expanding its operating domain.

My Takeaway

I’ve been covering Uber for over a decade, so I’m pretty familiar with the linguistic gymnastics they sometimes use to make their position look as strong as possible. And I don’t blame them. It’s their job to make the case to investors, and it’s my job to call them out when the data deserves a closer look.

And I actually think this is a perfect example of what my original tweet was getting at, even though I was mostly joking.

Yes, Uber has maintained or even increased its market share within the AV operating zones where it competes with Waymo. But during that same period, Waymo grew from roughly 200,000 to 300,000 paid trips per week in California and significantly expanded where it operates. That’s hardly evidence that Waymo has stopped growing or that it isn’t taking meaningful ride volume.

The ironic part is that CPUC data does suggest Waymo’s growth may have started slowing by Q1 2026. So Uber’s argument would have been more convincing if it looked back over the past six months instead of the past year.

San Francisco is obviously an especially good market for Waymo. It’s a dense 7-by-7-mile city with lots of tech evangelists and early adopters, and those dynamics won’t necessarily translate to every market. But I think Waymo has already demonstrated that it can enter a major rideshare market and gain meaningful share relatively quickly. And as it adds airport access and re-starts freeway service, the gap between what Waymo and Uber can offer gets smaller.

If anything, the bigger question is whether Waymo can scale supply fast enough to actually compete head to head with Uber. Vehicle availability has arguably been a bigger constraint than demand, and Waymo is still priced at a premium to Uber. If Waymo can add vehicles, expand its operating domains and eventually bring prices down, there’s still plenty of room for it to take additional share in my book.

Readers, what do you think? Are you Team Balaji or Team Harry on this one now that I’ve had my coffee?

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, The Driverless Digest Editor Ben Hubbard speaks with Brandon Basso, VP of Engineering at Wayve. Brandon breaks down the role of world models in autonomous driving and how Wayve is using them to train, test, and validate its AI Driver.

He explains what problems world models can solve, where their limitations still lie when it comes to rare edge cases, and how Wayve’s latest model, GAIA-4, advances the company’s approach to simulation. The conversation dives into GAIA-4’s closed-loop capabilities, how it can simulate multiple sensor modalities, and why closed-loop testing could give Wayve an edge in virtual testing and validation.

Brandon also explains the distinction between “AV 1.0” and “AV 2.0,” including how an AI-first approach changes the role of data, development, and validation. Ben and Brandon also discuss the challenge of optimizing autonomous vehicles for passenger comfort, what it takes to validate a more generalizable AI driving system, and how Wayve sees GAIA-4 playing a central role in the future of AV development and validation.

Chapters

• (00:00) Introduction to Brandon Basso

• (02:20) Brandon’s day-to-day role at Wayve

• (03:00) What is the Wayve AI Driver?

• (04:05) What is a world model, and why is it used to teach an AI how to drive?

• (05:11) What problem does the world model solve?

• (08:04) How some edge cases constrain what can be tested using world models

• (09:21) How far off are we from capturing all edge cases using world models?

• (11:35) GAIA-4, Wayve’s latest world model and its new features

• (12:06) Closed-loop vs. open-loop testing with GAIA-4

• (14:30) Multimodality and how GAIA-4 simulates different sensors

• (15:27) How closed-loop simulation gives an edge in virtual testing

• (17:38) The difference between AV 1.0 and AV 2.0

• (26:13) How difficult is it to optimize for comfort in the AV industry?

• (27:02) How important data is to the AI Driver

• (27:38) The challenges of developing and validating an AV 2.0 system

• (30:55) The future of validation and GAIA-4’s central role

• (32:18) Conclusion and final thoughts

Notes/Links:

• You can find Brandon Basso on LinkedIn.

• Wayve’s article about GAIA-4 (link).

• The diagram explaining the difference between closed-loop and open-loop, mentioned in the 12:08 timestamp (link).

• Check out Harry’s earlier interview with Wayve CEO Alex Kendall (link).

Top Story of the Week

Moove raises $250M to become the backbone of the robotaxi industry (link).

Moove is the fleet operator for Waymo in Phoenix, Miami, and Las Vegas, and in the future, London… This latest funding will be used to scale the company’s autonomous vehicle fleet management business, including hiring about 350 people.

Related: My podcast interview with Ming Maa - Moove AV CEO (link).

Other Stuff

Waymo autonomous rides now open to public in Dallas as waitlist ends (link).

Anyone can download the Waymo app and request a driverless ride within the company's Dallas service area as Waymo says nearly 150,000 people rode in its autonomous vehicles during the invitation-only phase.

Uber is building an autonomous vehicle empire, and here’s every company it’s using to do it (link). Comprehensive list, if anyone wants to help me rank all of Uber’s AV partners, please reach out.

Atoms and Joby Aviation Partner to Build Nationwide Vertiport Network (link, no paywall). Interesting to see Travis Kalanick move into transportation infrastructure through Atoms, with plans for vertiports that connect air taxis, AVs and rideshare.

Uber and Wayve get TfL approval for supervised robotaxi service in London (link).

TfL confirmed that the vehicles have been licensed as PHVs, not as automated vehicles. The approval allows supervised passenger trips with a licensed PHV driver present, but does not permit fully driverless passenger services.

Amazon’s Zoox to launch paid robotaxi service in Las Vegas on Monday Aug. 10 (link).

Faster Data Off the Vehicle—Faster AI on the Road

Virewirx’s VX60 High Performance Wireless System has piqued our interest—that’s why we’re partnering with them this week to share their unique millimeter wave system built to move massive AV fleet data quickly and reliably. Reach out to Brian Lee at info@virewirx.com to learn more.

Uber CEO brushes off reports of a Waymo break-up (link, no paywall). Sometimes I forget how good Dara is with his words. Calling the “on-the-ground partnership” with Waymo “very strong” gives him some plausible deniability, but it still feels rosy given the recent reports of tensions between the two companies. Uber also seems increasingly eager to point out that robotaxis still account for less than 0.5% of rides on its platform. But that only tells you how many AV rides happen on Uber. It says nothing about all the Waymo rides happening outside Uber, which is kind of the whole point.

“Waymo is a very important partner of ours, and we continue to operate in Austin and Atlanta,” Khosrowshahi said in an earnings call today. “We believe we’ll continue to operate next year in those marketplaces. It’s a terrific product, and the on-the-ground partnership continues to be very strong.”

‘AVs were back at the center of Uber’s Q2 earnings call. 75% of analyst questions included autonomous vehicles as a central topic, the highest share since I started tracking these calls four years ago..’ (link).

“Importantly, in more mature AV markets including Los Angeles, San Francisco, and Phoenix, Uber’s category share, both citywide and in AV operating zones, is higher today than it was a year ago.” I initially misread this tweet and was skeptical of the claim, but after Balaji dropped the hammer and shared the latest Yipit market share data (come on, you guys know I can’t afford Yipit data), it makes a lot more sense. Waymo quickly took share from Uber and Lyft in San Francisco, but over the past year its share has actually fallen while both Uber and Lyft have gained ground.

The chart isn’t quite as bad as it looks for Waymo though (link), since Yipit only counts rides that start and end inside Waymo’s operating zone. Waymo expanded its ODD in June and November 2025, which means the denominator suddenly included a bunch more Uber and Lyft rides in areas Waymo had only just entered. Too late though, damage done to my ego. Balaji wins this round, maybe next quarter he’ll come on the podcast.

Related: Lyft Q2 2026 Earnings Prepared Remarks (link).

We believe the future is hybrid and as AVs scale the market will expand. Take San Francisco, where Lyft Rides within the Operational Design Domain (ODD) grew approximately 20% year over year, the strongest growth across all current markets with AVs.

Waymo CEO explains why Tesla’s camera-only self-driving falls short (link). FSD Supervised is an amazing product, but I’m Team Waymo on this debate. Now that LiDAR and radar costs have come down sharply, it’s hard for me to see the argument against using more sensors and more types of data when you’re trying to build the safest possible driver.

His answer draws the line that camera-only advocates tend to skip right past. “Humans of course can drive with just eyes, so there’s that proof of existence,” he said. “If the goal were to just approximately match human performance or to build an assist product, that’s a very reasonable way to go.” Then the catch. If you’re targeting full autonomy and strongly superhuman performance, he said, “you find that weak sensing just leads to a safety curve that flattens out way too early.”

Related: Waymo Co-CEO Dmitri Dolgov: The Demo Is Only 1% Of The Work (link).

A London Taxi Family Confronts the Driverless Future (link, no paywall).

‘Few words on $UBER: It actually started to look like a no-brainer’ (link) via .

Zipline CEO Keller Cliffton: Zipline update, from our command center (link). Looking forward to our interview!

• Just crossed 2.5 million deliveries.

• Completed 140 million+ commercial autonomous miles.

• Certain individual aircraft have flown 1 million+ autonomous miles.

• Businesses delivering with Zipline increased 13x in H1 2026.

Within Uber-Waymo split, a key labor battle over AVs is being waged in the nation’s capital (link).

Zoox’s regulatory win is a win for Tesla, too (link).

In other words, exemption decisions must be focused narrowly on vehicle performance, not the overall safety of the self-driving technology… That regulatory limitation will make it easier for Tesla to self-certify the safety of its camera-only driving system, which has been heavily scrutinized.

Uber’s Robotaxi Ambition to Face Scrutiny as Shares Languish (link, no paywall). Nice chart ;)

Inside the Luxury Robotaxi Uber, Lucid and Nuro Are Testing (link).

Engineering Tour: Motional IONIQ 5 Robotaxi (link).

Portland moves to regulate robotaxis (link). Requiring a human driver in an AV defeats the point, but I think concerns about displaced rideshare drivers are legitimate. Personally, I like the idea of a small fee on AV trips to support affected drivers. Cities should also have more say and access to AV data, especially when these vehicles are operating on their streets.

‘Americans Can’t Buy Chinese EVs. Waymo Is Importing Thousands’ (link). Great breakdown from Alan Ohnsman at Forbes (got me to pay for a subscription!). The Ojai deal was done well before the huge tariffs on Chinese EVs, so Waymo is now kind of stuck with them. They’ve already imported 3,200+, which is a big number, but with the tariffs and the vehicles coming over stripped down, it’s hard to see Ojai as a great long-term vehicle solution for Waymo.

Related: ‘Just spotted upwards of 500 yes 500 Waymo Ojai today in Mesa, AZ’ (link).

Deep Dive: What Happens When Cars Drive Themselves (link).

Atlas, Kodiak grow autonomous truck fleet in Permian Basin (link).

Atlas now runs driverless trucks concurrently from depots in Texas and New Mexico, roughly 90 minutes apart, widening its delivery reach across one of the world’s most active oil and gas basins. Since deployments began in 2024, the fleet has completed around 7,000 loads across 15 routes, hauling more than 450,000 tons of sand and logging over 23,500 driverless hours in the first quarter of 2026 alone.

RoboDock CEO Zinny Weli: Big milestone: we moved into our own lab space. Building, testing, and iterating all under one roof has been a game changer for how quickly we can move (link). Is that a Jaguar I-Pace I see? 👀

Waymo expands D.C. footprint amid lobbying for council bill (link).

NVIDIA Alpamayo 2 Super, the Frontier Open Model for Robotaxis and Autonomous Vehicles, Now Available for Commercial Use (link).

Related: How NVIDIA Alpamayo Is Accelerating Autonomous Driving with Matt Cragun (link).

‘Zoox updated its Terms of Service today’ (link).

‘Autolane just launched in Austin. It’s 13 Tesla Model Ys making autonomous restaurant deliveries, and it’s live now with Tso Chinese Takeout & Delivery’ (link).

What else we're reading/listening to

Why Waymos Keep Driving Into Floods (link). Our latest guest piece where founder of DeterministicAI.tech Sanjay Kumar explores why flooded roads expose one of autonomous driving’s biggest blind spots, breaking down the sensor limitations behind recent Waymo incidents and the technologies that could eventually solve them. It’s already getting some love on Reddit.

Building the Energy Infrastructure for Autonomous Vehicles with Alex Pischalnikov (link). My latest podcast where I sit down with MN8 Energy’s Senior Director of Strategy to discuss why charging and energy infrastructure are becoming critical enablers of autonomous vehicles, how depot design and charging strategy impact robotaxi and autonomous trucking economics, and what the AV industry can learn from years of EV charging deployment, among other things.

Qasar Younis, Applied Intuition & Mamoon Hamid, Kleiner Perkins | RAISE Summit 2026 (link).

AVs/Humans behaving badly

‘Ojai continues to scrape on this hill - sf’ (link). Turn sound on.

‘Cracked Ojai Screen - LA’ (link).

‘Man smoking on top of Waymo with rider inside’ (link).

‘Another Robotaxi Supervisor Intervention: Hate to put these videos out but after being dropped off at SFO airport on my ride that had an intervention at the end, the driver leaving took over as well to leave departures drop off’ (link).

Shout-outs

Big thanks to TDD readers at Kindred Ventures for referring new subscribers. If there’s someone you think would enjoy TDD, feel free to forward this email or use the referral button below, and we’ll make sure to shout out your company.

Refer a friend

Neat Jobs

Multiple Strategic Partnerships roles at Waymo, via Jimmy Lyons.

Program Manager, Contingent Workforce at Zoox (link) via Steven Kekich.

HR Generalist at Atoms (link) via Chord Fonvielle.

Job Moves

Kimberly Wasiljew: Google -> Waymo (link).

Philip Reinckens: Atoms (link).

Ryan Wallace: FlexAI -> Waabi (link).

We’re now at 11,000+ subscribers! Interested in advertising with TDD? Check out our media kit here and reach out to us. Newsletter spots are currently sold out until September but we do have podcast advertising opportunities available.

Reach out

Until next week :)

-Harry

Imagine you are cruising down an Atlanta thoroughfare in the middle of a sudden, blinding downpour. The rain is drumming on your windshield, and up ahead, the asphalt vanishes beneath a deep, murky pool of rushing water.

As a human driver, your brain instantly fires off basic survival instincts: “Turn around, don’t drown.” You scan the surrounding environment, measuring how high the water is rising against the curbs, watching the cars ahead of you, and feeling the sheer force of the storm. You press the brakes.

Now consider the Waymo robotaxi pictured above. Backed by millions of miles of autonomous training data, a suite of custom-engineered microchips, and a physical array of cutting-edge sensors, the vehicle approaches the exact same hazard - and drives straight in.

This dramatic scene isn’t a one-off stunt. It is a real-world snapshot of a fascinating, high-stakes edge case in autonomous vehicle (AV) perception. Following a series of incidents - including a vehicle being swept off a roadway into Salado Creek in San Antonio and cars getting stranded mid-flood in Austin - Waymo officially issued a voluntary recall with NHTSA (National Highway Traffic Safety Administration) covering 3,791 of its driverless vehicles.

How does a multimillion-dollar supercomputer on wheels fail a test that a teenager with a fresh driver’s license could pass? Let’s explore the autonomous systems engineering behind why this happens - and how the industry plans to fix it.

1. Physics vs. LiDAR: Water Acts Like a Mirror

To understand why a self-driving car struggles in a deluge, we have to look at how it sees the road. Waymo’s vehicles rely on sensor fusion, combining cameras, radar and LiDAR (Light Detection and Ranging).

LiDAR is the crown jewel of AV tech. It fires millions of invisible infrared laser pulses every second, measuring how long they take to bounce back to construct a highly detailed 3D point cloud of the vehicle’s surroundings. But fluid physics has a way of breaking lasers:

• Specular Reflection (The Mirror Effect): When a laser hits dry, rough asphalt, the light scatters in all directions, sending a strong return signal back to the car's sensors. However, when water covers the road, it acts like a mirror. The laser hits the flat surface of the water and reflects away from the car.

• The "Disappearing" Road: Because laser light never bounces back, LiDAR receives zero return data. To the car's perception stack, a deep body of water can register as a flat, empty, clear road. If the car's high-definition (HD) maps say a road should be there, and the sensors detect no obstacles, the computer assumes the path is clear.

2. The Depth Blind Spot

Even if the vehicle’s onboard cameras detect a change in road texture or notice a glossy surface, current computer vision models cannot inherently calculate water depth.

Distinguishing between a harmless one-inch puddle and a catastrophic two-foot torrent requires contextual reasoning. Humans naturally look for physical reference points - how high the water is on a nearby fire hydrant, whether the tires of the car ahead are submerged, or how fast the current is moving. Current vision stacks struggle to compute these abstract, real-time comparisons.

3. The False Positive Dilemma

Every autonomous vehicle programmer must solve a delicate equation: balancing safety with traffic flow.

If a robotaxi executed an emergency stop, swerved, or triggered its hazards every time it encountered a shiny patch of road, a minor puddle, or a wet metallic construction plate, it would cause a nightmare of constant “phantom braking.”

In high-traffic environments, phantom braking is incredibly dangerous, frequently causing human-driven vehicles behind the AV to rear-end it. To keep the ride smooth and predictable, the software is tuned to tolerate minor surface anomalies - which means deep water sometimes gets incorrectly bucketed as an ignorable puddle.

4. The Logic Gap: "Decelerate and Proceed"

The physics of water is only half the story. The trigger for Waymo’s voluntary recall was a specific software logic error.

According to the NHTSA filings, Waymo’s system was designed with a specific instruction for higher-speed roadways (like 40+ mph highways). When the vehicle did detect an untraversable, flooded lane, its programming prioritized keeping traffic moving safely rather than locking up the brakes in the middle of a highway.

The system’s compromise? Decelerate, and attempt to proceed through the water anyway.

This exact logic is what caused an unoccupied Waymo to slow down and then slowly, methodically crawl directly into a high-water area in San Antonio - leading to it being swept away.

The Operational Workarounds - and Why They're Not Enough

While Waymo was quick to respond to the issue, it has not yet solved the fundamental physics problem of water. Instead, the company deployed a combination of operational workarounds and interim measures.

The Final Remedy Is Still Under Development

According to official NHTSA safety recall reports, Waymo's final remedy to resolve how its 5th- and 6th-generation automated driving systems (ADS) handle flooded lanes is still in active development. Because teaching an AI to instantly measure the physical depth of mud-opaque water in real-time is an incredibly complex engineering challenge, there is no set release date for a permanent sensor-based patch.

The Current Workaround

To keep cars out of deep water in the meantime, Waymo implemented an interim remedy:

• Dynamic Map Bans: Waymo has updated its high-definition maps to completely block off areas known to be prone to flash flooding.

• Weather-Alert Geofencing: The fleet's dispatch system is tied to local National Weather Service (NWS) alerts. If a severe storm or flood warning is issued, robotaxis are restricted from operating in those zones.

The catch: this approach is only as fast as the weather reports. If a sudden cloudburst floods a street before the NWS can issue an official flash flood warning, the geofence won't trigger - leaving the vehicle entirely dependent on its onboard sensors, which, as we've seen, are blind to standing water.

Service Pauses and Current Operations

Following these flood incidents, Waymo has had to temporarily pause services in affected cities during severe weather threats. Yet, despite these weather vulnerabilities, Waymo continues to scale up its operations aggressively, choosing to manage flood risk through strict operational shutdowns rather than a vehicle-level sensory fix.

How Engineers Plan to Fix the "Drowning Robotaxi"

To truly make autonomous vehicles flood-proof, engineers are targeting the physical and computational roots of the problem with three emerging technologies:

1. High-Resolution LiDAR-Like Radar

While LiDAR’s infrared lasers are physically blocked or reflected away by water, millimeter-wave radar can easily penetrate rain, fog and water surfaces.

Researchers are developing systems using multiple, widely spaced radar sensors on the vehicle’s front end. By overlapping the fields of view of two or more radar sensors and applying noise-filtering algorithms, the vehicle can generate a highly detailed 3D imaging point cloud - similar to LiDAR - but unaffected by wet road surfaces. This allows the car to effectively see through the water’s surface to scan the structural roadbed beneath.

2. Relative Contextual Vision (Landmark Comparison)

Since cameras cannot directly measure depth like a ruler, software engineers are training neural networks to infer water depth indirectly using the surrounding environment - a technique called Visual Depth Inference via HD Map Anchors.

The vehicle compares its real-time camera feeds against its highly detailed, pre-loaded dry HD maps. If the camera detects water and notices that a physical landmark - like a curb, a fire hydrant, or a traffic sign - is partially submerged compared to its dry baseline map coordinates, the algorithm can calculate the exact height of the floodwater based on how much of the reference object has disappeared beneath the surface.

3. In-Sensor Edge Computing

Currently, self-driving cars stream massive amounts of sensor data from cameras, LiDAR and radar to their onboard computing platform. However, standard LiDAR pulses bounce off wet surfaces like a mirror, as mentioned, returning empty signals that the car’s software incorrectly interprets as a completely clear, dry road.

To solve this visual blind spot, scientists are developing smart optoelectronic sensors using advanced materials, such as rhenium disulfide. These sensors detect the polarization, the specific angle, of reflected light rather than just brightness, allowing them to physically distinguish between dry asphalt and standing water with over 95% accuracy. By analyzing light behavior directly at the hardware layer via in-sensor computing, the device can flag a flooded surface to the central processor before the vehicle enters a danger area.

The "Matrix" Solution: Waymo Is Simulating the Flood

If you can’t safely teach a physical robotaxi to navigate a flood in the real world without drowning it, how do you fix its brain? You build a virtual world and flood it.

In early 2026, Waymo and Google DeepMind unveiled the Waymo World Model (WWM), built upon DeepMind’s ultra-advanced Genie 3 generative AI engine. WWM acts as a generative simulator for self-driving cars. Rather than relying on simple pre-programmed track scenarios, this model can ingest ordinary 2D video footage - such as a passenger’s phone video or dashcam footage of a flooded street - and convert it into a fully interactive, multimodal 3D simulation.

The key breakthroughs of this simulated training ground include:

• Multi-Modal Realism: The model doesn't just generate realistic camera images; it simultaneously synthesizes physically correct LiDAR point clouds. This allows the virtual Waymo Driver to learn exactly how laser light reflects off water surfaces.

• Controlling the Uncontrollable: Engineers can use language prompts to instantly manipulate the layout, control traffic, or summon extreme weather events - including massive flash floods, heavy snowstorms, or even tornadoes.

• Zero-Risk "What-If" Training: Waymo uses this system to safely run thousands of "counterfactual" scenarios. They can ask: "What if the driver didn't decelerate on this flooded highway lane?" or "How would the LiDAR react if the water rose by six more inches?" All of this training happens in the safety of a server rack, without risking a single scratch on their real-world vehicles.

Building an AI that can handle 99% of predictable, dry-weather driving is incredibly difficult. But solving for the fluid, unpredictable physics of the real world - where a road can transform into a river in a matter of minutes - is the true engineering frontier.

-Sanjay

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, I’m speaking with Alex Pischalnikov, Senior Director of Strategy at MN8 Energy. Alex explains how MN8 has grown into one of the country’s largest renewable energy infrastructure companies and why its experience building and operating energy assets positions it to play a critical role in the AV ecosystem. We discuss the company’s business across renewable generation, battery storage, EV charging, and energy management, along with what differentiates MN8 in an increasingly competitive market.

We also dive into the growing intersection between energy infrastructure and autonomous vehicles. Alex explains why charging is becoming one of the most important considerations for commercial AV deployments, how the needs of robotaxi and autonomous trucking fleets compare to large data centers, and why the location, design, and operation of depots can have a major impact on fleet economics. We discuss deadheading, charging strategies, fleet utilization, and the operational challenges companies face as they scale from hundreds to thousands of autonomous vehicles.

Finally, Alex shares lessons the AV industry can learn from years of EV charging deployment, how MN8 works with autonomous vehicle companies to accelerate commercialization, and what the next generation of AV fleet hubs could look like as autonomous transportation scales over the coming decade.

Chapters

• (00:00) Introduction to Alex Pischalnikov and MN8 Energy

• (03:05) What verticals is MN8 Energy focused on?

• (03:58) MN8 Energy’s unique differentiator in the energy sector

• (06:00) Inside MN8 Energy’s renewable energy infrastructure business

• (07:33) The different types of customers MN8 Energy caters for

• (10:00) Similarities between powering data centers and AVs

• (11:45) How early are we in the AV cycle?

• (14:25) The importance of charging to AV deployments

• (16:20) The major cost factor in AV deployments

• (18:52) How MN8 Energy helps AV companies deploy better

• (21:40) The importance of different depot layouts for effective fleet management

• (23:15) How to locate depots effectively to balance costs and reduce deadheading

• (25:45) The biggest fleet management challenges when scaling AV fleets from hundreds to thousands of vehicles

• (29:44) Lessons the AV fleet charging industry can learn from EV customer charging

• (32:27) The future of AV fleet hubs and what they might look like

• (35:40) Conclusions and final thoughts

Notes/Links:

• You can find Alex Pischalnikov on LinkedIn.

• MN8 Energy’s website (link).

• MN8 Energy’s article about AV fleets: Autonomous Fleets May Be Self-Driving, But They’re Not yet Fully Self-Reliant (link).

-Harry

Top Story of the Week

Amazon’s Zoox wins first US approval for paid robotaxis without human controls (link).

Zoox has won U.S. approval for limited commercial deployment of its novel steering-wheel-free robotaxis, ​a first for the autonomous ride industry, the U.S. auto safety agency said on Thursday… NHTSA Administrator Jonathan Morrison told Reuters that Zoox had received clearance to commercially deploy up to 2,500 ​vehicles in each of the next two years… All remote operators must be located in the United States, NHTSA said, and Zoox must publish maps of areas indicating where the vehicles are operating.

“We can say pretty clearly that the systems in place on the Zoox exceed the equivalent performance ​requirements of a compliant vehicle,” said NHTSA Administrator Jonathan Morrison. “But we still want to make sure that the automated driving system will operate appropriately.”

Other Stuff

Baidu’s Apollo Go begins London robotaxi testing with Uber and Lyft (link).

The London deployment comes out of two separate deals Baidu signed in 2025. The Lyft partnership is meant to scale thousands of Apollo Go vehicles across key European markets. The Uber partnership targets thousands of vehicles on Uber’s platform across global markets outside the US and mainland China.

The move highlights London's growing role as a neutral testing ground for US and Chinese autonomous vehicle companies. It also reinforces the increasingly two-horse nature of the global robotaxi race, with Waymo leading commercial deployments in the US while Baidu has quietly built the broadest international footprint, spanning China, Hong Kong, the UAE, Switzerland and now the UK.

Waymo robotaxis are starting to return to freeways (link).

Waymo announced Wednesday that it would gradually add freeway routes, starting with Phoenix — the company’s first driverless ride-hailing market. More cities, including Los Angeles and the San Francisco Bay Area, would follow in the coming days.

DoorDash is building its own drone delivery business (link). This was a big announcement, and I’m bullish on drone delivery over the long term. The technology already seems to be working well and if the regulatory hurdles can be cleared, I think drones have the potential to become the fastest and lowest-cost way to get small orders to customers.

I also find it interesting that DoorDash is pursuing both a partnership strategy and a build-it-yourself strategy. The company already works with Coco and Serve on sidewalk robots while also developing its own Dot autonomous delivery vehicle. On the drone side, it partners with Wing and Flytrex but is now adding DoorDash Air to the mix. Rather than picking a single winner, DoorDash seems content to partner where it makes sense while building its own capabilities in parallel.

The question is what that means for partners over the long run. If DoorDash builds strong in-house robots and drones, does it keep relying on partners or start shifting toward its own tech?

The company unveiled the new business, called DoorDash Air, after receiving a Part 135 air carrier certification from the U.S. Federal Aviation Administration. The certification allows the company to legally operate a commercial drone delivery service in the United States…If DoorDash wants its drones to fly autonomously over longer distances, it will need the FAA to approve its Beyond Visual Line of Sight technology, a certification that companies like Amazon, Wing, and Zipline have received in recent years.

‘A few months ago I had an idea to build a self-driving golf cart’ (link).

Waymo explores split with Uber as robotaxi tensions deepen (link, no paywall).

The self-driving car company has held internal talks about ending its current deals with the ride-hailing group, which operates services in Austin and Atlanta, according to multiple people familiar with the matter. Waymo has already notified Uber that it plans to enter these markets independently in January 2028 when their contract allows, the ride-hailing group said in a statement.

Designing Energy Infrastructure for Autonomous and Electric Fleets

MN8 Energy is one of the largest renewable energy and infrastructure owners in the U.S., supporting fleet operators as they scale electrification and autonomy. With expertise in delivering custom, turnkey solutions tailored to each fleet’s duty cycle and operational needs, bringing reliability, uptime guarantees, and long-term operational ownership, MN8 is the choice partner for fleets as they transition to electric and autonomous operations.

To learn more about MN8’s work with fleet and infrastructure partners, reach out to the Distributed Energy Solutions team at DES@mn8.com

China Resumes Issuing Robotaxi Licenses After Long Freeze (link, no paywall). I’ve been critical of Waymo for their lack of transparency around the recent July 4 issues but we still have no idea what happened with this much more serious incident in China.

Neither the government nor Baidu has publicly disclosed details of the cause of the Wuhan robotaxi outage in March, which triggered the halt in licenses.

Waymo Is Racking Up Thousands of Dollars of Parking Fines in Austin (link, no paywall). Just a drop in the bucket considering Waymo has hundreds of vehicles operating in the city. But it does make me wonder: is Waymo paying those tickets, or is Uber getting the bill?

In Austin, Texas, Waymo’s robotaxis have struggled to follow parking rules. Since the company came to town in 2024, its vehicles have been fined $9,325 in parking tickets, according to documents The Wall Street Journal obtained through an open-records request.

‘In hindsight it was a huge mistake for Rivian to commit to deploying their Robotaxis exclusively on the Uber network’ (link). Of all Uber’s AV partners, Rivian has the furthest hill to climb. As a happy R1S owner, I can confirm they make a great vehicle, but the autonomy features are the same as what you’d find on any other average car (lane keep and adaptive cruise control). We’ll see if all the hype and investment translate into a better product though over time.

Japanese AV truck startup T2 raises $30M (link, no paywall).

Momenta wins Germany-wide level 4 testing approval (link).

Waymo Registers Korean Subsidiary as Asia-Pacific Push Accelerates (link).

Wayve CEO Alex Kendall: A frontier ecosystem for embodied AI (link).

Say hi to the new Ojai in-car experience (link).

What else we're reading/listening to

The AV Market Strategist by : Tesla Q2 2026: 3 New Robotaxi Markets, 36% Fewer Paid Miles & Waymo Puts a Date on the Uber Divorce (link).

The InEVitable on MotorTrend: The Rideshare Guy: How Uber Drivers Make Money—and When Robotaxis Take Over (link).

Electrons & Asphalt by Josie-Dee Li: EV/AV Learning Resources (link). Cool list of electric & autonomous mobility resources, featuring The Driverless Digest newsletter and podcast.

TBPN: Inside Travis Kalanick’s Wild New Al Company (link).

Autonocast by , , and Kirsten Korosec: #368: How Aseon Labs Is Solving AV Operations, w/CoFounder George Kalligeros (link).

AVs Are Coming to Southeast Asia. The Global Playbook is Not (link). Our latest guest piece where Grab’s GM for AVs, Dominic Ong, explores why he thinks the future of ride-hailing in Southeast Asia will not be fully autonomous, but rather a hybrid fleet.

A single monsoon downpour can spike ride requests 9-10x within ten minutes. This is a bind for a standalone robotaxi fleet. If you build your capacity for the peak monsoon surge, you risk underutilizing cars. And if you build for the quiet average, you fail your riders whenever demand surges around events.

How NVIDIA Alpamayo Is Accelerating Autonomous Driving with Matt Cragun (link). My latest podcast where I sit down with NVIDIA's Director of Product for AVs to discuss the company's AV strategy, how its open-source Alpamayo models are helping developers accelerate autonomous driving, and why NVIDIA is combining end-to-end AI with more classical autonomy approaches, among other things.

AVs/Humans behaving badly

‘First public-road accident involving Bliq’s remotely driven passenger car – an important learning opportunity for the entire industry’ (link).

2 hurt after Waymo driverless car catches fire in Atlanta (link).

Data from Waymo shows the autonomous vehicle was turning left off Cheshire Bridge Road when the other car rammed it from behind at 33 miles per hour. The impact triggered a rapidly growing fire in the back of the Waymo.

‘Wrong Way Driving in SF’ (link).

‘Waymo, doing waymo things’ (link).

Shout-outs

Big thanks to TDD readers at NGV Talent, Uber and to Li X for referring new subscribers. If there’s someone you think would enjoy TDD, feel free to forward this email or use the referral button below, and we’ll make sure to shout out your company.

Refer a friend

Neat Jobs

Staff Tech Lead Manager at Waymo (link) via Claire Yan.

Manager, Technical Program Management at Zoox (link) via Ernie L.

Data Engineer at Wayve (link) via Marius Cordts.

Staff Engineer, OEM Platform Integration at Motional (link) via Aimee Henley.

Job Moves

Chetan Shankar: Mercedes-Benz -> Bliq (link).

Andrew Beck: Perplexity -> Bedrock Robotics (link).

Yuka Hamanaka: Haas@Work -> Moove AV (link).

Bree Jones: The Browser Company -> Waymo (link).

Gaurav Jurani: AWS -> Zoox (link).

We’re now at 11,000+ subscribers! Interested in advertising with TDD? Check out our media kit here and reach out to us. Newsletter spots are currently sold out until September but we do have podcast advertising opportunities available.

Reach out

Until next week :)

-Harry

Today, the autonomous vehicle industry has an established model for how adoption unfolds. Deploy in a geo-fenced suburb. Improve the operator-to-vehicle ratio. Ride the unit economics down the curve. Reach parity with human drivers. Scale until the robot wins.

And it seems to be working. Waymo proved it in Phoenix, San Francisco and Los Angeles. Baidu is proving it in Wuhan. Investors are pricing in its replication across every major city.

But after a decade of mobility experience in Southeast Asia – including introducing some of the region’s first autonomous vehicles – my view is that this playbook won’t apply here; not because autonomous technology won’t arrive, but because this model was built for conditions that simply don’t exist on our shores. Global ridesharing platforms had already tried and failed to run a one-size-fits-all template here. Now, autonomy faces that gauntlet too.

Fixed Fleets vs Fluid Markets

The global playbook assumes a predictable environment where mobility players gain market share with a standardized fleet of readily deployable vehicles. But Southeast Asian cities demand an elasticity that a fixed autonomous fleet alone cannot provide.

Consider Singapore. During normal peak hours, ride requests can increase up to 4-5x that of quiet hours, according to our 2026 data. Human supply matches this chaos dynamically, with 3-4x more drivers logging on precisely when the city needs them. It’s a remarkably capital-efficient system: human drivers absorb the spikes and log off when things quieten down, rather than idling an expensive asset.

Equatorial weather exacerbates this volatility. Singapore is 12x wetter than Phoenix and 3x wetter than Seattle. Here, torrential rain isn’t an edge case, it is the baseline environment. A single monsoon downpour can spike ride requests 9-10x within ten minutes. This is a bind for a standalone robotaxi fleet. If you build your capacity for the peak monsoon surge, you risk underutilizing cars. And if you build for the quiet average, you fail your riders whenever demand surges around events.

Furthermore, human knowledge is often required to navigate extreme, fast-changing, driving conditions. Flash flooding in megacities – urban metropolises of over 10 million people, like Jakarta, Manila, and Bangkok – can affect routing on short notice, breaking an autonomous vehicle’s Operational Design Domain (ODD) and turning pre-mapped data obsolete. Ride-hailing drivers can better handle unexpected driving conditions, like navigating floodwaters, or even bypassing mosquito-fogging trucks which can confuse vehicle sensors without warning.

In Southeast Asian cities, the backbone of urban mobility is the two-wheeled motorbike. Millions of commuters rely on these to lane-split through traffic gridlocks where larger vehicles sit paralysed. Because autonomous motorbikes do not yet exist – nor are they coming soon – any global playbook built exclusively around four-wheeled vehicles completely misses how the region actually moves.

And, finally, some transport needs should never be left to a pure autonomous fleet. Wheelchair users need help from door to car. Elderly riders need support while heading alone to hospital appointments. Or, imagine a family of five with a stroller and seven bags at Changi Airport. Some trips will always require a person, not just a vehicle.

This is why a hybrid model – autonomous vehicles paired with a flexible human network – is the steady-state model for Southeast Asia. Here, the autonomous playbook cannot be about replacing human supply; it must be about the orchestration layer that unites them.

The Winning Formula Is About Orchestration

When an operating environment is defined by radical fluidity, the competitive advantage belongs to the platform that can coordinate the chaos.

This begins with predictive marketplace intelligence. Even in structured Western markets, early data shows that plugging autonomous vehicles into an established marketplace can deliver more trips per vehicle per day than running them standalone. That’s because these platforms already know where demand pools and how to route into it. After many years of operations, regional mobility incumbents like Grab have accumulated an irreplaceable operational memory. A platform with scale and hyperlocal operating data doesn’t just react to live traffic; its routing infrastructure anticipates these shifts. Global AV providers can plug into these local networks and get an efficient fleet from day one.

Multi-service flexibility will also be important to optimize supply elasticity. An autonomous vehicle fleet can manage predictable volumes, while human drivers can work across services, from delivering food during the lunch lull to carrying passengers in the evening rush. Being able to orchestrate a hybrid fleet across different services will best maximize utilization across the entire network.

Furthermore, real scale means being tightly woven into the local economy. In Southeast Asia, ride-hailing is a primary income source for millions of people everyday. Their economic stability is a priority for every government here. And government permission – the slowest asset to build, and the easiest to lose – is earned by doing right by citizens, drivers and cities over many years. So, a mature AV strategy has to drive a deliberate transition that works with and for local gig workers. This includes preparing them for specialist AV roles – including fleet operations and depot management roles, like safety and remote operators. By integrating these citizens into the backbone of deployment, you turn potential public friction into real operating strength.

Rewriting The Global Playbook

The prize for the next generation of mobility belongs to platforms that run the orchestration layer autonomous vehicles need to locally work in.

The dense, informal, chaotic, reality of Southeast Asian cities is arguably more of a global rule – closer to the conditions of Lagos, São Paulo, and Mumbai, than outliers of urbanization like Phoenix. By building a marketplace that masters these conditions, you aren’t just solving for one region. You’re building a hybrid model that could eventually determine how autonomy scales across much of the world.

So, to me, the common question: “When will Southeast Asia be ready for autonomous vehicles?” is backwards. The real question is whether the autonomous vehicle industry is ready to be changed by the operating realities of regions like Southeast Asia. It is critical for platforms to lean into local and cultural complexities to translate the promise of autonomy into practical, real-world mobility.

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, I’m speaking with Matt Cragun, Director of Product for Autonomous Vehicles at NVIDIA. Matt breaks down NVIDIA’s strategy for autonomous vehicle development and how the company is approaching both the hardware and software sides of the AV stack. We discuss how NVIDIA works with AV developers to accelerate development, the role of its Hyperion platform, and how NVIDIA is making its technology easier to integrate across different vehicle platforms.

We also dive deep into Alpamayo, NVIDIA’s open-source family of AI models for autonomous driving. Matt explains how Alpamayo works, how it was trained, and how developers can adapt the models to their own use cases. We discuss the limitations of the technology, how it approaches edge cases, and how Alpamayo combines end-to-end AI with more classical autonomous driving approaches rather than relying entirely on one or the other.

Matt also discusses how Level 4 autonomy will eventually make its way from robotaxis into consumer vehicles, where the AV industry stands today, and what still needs to happen before AV technology can scale more broadly. Finally, he shares what excites him most about the next few years of autonomous driving and where he expects the technology to make the biggest advances.

Chapters

• (00:00) Introduction to Matt Cragun

• (02:10) What is Alpamayo?

• (03:50) Is NVIDIA more focused on building the hardware (chips) or software (models) for AVs?

• (05:29) NVIDIA’s strategy for AV development

• (07:03) How NVIDIA helps new AV developers accelerate development

• (08:44) How open-source is Alpamayo, and can it be run on non-NVIDIA chips?

• (09:28) Hyperion and how NVIDIA democratizes access to AV hardware

• (10:40) How Hyperion is vehicle-agnostic and easy to integrate into different vehicle platforms

• (11:20) How Level 4 autonomy will make its way from robotaxis to consumer vehicles in the coming years

• (15:10) Digging deep into Alpamayo: How it works and how it was trained

• (17:16) The limitations of Alpamayo and how AV developers can tailor it to their use cases

• (19:20) The different use cases for Alpamayo, and its approach

• (22:15) How Alpamayo combines the end-to-end AI and classical AV approaches

• (27:23) How their AV models handle edge cases

• (30:46) The current state of AV development

• (32:52) What excites Matt most about the next few years for AVs

• (33:50) Conclusions and final thoughts

Notes/Links:

• You can find Matt Cragun on LinkedIn.

• An article about Alpamayo 2 Super on Hugging face, mentioned at the 04:43 timestamp (link).

• NVIDIA’s article about Alpamayo 2 Super (link).

• NVIDIA Alpamayo landing page (link).

-Harry

Top Story of the Week

Aurora Launches Second-Generation Driverless Trucks in U.S. to Meet Customer Demand (link).

The new fleet will be deployed across Aurora’s commercial network of 10 driverless routes in the U.S. Sun Belt, allowing the company to serve more customers as it expands its autonomous freight operations. The trucks replace Aurora’s first commercial fleet, which entered service last year when the company became the first to operate Class 8 trucks without a driver on public roads in the U.S. Those vehicles have since completed nearly 440,000 driverless miles. Aurora expects the new fleet to grow to 200 driverless trucks by the end of the year.

“Last year’s driverless launch proved our technology could operate safely on public roads – our new platform now provides the foundation to deliver at scale,” said Chris Urmson, CEO and co-founder of Aurora.

Other Stuff

Zoox issues software recall after a robotaxi got confused by heavy smoke (link).

The software update “enhances the existing capability of detecting active [emergency] scenes by adding the ability to detect and respond to heavy smoke in certain situations.”

Report: Waymo Cars Mean Way Mo’ Traffic (link).

Waymo cabs spend far more time driving around without passengers than comparable human-driven Uber and Lyft cabs. In California, where Waymo’s autonomous taxis have been operating and churning out trip data for years, a Waymo will travel 0.71 miles empty for every one mile. But in New York City, so-called high-volume taxis — Uber, Lyft and the like — drive empty for 0.49 miles for every one mile with a passenger. That 14.8-percent gap would translate to tens of thousands of additional vehicle miles every single day.

Related: Are Waymo robotaxis actually safer than NYC Uber and taxi drivers? (link). A lot of the safety analysis cited in this article comes from the report referenced above, but the researchers have since pulled that entire section after discovering discrepancies in the NYC crash data and definitions they used. So for now, I’d take those safety findings, including the claim that Waymo had 1.5x the serious injury/fatal crash rate of NYC for-hire drivers, with a pretty big grain of salt. The report’s authors say they’re working to resolve the discrepancies and update the analysis.

Meanwhile, there has also been some recent research from the Insurance Institute for Highway Safety (IIHS) reinforcing Waymo’s safety claim: Waymo’s driverless cars crash less often than people (link). They found Waymo’s driverless vehicles had 68% fewer police-reportable crashes and 81% fewer injury crashes per mile than human drivers across roughly 50 million driverless miles. This is consistent with Waymo’s own safety research, which found a 79% reduction in injury crashes over 56.7 million miles (link).

Trial Lawyers Lobby Against Autonomous Vehicles (link).

Remarkably the trial lawyers saw the writing on the wall very early and they have been lobbying against AVs for nearly a decade! The trial lawyers earn a huge amount litigating ordinary auto accidents–Annual U.S. auto insurance payouts (liability + PIP/MedPay) are on the order of $180–220B and trial lawyers are very eager to retain the right to sue car manufacturers for product liability.

Tesla spins up robotaxi pilots in Orlando and Tampa ahead of Q2 earnings (link).

Waymo expects D.C. robotaxi talks to continue this summer (link).

Terawatt: How EV Fleet Operators Can Turn Energy Costs Into an Advantage (link).

Hertz’s next act: From rental cars to robotaxi depots (link).

Faster Data Off the Vehicle—Faster AI on the Road

This July 27-30, Virewirx is bringing VX60 High Performance Wireless to ATS 2026. Visit Booth 6 and discover advanced mmWave networking that delivers the massive, real-time data movement that autonomous fleets and robotaxi fleets demand. Reach out to Brian Lee at info@virewirx.com for more info.

Waymo And Uber Support Bad, Incumbent-Protecting Laws In DC And NJ (link, no paywall).

Stellantis taps Mobileye for hands-free driving assist (link, no paywall).

Mobileye CEO Amnon Shashua to step aside as company pushes into robotaxis, robotics (link).

Pony.ai ties up with JD.com for China robotaxi maintenance (link).

Waymo meltdowns, Lurie pressure, spur push for tougher robotaxi rules (link, no paywall).

Under Cortese’s bill, AV companies would be required to immediately dispatch a technician to move a car if it has a steering wheel or allow first responders to move it. Each failure to move cars or ensure that first responders can move them would result in a $5,000 fine, and if an AV causes a roadblock for more than 30 minutes and blocks first responders, the fine would be $10,000. Failure to comply with the law would result in a civil action against the AV company.

Just a few days until SAE International’s Automated Transportation Symposium in San Diego (link).

‘Uber robot taxi, new competitor incoming?’ (link). Closeup of an Uber-Lucid-Nuro robotaxi.

They can do that because they operate using a Limo license like Uber Black, and not an AV permit like Waymo (link).

Related: How the CPUC Handles AV Permits, Data, and Enforcement - Pat Tsen (link).

‘Updated Ojai Cup Holders’ (link).

What else we're reading/listening to

Waymo’s Lead Is Real. Its Biggest Test Is What Comes Next (link). Our latest guest piece where former founding member of Uber’s Autonomous Mobility & Delivery team, Eduardo Rojas, analyzes Waymo’s lead in the Robotaxi space. What strikes me here is the gap between impressive operational scale and actual market penetration. Waymo is currently completing 500,000 rides per week, which annualizes to about 26 million rides. That sounds significant, but against an estimated four billion annual U.S. ride-hail trips, it represents only 0.65% of total demand.

Rethinking AV Safety as Robotaxis Scale with Junko Yoshida (link). My latest podcast with the veteran tech journalist discusses what journalists still get wrong in their coverage of AVs today, the state of the industry, what needs to happen to scale robotaxis, and why transparency remains one of the industry’s biggest challenges.

Related: I’ll be on stage with at SAE International’s Automated Transportation Symposium on July 30th at 11:40 am on the ‘Journalist Panel’, alongside ATS co-founder Jane Lappin, Alan Ohnsman of Forbes, and David Welch of Bloomberg. Feel free to reach out if you’ll be in attendance!

AVs/Humans behaving badly

‘They’re going to kill me’: A Waymo rider was trapped inside as vandals smashed the robotaxi (link, no paywall).

The car’s remote assistant had noticed the car was not moving, and decided to call the police. Watson grew frantic. He asked Waymo’s support team when law enforcement would arrive, but they didn’t know. He asked if they could move the car, and they declined, saying it was obstructed.

‘Absolutely Insane Route’ (link). I’ve had several weird routing issues with Waymo lately and it seems like I’m not the only one. Hard to tell if routing is getting worse for some reason or these are just anomalies.

‘What’s the longest you’ve had to wait for a Waymo?’ (link).

‘My Ojai re-routed into the Waymo backrooms and rider support had to step in’ (link).

Shout-outs

Big thanks to TDD readers at Element Fleet Management, Uber, and to Andy L and Mark M for referring new subscribers. If there’s someone you think would enjoy TDD, feel free to forward this email or use the referral button below, and we’ll make sure to shout out your company.

Refer a friend

Neat Jobs

Staff Product Data Scientist, Paid Media at Waymo (link) via Hongying Z.

Physical Infrastructure Engineer at Waabi (link) via Gregory Carvlin.

Senior Software Engineer, Cloud, Simulation at Wayve (link) via Arvind Pereira.

UK Operations Lead at Waymo (link) via Leticia Cavalcante.

Job Moves

Mason Lee: Nvidia -> Rivian (link).

Laura Salazar Lawton: Torc Robotics -> Avride (link).

Sendil Palani: Tesla -> Zipline (link).

Girish Venkataramani: Wayve -> OpenAI (link).

Neha Sinha: Thermo Fisher Scientific -> Waymo (link).

We’re now at 11,000+ subscribers! Interested in advertising with TDD? Check out our media kit here and reach out to us. Newsletter spots are currently sold out until September but we do have podcast advertising opportunities available.

Reach out

Until next week :)

-Harry

Waymo is the clear leader in the U.S. robotaxi race.

It is the only company operating commercial no-vehicle-operator (NVO) services across 10+ U.S. markets. Today, Waymo is completing more than 500,000 paid rides per week, and its U.S. fleet appears to be approaching 4,000 vehicles.

That lead is real, and it is not small. Its closest U.S. competitor, Zoox, reportedly has a fleet in the low hundreds of vehicles, less than 10% of Waymo’s fleet, and while Zoox is offering public rides in Las Vegas and San Francisco, the company still cannot charge customers as it awaits final regulatory approval for its purpose-built robotaxi.

Waymo built that advantage through patient capital, engineering depth and sustained execution. Alphabet’s profitability gave it room to absorb years of losses, keep investing through setbacks and pursue a problem that most companies could not afford to chase for that long. The engineering made driverless driving work; the execution turned it into a commercial product across 10+ U.S. markets and supported plans for an aggressive footprint expansion this year.

But the more interesting question is what happens next. Has Waymo built a lead that is structurally hard to catch, or has it simply won phase one of a much longer race?

I think the next 12 months will be critical to answering that question. Waymo is clearly ahead, but it has also been operating in the U.S. with limited direct competition. That is starting to change. Zoox is preparing to scale, Tesla remains the wildcard, Uber is assembling a portfolio of AV partners, and Chinese players have already shown that large-scale NVO deployments are possible.

Waymo won the first stage, but the market is still early. The next test is how it performs as the U.S. robotaxi landscape becomes more competitive.

Robotaxis vs. Ride-Hail

Waymo leads the autonomous mobility category in the U.S., but robotaxis still represent a very small part of the broader ride-hail market.

Waymo is completing roughly 500,000 rides per week, which annualizes to about 26 million rides. That sounds large because, for autonomous vehicles, it is. But against an estimated four billion annual U.S. ride-hail trips1, it represents roughly 0.65% of the market.

Even if Waymo reaches its stated goal of one million weekly rides, that would annualize to roughly 52 million rides, or about 1.3% of estimated U.S. ride-hail demand. That would be a significant milestone for the AV industry, but still only a small fraction of the broader rideshare market.

That said, Waymo already has a meaningful presence in key cities. San Francisco is the clearest example, where it has reportedly reached roughly 25% market share and is continuing to expand across the Bay Area.

But San Francisco is also Waymo’s cleanest product-market fit. It is dense, expensive, tech-forward, and filled with high-income riders who are more willing to try new technology and pay premium prices. Obi has reported that Waymo rides can cost 13–27% more than Uber and Lyft, which makes San Francisco a useful but imperfect test case. It is less clear that riders in less affluent or less tech-forward markets will make the same tradeoff, especially while Waymo’s coverage remains limited.

That is why the next phase is so important. The question is no longer whether Waymo can make driverless rides work. It can. The question is whether the company can become the default AV option in the most important U.S. markets before competitors establish a real presence. Waymo has proven that customers value the product in its strongest markets; it has not yet demonstrated that it can translate that lead into meaningful share of the broader U.S. rideshare market.

Waymo’s Lead Is Real. So Why Isn’t It Bigger?

If Waymo’s service works and consumer demand exists, why hasn’t the operator expanded faster?

I think the answer is supply.

Waymo’s fleet is still small relative to the size of the opportunity. A few thousand vehicles is a lot for robotaxis, but it is not enough to serve the largest U.S. ride-hail markets at real density. Expanding into more cities matters, but so does having enough vehicles in each market to offer short wait times, broad coverage and reliable availability.

I think its challenge has been finding the right vehicle partners. The Jaguar I-PACE helped Waymo prove the category but it was never a long-term scaling platform. Jaguar was winding down the platform, and Waymo needed a partner capable of supporting its expansion plans.

That is harder than it sounds. Traditional automakers are built around programs measured in hundreds of thousands, or millions, of vehicles. Robotaxi companies are still talking about fleets in the thousands, and eventually tens of thousands. Those volumes are meaningful for autonomous mobility but relatively small for a global automaker.

There is also a strategic tension. If autonomous mobility works, it may shift some demand away from car ownership and toward mobility-as-a-service. Automakers want to sell vehicles, while robotaxi companies want to turn those vehicles into service infrastructure. Their interests overlap but they are not perfectly aligned.

Waymo is now trying to address the supply constraint with two new platforms: the Zeekr-built Ojai and the Hyundai IONIQ 5. In a June ‘26 interview with Ben Thompson, Michael Morton from Moffett Nathanson estimated Waymo is importing Ojai vehicles at a rate of roughly 300 per month. If that pace holds, it could add more than 3,000 vehicles over the next year, roughly doubling the current fleet.

But imported vehicles are not the same as deployed robotaxis. The real test is how quickly Waymo can validate, permit, and put them into service—and whether the new platforms preserve the product experience that helped build demand.

The Ojai introduces two additional risks. Because it is built in China by Zeekr, it is subject to the 100% China-specific Section 301 tariff in the U.S., making the vehicle materially more expensive. At least on first impression, it appears less premium than the I-PACE. That may not matter to all customers, but it could become more important if Waymo wants to sustain a 10–20% pricing premium as it expands beyond San Francisco.

The IONIQ 5 creates a different challenge. Avride, Motional, and potentially other AV companies are also integrating on the same vehicle, which means the physical product itself may become less distinctive. If several operators use the same platform, Waymo will need to differentiate more clearly through the Driver, the app, service reliability, and the overall rider experience.

The new vehicles should help Waymo scale. The open question is whether they can do so without diluting the product advantage that got them here.

The Opportunity: Win The Largest U.S. Ride-hail Markets

These new vehicles give Waymo a real opportunity to expand across the most important U.S. ride-hail markets.

Waymo has already entered or announced many of the largest markets. Based on my demand-weighted view of U.S. ride-hail, it is currently live in metros representing roughly 27% of estimated demand and has announced or signaled markets representing another 43%. In other words, Waymo now has a presence—or a stated intention to establish one—across close to 70% of estimated U.S. ride-hail demand.

That does not mean Waymo serves 70% of the market today. Nor does it mean its service areas cover entire metros. This is a metro-level demand proxy, not a measure of actual coverage. I count Los Angeles as a live market because Waymo operates there, for example, even though its service remains concentrated within a limited portion of the broader metro area.

But that still matters. Waymo is already in, or heading to, many of the markets that drive U.S. rideshare demand. Now it needs to make the service meaningfully available in each one: more vehicles, shorter wait times, broader coverage, airport access, and more consistent reliability.

In robotaxis, availability is the product. A service that works great but comes with long wait times or is not available in your area is still not a true substitute for Uber or Lyft.

The key point is simple: Waymo has established the market footprint. The next 12 months are about building density within it.

That is both the opportunity and the risk. Waymo’s next chapter is less about proving the Driver and more about industrializing the fleet.

Competition Is Closing In

Waymo is not facing one obvious “Waymo killer.” The threat is broader than that. Several credible players are moving into deployment over the next 6–12 months.

Zoox is carrying public riders in Las Vegas and select riders in San Francisco, although rides remain free. Tesla’s Robotaxi service is available in limited areas of Austin, Dallas, and Houston. Avride has offered rides through Uber in Dallas since December 2025, initially with an onboard specialist behind the wheel. Nuro has secured both a California driverless testing permit and a permit to test passenger services with a safety driver. MOIA is testing autonomous ID. Buzz vehicles in Los Angeles ahead of a planned Uber launch in late 2026, while Wayve and Uber are preparing passenger trials in London.

None of these companies are at Waymo’s level today, and early deployment does not mean the technology is mature. NHTSA has opened a preliminary evaluation into 16 reported crashes involving Avride vehicles in Austin and Dallas. The crashes resulted in property damage and one alleged minor injury, and all occurred while an in-vehicle operator was supervising the system.

Zoox is the runner-up. It is probably the most credible U.S. challenger to Waymo’s rider experience, with a purpose-built robotaxi, Amazon backing, and a factory capable of assembling more than 10,000 vehicles per year at full scale. If Zoox secures the approvals needed to charge riders and expands beyond its limited footprint, it becomes a real second player in the U.S. robotaxi market.

Tesla is the dark horse. It is no longer competing only through announcements: its service is live in three Texas markets, although coverage remains limited. If Tesla can prove the product works reliably at scale, it would be difficult to compete with: low-cost vehicles, massive production capacity, a direct consumer app, a powerful brand, and a leader who can shape the industry narrative. But the rollout remains small and has included reported crashes, so this is still an execution story rather than a solved one.

Uber’s bench is the platform threat. It may not need to own an autonomous-driving stack if it controls the demand layer. By backing Nuro, Avride, MOIA, Wayve, WeRide, Pony.ai, and other suppliers, Uber is building alternatives to vertically integrated players such as Waymo, Zoox, and Tesla. Combining those autonomous fleets with Uber’s structural advantage—human drivers who can cover peaks, edge cases, and markets where AV supply remains thin—could allow it to keep wait times low while maintaining high AV utilization.

The Chinese players matter because they show that Waymo’s technology path is not unique. They are already operating across several markets in mainland China and are beginning to expand internationally. Pony.ai has launched paid robotaxi rides in Zagreb with local operator Verne, with Uber integration expected to follow. Lyft also plans to deploy Baidu Apollo Go vehicles in Germany and the UK.

The UK could become the first Western market where these companies compete from roughly the same starting line. Unlike San Francisco, where Waymo has had years to build brand recognition, operating experience, and customer habits before a serious competitor arrives, London could see Waymo, Wayve, and Apollo Go enter within a similar window. That would give us the clearest test yet of whether Waymo’s product is truly differentiated—or whether part of their advantage has simply come from being first.

Mainland Europe could become an important proving ground. Waymo and U.S.-backed providers are exploring the region, but so are European operators and Chinese AV companies. Europe’s fragmented regulatory landscape means deployment will likely happen market by market, giving new entrants multiple paths to build a safety record and commercial credibility. More importantly for the U.S. debate, if Chinese companies can prove themselves across respected European markets, that could eventually strengthen their case with U.S. regulators.

Waymo has already won the race to prove that robotaxis can work. The next race is to turn that technology lead into a durable market lead. The next 12 months will show whether it can challenge Uber and Lyft as the dominant ride-hailing platform—or whether it becomes just another autonomous fleet supplying those platforms.

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, I’m speaking with Junko Yoshida, a veteran technology journalist and one of the industry’s leading voices on autonomous vehicles. We discuss where the AV industry stands today, why Waymo has emerged as the clear commercial leader, and whether autonomous driving is really a solved problem, or if the industry’s biggest challenges are still ahead. Junko shares her perspective on the importance of edge cases, what the recent Fourth of July Waymo incident in San Francisco reveals about scaling robotaxi operations, and how AV companies should be judged as commercial deployments continue to expand.

We also dive into hybrid autonomy, the role of remote operators, and whether removing the safety driver is truly the right benchmark for autonomy maturity. Junko explains why transparency remains one of the industry’s biggest challenges, how journalists can better cover autonomous vehicles, and why there is often a disconnect between industry reporting and academic research. Finally, we discuss end-to-end AI, exciting developments in London’s AV ecosystem, and preview our upcoming ‘Journalist Panel’ at ATS 2026.

Chapters

• (00:00) Introduction to Junko Yoshida

• (03:30) What journalists still get wrong in their coverage of AVs today

• (05:03) Where we are in the AV landscape today and the factors still being overlooked

• (08:30) Are AVs a solved problem? Why edge cases still matter

• (10:26) Waymo’s 4th of July incident, and what they could have done better

• (17:00) Should all AV companies be judged by the same standard as Waymo?

• (17:48) Hybrid autonomy as a path to scaling AVs

• (19:00) Is removing the safety driver proof of autonomy maturity?

• (19:30) The importance of remote operators in scaling AV operations and boosting passenger confidence

• (22:40) AV regulation and how to ensure the safest deployments

• (26:29) London’s AV landscape and its key players

• (27:14) Transparency and why it’s so difficult to cover the AV industry as a journalist

• (28:30) The gap between industry reporting and academic research in the AV space

• (29:17) End-to-end AI and the potential it could unlock if successful

• (30:35) Exciting AV developments in London

• (31:40) ATS 2026: What to look out for and our ‘Journalist Panel’

• (35:38) Conclusions and final thoughts

I’ll be on stage with at SAE International’s Automated Transportation Symposium on July 30th at 11:40 am on the ‘Journalist Panel’ alongside Jane Lappin, Alan Ohnsman of Forbes, and David Welch of Bloomberg. Feel free to reach out if you’ll be in attendance!

Notes/Links:

• You can find Junko Yoshida on LinkedIn and X.

• Junko’s blog: Junko’s Tech Probe (link).

• Junko’s podcast/newsletter with Prof. emeritus : Phil & Junko on AV Safety (link).

• Waymo’s article on their remote operators ratio, mentioned at the 20:00 timestamp (link).

• Here are some of my food recommendations for ATS 2026 in San Diego, mentioned at the 32:17 timestamp:

  • Downtown/Midtown

    • Lucha Libre

      • Luchador themed taco shop. Must get the Surf and Turf Burrito with fries (California Style) and try the cilantro crema (great salsa bar).

    • Las Cuatro Milpas

      • Longtime hole in the wall spot but recently renovated. Love the homemade tortillas and rolled tacos.

    • Phils BBQ (multiple locations)

    • Bronx Pizza

    • Blue Water (middle town)

      • Great fish tacos

    • Hodad’s

  • PB/La Jolla

    • Leilani’s cafe

    • El Pescador Fish Market & Restaurant

      • Famous Fish Sandwich

  • Other

    • Lolita’s Mexican Food (Clairemont Mesa)

    • Punjabi Tandoor (multiple locations)

-Harry

Top Stories of the Week

Volkswagen enters the robotaxi race with its first passenger service launch in Germany (link).

MOIA announced Wednesday that select, preregistered Hamburg residents can now book rides in a self-driving Volkswagen ID. Buzz van through the company’s app. Up to five vehicles will operate at initial launch, with the fleet expected to expand to as many as 10 cars…Each car will have a trained safety monitor behind the wheel who can intervene when necessary

The pilot will function more like an autonomous shuttle — or a shared robotaxi — than a private ride like those offered by Alphabet’s Waymo. Passengers traveling in the same direction may share a vehicle, while pickups and drop-offs will take place at designated virtual stops.

Waymo Wants in on DC

Waymo really wants to operate in D.C. But does D.C. want Waymo? (link, no paywall).

Here’s a full recording of the hearing if you’re bored this weekend (link).

Related: Waymo says it’ll hire hundreds in D.C. if robotaxis are approved (link). A good reminder that AVs will create new jobs, though over the long run they’re likely to displace far more driving jobs than they create.

Waymo co-CEO Tekedra Mawakana: We’ve autonomously driven more than 300,000 miles in Washington, D.C., safely navigating everything from crowded evenings in Chinatown to emergency scenes in the Third Street Tunnel (link).

Other Stuff

Uber’s Autonomous Vehicle Strategy: Slow Their Adoption (link, no paywall). Well, that’s convenient. Uber’s Waymo partnership has already ended in Phoenix, and most of its other AV partners are still years away from meaningful scale. Slowing robotaxi adoption would certainly give the company more time for its broader AV strategy to catch up.

In New Jersey, a lobbyist representing Uber took the strategy a step further, circulating legislative language that would, for a period of three years, require any platform offering driverless ride-hailing services to have human drivers serve 85 percent of its rides.

Related: Uber’s CFO responds to regulatory capture claims (link).

Bliq wins approval to test driverless cars in Finland (link).

The approval makes Finland the second EU member state after Estonia to permit Bliq’s vehicles on public roads.

Related: Europe’s First Driverless Vehicle Approval with Bliq CEO Julian Glaab (link).

6 in 10 Gig Drivers Borrow to Make Payday Amid Rising Gas Prices and AV Uncertainty, New Report Finds (link).

Everee…released its 2026 Driver Workforce Report, a nationwide survey of 413 U.S. professional and gig drivers examining how rising fuel costs, pay timing, and the expansion of autonomous vehicles are affecting driver earnings and retention… 52.5% are concerned that autonomous vehicles will reduce the number of driving jobs available to them, as services like Waymo and Tesla Robotaxi continue expanding into new U.S. markets.

Restaurant delivery robots return to D.C. with AI upgrade (link).

Coco is launching about 25 autonomous robots serving Logan Circle, and wants to expand into Dupont Circle and beyond if demand is there.

Built for Autonomous Fleet Deployment

Voltera develops, owns, and operates high-capacity charging infrastructure purpose-built for autonomous vehicle fleets. By managing site selection, power coordination, construction, and operations, Voltera enables rapid deployment in dense urban markets where land, power, and permitting are the biggest constraints. As autonomy scales, infrastructure becomes the bottleneck. Voltera delivers high-performance fleet hubs that remove upfront capital and execution risk—enabling faster market entry and scalable deployment.

Please reach out to learn more about partnering with Voltera.

Contact Voltera

Had a great time celebrating the launch of Chip yesterday! Chip is a new all-electric “life utility vehicle” built for neighborhoods, campuses, resorts, and anywhere you don’t necessarily need a full-sized car, including places you do. One of the more interesting parts is Chip Go, a remote-driving service where a real human can remotely drive the vehicle for things like parking, pickups, or errands today, with the long-term goal of Level 4 autonomy in the future (link).

Avride is officially on the ground in Miami! In partnership with Uber Eats (link).

Shared Roadways, Platforms, and Benefits: Integrating Autonomous Travel Choices - The Eno Center for Transportation (link).

‘This is a genuinely interesting research survey to share your ideas for how Waymo vehicles can collect data to enhance urban environments’ (link).

Autonomous Fleets May Be Self-Driving, But They’re Not yet Fully Self-Reliant (link).

Sharing economy in the era of full automation: Evidence from autonomous vehicle on-demand mobility services (link). This research paper explores what a future city could look like with a mix of company-owned and privately owned autonomous vehicles sharing the road. Using Chicago as a case study and ridehail data from November 2024.

Waymos are everywhere in much of LA. But how many people are taking them? (link).

Though people felt less safe in Waymos compared to cars with drivers in them, that perception flipped when people were asked about how safe they felt from harassment — at least for one major demographic. “Women feel a lot safer in a Waymo compared to an Uber or Lyft,” Thomas said.

ROCSYS has been named as one of TIME’s Top 250 GreenTech Companies (link).

Ojai back seat bike storage (link).

What else we're reading/listening to

Ride AI: Ground Truth vol.3 | Building the “Air Traffic Control” for AVs with Autolane (link).

Why Are Waymos Still Blocking Bike Lanes? (link). Our latest guest piece where takes a closer look at Waymo’s repeated incursions into bike lanes, what they mean for cyclists, and how the company’s behavior compares to human rideshare drivers, despite AVs arguably being held to a higher standard. He also explores why Waymo is incentivized to keep operating this way, and what cities could do to encourage safer curbside behavior.

Building Autonomous Vessels for Cleaner Waterways - Michael Arens, Clean Earth Rovers CEO (link). My latest podcast where we explore a very unique application of autonomy: cleaning debris from waterways, highlighting how versatile autonomy is beyond robotaxis.

AV Events

On Thursday, July 23, MobilitySF is hosting Chris Yu of ALSO and Alicia John-Baptiste for a conversation on urban mobility, smaller vehicles, better streets, and the future of cities. Rivian SF Showroom, Hayes Valley. RSVP here (link).

The Austin edition of our Urban Autonomy Summit, happening on October 14, is shaping up to be a gathering of some of the brightest minds in the AV industry, all coming together in one of the most tech-forward cities in the US. Our Detroit edition was a success and we’re bringing that same energy to Austin. Apply now to secure your spot before registration fills up (link).

‘Excited to head back to SF for [TaskUs] Forward ‘26. The conversations we had with AV and robotics leaders last year were incredible, and this year is going to be even better now that everyone is talking less about promises and more about the real operational engine required to run the technology’ (link). It’s coming up on October 28, 2026, at Above the Rim at Chase Center in San Francisco, CA (link).

‘40 30 days until #Actuate26! The full speaker lineup is here!’. Some notable speakers include Wayve CEO Alex Kendall, Aurora CEO Chris Urmson, and NVIDIA VP of AI Research Sanja Fidler. The event will take place August 18–19, 2026, at Fort Mason in San Francisco (link).

‘On July 22, Signal. returns to Kodiak HQ with Kasper Sage, Managing Partner at BMW i Ventures. Join Kasper and Kodiak CEO Don Burnette for a conversation focused on how the best bets in autonomous technology are made, what investors look for, and what separates the companies that get funded from those that don’t’ (link).

AVs/Humans behaving badly

Shirtless man vandalizes Waymo in East Hollywood (link).

‘Trusted Testers, what are y’all doing???’ (link). Waymo may want to re-think cloth seats in the Ojai.

No, a Waymo did not hit a bicyclist in Miami. It was a human driver, police say (link). While police confirmed the Waymo wasn’t involved in the collision, I’d still love to see the video because I’m not entirely clear on what actually happened.

Miami Beach Police have debunked an Instagram video posted by a popular “citizens journalism” account claiming that a Waymo hit a bicyclist on Tuesday in South Beach. Miami Beach Police spokesperson Officer Christopher Bess says the driver of a Mercury Grand Marquis was parked on the side of the road and hit the cyclist when trying to make a U-turn. The driver “was determined to be at fault and was cited accordingly,” Bess said.

Robotaxi Riders Are Falling Asleep, Sparking Frantic 911 Calls (link, no paywall). Feels like something the operators need to solve. First responders shouldn’t have to spend time dealing with sleeping robotaxi passengers.

AVs behaving “goodly” :)

‘Roads are unpredictable. People run red lights, drift out of bike lanes, step off curbs without looking. The question isn't whether it happens. It's whether the vehicle is ready when it does’ (link).

Kodiak: Three trailers. 137 tons. Nearly half a football field in length… Driverless (link).

Collision Involvement Note (link).

Shout-outs

Big thanks to TDD readers at Ono, and to Mark G for referring new subscribers. If there’s someone you think would enjoy TDD, feel free to forward this email or use the referral button below, and we’ll make sure to shout out your company.

Refer a friend

Neat Jobs

Senior Strategic Operations Associate, Autonomous Mobility & Delivery at Uber (link) via Ana Loibner.

Senior Staff / Principal AI & Compute Architect at Zoox (link) via Zoox.

Strategic Partnerships Development Manager at Waymo (link) via Matthew Fisher.

Head of Product at Nuro (link) via Jiajun Zhu.

Mobility Manager at Metropolis in Los Angeles and New York, via Robin Riedel.

System Disruption Manager at Waymo (link) via Jackie Lindsey.

Job Moves

Jeff Beck: Honda -> Waabi (link).

Ryan Browne: Wayve -> ❓ (link).

Waqas Malik: Valeo -> Nuro (link).

We’re now at 11,000+ subscribers! Interested in advertising with TDD? Check out our media kit here and reach out to us. Newsletter spots are currently sold out until September but we do have podcast advertising opportunities available.

Reach out

Until next week :)

-Harry

In 2023, Christopher White, the executive director of the San Francisco Bicycle Coalition (SFBike), received a phone call from a longtime donor who was threatening to yank his funding.

At the time, Waymo was seeking approval to operate its driverless taxis around the clock in San Francisco despite opposition from local officials. The donor was frustrated that the bicycle coalition had declined to endorse Waymo’s expansion. Self-driving cars, he argued, would ultimately make cycling safer because they caused fewer serious crashes than human drivers. After a heated conversation, he invited White to experience a Waymo ride firsthand.

For their joint Waymo trip, the donor selected a dropoff location on Cesar Chavez Avenue, a San Francisco arterial where a painted cycle lane is sandwiched between two lanes of car traffic and a row of on-street parking. “It’s one of our least favorite streets for biking,” White recalled recently. “He wanted to show me how well the robotaxi interacted with bike infrastructure.”

The dropoff did not go according to the donor’s plans. Upon reaching the requested destination, the Waymo “just pulled over into the bike lane,” White said. “I was mortified because somebody on a bike actually had to swerve out of our way into fast-moving traffic.”

Three years later, Waymos are still entering bike lanes when conducting pick ups and dropoffs (examples are here, here, and here). The issue became a point of contention in the District of Columbia during a recent hearing about robotaxis, after which Ethan Teicher, a Waymo spokesperson, posted on X that “riders in places like SF will tell you Waymo vehicles will frequently drive down the block or around the corner to find an appropriate spot.” White was skeptical about that assertion; he said that his members have observed Waymos entering bike lanes “frequently enough to cause concern.”

Beyond being illegal, such maneuvers endanger anyone traveling on two wheels. Last year a San Francisco cyclist filed a lawsuit claiming that she crashed into a Waymo that was blocking a bike lane – and that she then landed on a second Waymo that was also blocking that same bike lane.

This is not a good look for a company that stakes its reputation on the idea that its vehicles save lives. But it’s also true that human drivers pause in bike lanes all the time. If Waymo refused to do so when collecting or discharging a passenger, it could cede market share to competing services like Uber and Lyft that have no such restrictions.

Waymo’s ongoing bike lane controversy—which recently went viral on social media and was covered in outlets including San Francisco Chronicle and Futurism—highlights a tension in robotaxi operations: When autonomous vehicle companies face a choice between road safety and passenger convenience, what will they do?

At the risk of stating the obvious, the purpose of a bike lane is to separate people on bikes from big metal boxes that are moving much faster. A key tool in cities’ efforts to reduce bicyclist crash deaths (which hit an all-time record in 2024), bike lanes also confer a sense of security that can entice more people to start riding.

But those benefits dissipate if motor vehicles impede cyclists’ progress. Harried drivers can view bike lanes as tantalizingly open street space, particularly in places like San Francisco where parking is in short supply. Dedicated pick up/dropoff zones can help, but cities have installed relatively few of them since ridehail emerged almost two decades ago.

A driver seeing a curb full of parked cars may conclude that the bike lane is their next-best option. After all, they will only be there for a moment, and a cyclist can always go around.

Such thinking is deeply problematic. For one thing, it implies a double standard that was satirized in 2022 by a group of activists who gathered on San Francisco’s Valencia Street, a thoroughfare notorious for bike-car conflicts. When a driver halted in a bike lane, the activists dashed into the street and waved signs with apologetic messages—“Hang tight, we’ll be just a minute!”— that cyclists often hear.

The underlying issue is a serious one, said Ken McLeod, policy director at the League of American Bicyclists, a national association for people who cycle. “If a bicyclist comes upon a car in a bike lane, they may have to go into traffic that is going by at a very high rate of speed and a heavy volume.” Doing so is uncomfortable and potentially unsafe. If a passenger is inside the vehicle, there is also an added risk of them “dooring” a cyclist as they exit (which allegedly happened to the aforementioned San Francisco cyclist who is currently suing Waymo).

Because bike lane incursions are so dangerous, many (though not all) states have banned them. For instance, California Vehicle Code Section 21209 states that “no person shall drive a motor vehicle in a bicycle lane” unless they are about to turn, preparing to exit or enter a roadway, or intend “to park where parking is permitted.” Even so, ticketing is rare. Police are unlikely to witness—let alone penalize—an infraction that lasts only a minute or two. A driver may conclude that the time they save from blocking a bike lane is worth more than the risk-adjusted punishment. Such thinking may be rational, but it invites driving behavior that is hazardous and illegal.

So what does all of this mean for a robotaxi company? Waymo did not respond to a request for comment, but last December Chris Ludwick, a senior director of product management, offered insight into Waymo’s perspective when he told the Wall Street Journal that the company’s vehicles are programmed to violate the law in certain situations that require it, such as by crossing a double-yellow line to bypass a delivery van in the street. Fair enough; any reasonable driver would do the same. The only alternative is to remain stuck indefinitely.

But that’s a very different situation from a Waymo passenger requesting a pickup or drop off at a location with a bike lane out front. “Ninety-nine percent of the time there is somewhere just a little less convenient for them to pull over that would be far safer,” said White, of SFBike. Waymo doesn’t need to block bike lanes; doing so is a choice. Although Teicher, the company’s spokesperson, suggested that its vehicles avoid such conflicts whenever possible, White said that he has seen no data supporting that claim.

Of course, the company has a powerful incentive to treat bike lanes as auxiliary parking. If Waymo did not allow its riders to be picked up or deposited where they prefer, it would confer a modest but real advantage to the Uber, Lyft, and taxi services whose drivers willingly disregard the painted lanes on the street. Especially given the immense pressure Waymo faces to scale revenue, alienating customers is not an appealing prospect.

Still, based on his own observations White believes that Waymos may be a bit less likely to stop in bike lanes than Uber and Lyft drivers, who are notorious for doing so. If that is true (and anecdotal data seems to be the only kind available), Waymo defenders might reasonably argue that the company deserves credit for improving upon human driving behavior.

But a defense that “we don’t enter bike lanes as often as Uber and Lyft” seems to place the bar rather low, particularly for a company insisting that it sets a new and higher safety standard. On its website as well as network TV, Waymo claims that its vehicles reduce serious injuries by 94 percent (an assertion that some transportation and robotics experts contest).

These kinds of tradeoffs between street safety and passenger convenience are not unique to bike lanes. Waymos have also obstructed crosswalks, which is hazardous to pedestrians and a barrier to the mobility of those in wheelchairs. McLeod expressed unease about the juxtaposition of self-driving companies’ safety claims with their seemingly deliberate violations of certain rules. “We could have automated vehicle companies saying, ‘We are safer than humans, regardless of our compliance with traffic laws. Therefore, these traffic laws are unnecessary.’ I think that is a dangerously slippery slope.”

For now, Waymo is cultivating ties with street safety communities through its sponsorships of groups like the Vision Zero Network and the Governors Highway Safety Association. That outreach includes cycling organizations as well. Last year the company flew McLeod and other staff from the League of American Bicyclists to Los Angeles to participate in demos and the recording of a slick video. (White said that Waymo previously sponsored SFBike, but that the company ceased doing so in 2023 after the organization did not endorse its push for citywide robotaxi service. He added that his members are roughly evenly split on Waymo: Some think the companies’ vehicles make cycling less dangerous, while others worry they degrade all forms of mobility that do not involve a car.)

The good news is that enhancing the safety of cyclists and other vulnerable road users need not require Waymo to go against its own interests. “Hardening” bike lanes with fixed barriers can prevent cars from entering them in the first place (research suggests they also attract many more cyclists). Santa Monica recently followed the lead of cities including Pittsburgh and Chicago by deploying automatic cameras to issue citations to cars that block bike lanes, negating the need for police enforcement.

Even Waymo has voiced support for “efforts to further protect bike lanes” from incursions by motor vehicles. Rather than cajole drivers—whether human or robotic—to stay out of bike lanes, a better strategy might be forcing them to.

-David

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, I’m speaking with Michael Arens, Founder and CEO of Clean Earth Rovers. We discuss how Michael went from an early failed prototype to building an autonomous maritime robotics company focused on cleaning waterways and collecting environmental data. Michael explains how Clean Earth Rovers’ autonomous vessels operate, the types of debris they collect, and the different deployment models they use for ports, marinas, municipalities, and other customers.

We also dive into the company’s approach to autonomy, including where human operators remain in the loop, why a hybrid human-autonomous model makes sense for maritime applications, and how the hardware stack compares to what’s used in robotaxis. Michael shares his perspective on the broader maritime autonomy landscape, why the economics differ from autonomous driving, and where he sees the biggest opportunities for automation on the water.

We also discuss the growing role of environmental data in Clean Earth Rovers’ business, how remote operations could evolve over time, and what the future holds for autonomous systems beyond roads and into waterways.

Chapters

• (00:00) Introduction to Michael Arens

• (01:40) What inspired Michael to start Clean Earth Rovers

• (04:10) From a failed prototype to Clean Earth Rovers

• (04:50) Their Rover’s specifications, how it works, and its use cases

• (06:44) Where Clean Earth Rovers operates, and how debris migrates there

• (09:00) How the Rovers are deployed, and the different operating models for various clients

• (11:10) How autonomous are the Rovers, and where human intervention comes in

• (12:20) What kind of debris do the Rovers collect, and how is it handled?

• (17:00) Advantages of the Human-Autonomous hybrid approach

• (21:15) The hardware stack of the Rovers, and how they compare to robotaxis

• (25:30) The maritime autonomy landscape and the innovations popping out of it

• (28:20) The value of autonomy in the maritime industry

• (30:34) The data side of the business and how they use the collected data

• (35:22) How remote operation and autonomy would evolve for Clean Earth Rovers

• (39:18) Conclusions and final thoughts

Notes/Links:

• You can find Michael Arens on LinkedIn.

• Clean Earth Rovers website (link).

-Harry

Top Stories of the Week

Waymo cars stranded on San Francisco streets as Fourth of July gridlock drains batteries (link). There are a bunch of videos from this incident, including one where a Waymo drove over a firework, and another where a gentleman counted nearly 20 stuck Waymos in a row.

It’s been a few days since this incident and Waymo still hasn’t shared more details about what happened. How many cars actually got stuck? How many had to be towed because they ran out of battery? Why weren’t they able to make it to chargers? And what are they changing so this doesn’t happen again (again)?

This also comes just a few months after the PG&E outage in San Francisco left a number of Waymos stranded. Different cause, but a similar outcome.

Here’s what Waymo had to say immediately following the incident:

“Major traffic disruptions, a high volume of travelers, and unplanned road closures contributed to unexpected congestion,” a company spokesperson said in a statement. The spokesperson said Waymo’s roadside assistance team worked to clear the vehicles, adding that the company is “evaluating ways to strengthen Waymo’s resilience in major traffic disruptions.”

Other Stuff

Tesla Robotaxi now available in Miami (link), and they say it’s unsupervised (link).

Tesla’s Robotaxi Push Tests New Blueprint for Scaling Fast (link, no paywall). Good recap, but it leaves out one important detail: Tesla’s driverless fleet in Austin is reportedly down to around 14 active vehicles, with roughly 50 Robotaxis being deployed overall. That’s still tiny compared to Waymo, which has around 250 vehicles operating in the same market.

Waymo to start driverless rides in 4 more U.S. markets as expansion accelerates (link).

Alphabet’s robotaxi division will soon start offering fully autonomous rides in San Diego, Las Vegas, Tampa, Florida, and Denver, the company said Wednesday. While the launch will initially be for Alphabet employees, the rollout will then extend to the public.

Nexar and Nauto to Merge, Creating the Independent Infrastructure Platform for Physical AI (link).

Waymo registers entities in France, the Netherlands, Spain and Germany (link).

Feds demand autonomous vehicle companies stop interfering with first responders (link).

“Let me be clear: the inability to detect and appropriately respond to such situations represents a functional insufficiency,” Morrison’s letter reads. “Emergency scenes are not rare or extreme ‘edge cases.’ As such, NHTSA is today issuing a call to action for AV developers and operators to immediately focus their resources on fixing this issue.”

The Human Edge in AV: Scaling Swift Emergency Response

TaskUs partners with the top autonomous vehicle (AV) and AI organizations to manage the human-critical workflows that make advanced systems safer and more reliable. When one of the leading companies deployed its fleets in new cities, they also needed to scale emergency response teams. TaskUs helped by deploying expert operators, creating 24/7 incident frameworks and implementing a unique fatigue management program. Get all the details and results in their case study below.

Check it out here!

One robotaxi player has grown its share of users so far in 2026 — and it’s not Waymo (link, no paywall).

‘FSD would be twice as useful in neighborhoods if I could actually talk to the car and tell it which driveway to pull into, the same way I would with a person driving me home’ (link). Tesla says they’re working on it.

‘Waymo has launched freeways in Austin on the Uber app!!’ (link). This may be a sign that freeway rides are coming back soon.

The robotaxi law that could ban Tesla (link, no paywall). Happy to see Rani Molla at a new outlet. This reminds me though of Lyft’s stance and what I wrote at the time: “I’d rather see an outcome-based framework that judges systems on real-world safety performance, not the specific technology or sensor stack they use.”

A bill expected to come up for a vote later this year would require companies seeking to operate fully autonomous vehicles in New Jersey to use cameras plus two other sensing technologies, most commonly lidar and radar… The measure would also effectively prevent Tesla’s camera-only Robotaxi system from operating in New Jersey unless the company changed its hardware.

Consumers Embrace Driverless Delivery, But Infrastructure Lags (link, no paywall).

‘I’ve lived long enough to get rug pulled by an autonomous vehicle’ (link). It’s hard to compete with Uber’s density and ETAs, but Waymo can stand out with instant matching and ETA accuracy. The typical “Waymo experience” feels especially awesome because Uber’s matching and pickup algorithm can be so inconsistent and is one of the top complaints riders have. Thus, Waymo really needs to avoid developing that same reputation with situations like this.

I used a self-driving car for 100 days. These are the mistakes it kept making (link). I was surprised by how many issues the author experienced over 100 days. It’s a much more negative experience than you typically hear from Tesla FSD supporters, though it’s hard to know how representative either perspective is. The best way to form an opinion is probably to try it yourself.

We recorded more than 500 safety-critical events where the system required driver intervention or revealed an important limitation in how it interpreted the road environment. And In more than 100 days of daily testing, we did not complete a single trip in which FSD drove independently from start to finish without intervention.

Not All Miles are Equal: Why Time and Location Matter When Benchmarking Autonomous Safety (link).

Rivian R2 spotted with LiDAR near HQ, looks better than most LiDARs (link).

Robotaxi Companies Need To Stop It With The Right Seat Safety Drivers (link, no paywall).

What else we're reading/listening to

The AV Market Strategist by : Waymo’s Grand European Expansion & The Waymo-Uber Break-Up Begins in Phoenix (link).

The Real Impact of Autonomy Will Come From Buses (link). Our latest guest piece where analyzes the impact of autonomy on buses, the opportunities it unlocks, and what the industry still needs to get right.

Einride’s Autonomous Freight Strategy with CTO Henrik Green (link). My latest podcast where we discussed how Einride is redefining autonomous freight with their cabless autonomous truck design, their hybrid network of autonomous and human-driven vehicles, and how the company combines electric trucking, autonomy, and logistics into a single platform, among other things.

AV/Humans behaving badly

Waymo calls San Mateo police on teens for drinking, shooting toy guns from driverless vehicle (link). Good one Waymo!

‘First time getting in an Ojai and seeing the front screen not working’ (link). Seems like there are still quite a few bugs with the Ojai despite them testing it for what feels like forever.

Asleep at the wheel: BC driver caught sleeping in self-driving Tesla on busy highway (link).

Shout-outs

Big thanks to TDD readers at CharterUP, and to Mike F for referring new subscribers. If there’s someone you think would enjoy TDD, feel free to forward this email or use the referral button below, and we’ll make sure to shout out your company.

Refer a friend

Neat Jobs

City Fleet Manager - Autonomous Vehicle at Freenow (link) via Swagat Chopra.

Solutions Architect at Wayve (link) via Masa Kameyama.

Program Manager, Driver and Vehicle Compliance at Waymo (link) via Karen Isgrigg.

Autonomous Vehicle Partnership Manager at Lyft (link) via William Lynn.

Principal Engineer Tech Lead at Motional (link) via Gayatri K Sukumar.

Job Moves

Balajee Kannan: Motional -> May Mobility (link).

We’re now at 11,000+ subscribers! Interested in advertising with TDD? Check out our media kit here and reach out to us. Newsletter spots are currently sold out until September but we do have podcast advertising opportunities available.

Reach out

Until next week :)

-Harry

Public transit is the backbone of cities. In New York, buses carry around 1.3 million trips a day, more than ride-hailing and taxis combined at roughly one million. In London, the gap is even wider: about five million bus journeys versus approximately 0.6 million taxi and ride-hailing trips1.

Yet most of the autonomous conversation still centers on cars. Partly because the commercial upside is obvious and near-term, and partly because cars are sexier. But it is buses that carry far more people, and that’s where autonomy will have the real impact.

More Public Transport on the Same Budget

Turning buses autonomous holds real advantages. The first is solving the driver shortage. In 2023, the Geneva-based International Road Transport Union (IRU) estimated that Europe was missing around 100,000 bus and coach drivers, equivalent to 10% of the total workforce. By 2028, that number could more than double.

Then comes the business case. Removing the driver changes the economics: lower operating costs; longer service hours, including extended nights and weekends, which raise vehicle utilisation; and smoother, more consistent driving, which can cut fuel and maintenance costs by at least 7%.

Last, the social benefits. Greater efficiency and availability could make it viable to run services that are uneconomic today, especially in lower-density areas. Add to that the potential safety gains autonomy could deliver.

The Ecosystem Is More Robust Than It Seems

Across the world today, more than 30 autonomous bus and shuttle projects are running on public roads. As of June 2026, services were operating across Europe (Norway, Sweden, Finland, France, UK), Asia (Japan, South Korea, China) and the US (Atlanta, Jacksonville), to name a few. A smaller but important subset has moved into fare-collecting services, where passengers pay and the bus is treated as part of the transport network.

The supplier landscape is taking shape too. ADASTEC, Imagry, Fusion Processing and TIER IV are among those working on autonomous bus platforms. Add shuttle players such as Ohmio, WeRide and HOLON, and the market starts to look less experimental than many assume.

Technology still needs to improve, but it is not THE limiting factor. Buses have one important advantage over robotaxis: they run on fixed routes. That creates constrained operational design domains, which can be mapped, tested and prepared for in detail. Nor do operators need to automate the whole network at once. They can go route by route, starting with the simplest corridors: bus lanes, lower speeds, easy terrain and clear road markings.

In April 2026, Norway went a step further: on a regular scheduled bus route, it removed the safety driver. Yet the driver is still on the bus, sitting in a passenger seat. That’s because drivers serve many more functions than simply drive, and removing them is where the hardest problem lives.

Respect the Driver

A driver does much more than drive. They quietly handle the many small moments that keep a service running: elderly riders who need time to board and sit down, wheelchair users who need the ramp deployed, passengers who rely on a visible human presence when things turn uncomfortable or unsafe. They are also the default answer to “what happens if…” - the accident, the breakdown, the medical emergency, or the passenger in distress. And they are part of the reason people pay. With no one at the front of the bus, how do you ensure fares are collected?

Public transport is designed to serve everyone, and vulnerable riders often lean hardest on that human presence. Remove the driver, and these tasks do not disappear. They have to be redesigned into the service. Bus autonomy is not “just” a technology challenge; it is mainly an operating model challenge.

Show Me the Money

The second major challenge is procurement. Public transport is not bought like consumer technology, where a better product can spread quickly once users like it. Buses are procured through long, risk-averse public processes shaped by central and local governments. Operators never procure at scale without a clear public budget.

That creates a vicious cycle. Few authorities are ready to budget for scaled autonomous operations, which means operators won’t procure them. That, in turn, makes it harder for technology providers to raise capital, finance development and execute deployments. And unlike Waymo, bus autonomy companies cannot simply fund and run large fleets themselves while the market catches up. So the proposition stays nascent, while everyone waits for someone else to go first.

Digital Bus Driver

As for the major challenge, the driver, the industry is already at work. The answer has two parts.

The first is service design. There will not be one driverless model. Different routes and communities will need different levels of human presence. Some services may run fully autonomously; others may carry an onboard attendant to assist passengers, at least during certain hours or on certain routes. A service might run unstaffed during the day, then add an attendant at night. Remote support can flex too: from a single control centre in another city to local “rider-assistance stations” that can reach passengers at short notice.

The second is the emergence of a new operating layer. Companies such as Beti, Applied Autonomy and Beep are rebuilding the functions that used to sit with the driver. Remote supervisors can operate the air conditioning, open and close doors, and, where permitted, intervene in how the vehicle operates. Microphones, speakers and help buttons connect passengers to the control centre, while cameras and software flag anything from anti-social behaviour to fare evasion. These tools make up the digital bus driver: a bundle of tools, software, procedures and human support channels that keep the service running.

Looking Ahead: A Slow Transition to Commercial Services

For now, autonomous bus deployments will keep relying on public funding, national innovation programmes, operator sponsorship and regulatory exemptions. But over the next five years, expect the market to shift from experimentation to early deployment. Trials will multiply, technology and digital-driver tools will mature, more services will turn commercial, and more safety drivers will move out of the driver’s seat - either into passenger-support roles or remote operations.

Watch Japan, Switzerland and Scandinavia in particular: the driver shortage is acute, regulation is moving and trials are most advanced.

Cars will keep the spotlight. But if autonomy is going to change how cities move, it will do it on the bus. Safely, reliably and affordably, on a wet Tuesday night, with real passengers, real fares and no driver at the front.

-Barak

Listen to this episode on Apple Podcasts, YouTube, Spotify, and Substack. Transcripts are also available by clicking on the “transcripts” button in the top right corner of this post on the app/web version. You can also access them from the email version by clicking here.

In today’s episode, I’m speaking with Henrik Green, Chief Technology Officer at Einride. Henrik shares his journey from a long career in the traditional automotive industry to helping lead one of the most unique companies in autonomous freight. We discuss the differences between building vehicles for consumers versus commercial autonomy, what attracted him to Einride, and why he believes the AV industry is entering an exciting new phase.

We also take a deep dive into Einride’s business model, including its cabless autonomous trucks, hybrid network of autonomous and human-driven vehicles, and how the company combines electric trucking, autonomy, and logistics into a single platform. Henrik explains the technical and economic realities of Einride’s purpose-built vehicle design, how the company approaches scaling deployments, and the regulatory considerations unique to autonomous freight.

We also explore some of the biggest technical challenges facing the industry today, Einride’s hardware choices and operational design, mapping, AI, and how they handle edge cases in real-world deployments. Finally, Henrik shares his outlook on the future of autonomous freight, where he sees the industry over the next five years, and what he’s most excited about as commercialization continues to accelerate.

Chapters

• (00:00) Introduction to Henrik Green

• (02:33) Henrik’s background in the automotive industry

• (04:00) The differences between the B2C automotive industry and the AV industry

• (05:09) What drew Henrik to Einride

• (05:50) Trends in the AV career space

• (06:50) Einride’s background, and what makes them stand out in the industry

• (07:35) Einride’s footprint, and their focus

• (10:40) Einride’s cabless autonomous trucks, their specs, and current applications

• (13:11) How Einride plans to scale, and their goals for the next year

• (13:50) The economics of Einride’s autonomous trucks, and technical specs

• (15:29) What hardware/software is unique to their cabless autonomous truck design?

• (16:30) Regulatory requirements for autonomous freight vs robotaxi operations

• (17:12) Hardware/operational differences: autonomous freight vehicles vs. robotaxis

• (19:12) How Einride’s autonomous truck form factors could be optimized in the coming years

• (20:14) Einride’s competitive advantage: Hybrid network of human-driven and autonomous freight vehicles

• (22:50) Biggest technical challenges involved in building and operating autonomous freight vehicles today

• (23:30) How Einride’s autonomous freight trucks are trained for deployment, and edge cases

• (25:48) Pre-mapping vs. end-to-end AI: how Einride uses a hybrid approach for safe deployments

• (28:40) Balancing commercialization with scaling: Einride’s approach

• (31:00) What are you most excited about over the coming years

• (34:55) 5-year outlook for the autonomous freight and robotaxi industry

• (36:24) Conclusions and final thoughts

See You at ATS 2026

The SAE Automated Transportation Symposium (ATS) is one of the premier gatherings for the autonomous vehicle industry, bringing together engineers, researchers, regulators, policymakers, and executives from across the ecosystem. I’ll be speaking on the Journalist Panel on Thursday, July 30 at 11:40 a.m. This year’s keynote speakers include Uber’s Head of Autonomous Mobility & Delivery, Sarfraz Maredia, Zoox Co-Founder and CTO Jesse Levinson, along with other leaders helping shape the future of automated transportation. Drop me a note if you’ll be in attendance, and registration is still open if you haven’t signed up yet. I hope to see you in San Diego.

Register now

Notes/Links:

• You can find Henrik Green on LinkedIn.

• Einride’s website (link).

• You can find details/pictures of their cabless autonomous trucks here.

-Harry