BYD Will Pay You If Its ADAS Crashes
Plus: Tesla self-certifies L4 but only in Texas, and you can finally try Waymo’s Ojai
This is the Ride AI newsletter: The most comprehensive weekly digest of news and intelligence at the intersection of technology and transportation.
ICYMI, early bird tickets for Ride AI 2027 are available. If you’d like to apply for an early spot, please do so here.
Top Story
BYD has unveiled Xuanji A3, China’s first 4nm smart-driving chip at its May 28 intelligence strategy event. The Xuanji A3 supports L3 and L4 functions, and a three-chip cluster delivers over 2,100 TOPS. BYD says it’s already in mass production, making BYD the only automaker with full-process chip capability, from design through wafer manufacturing.
But new hardware is always getting announced. Perhaps the bigger deal is that BYD is putting its money where its mouth is by announcing it will cover all costs of at-fault accidents while its “God’s Eye” urban pilot and parking systems are engaged, with no payout cap, no separate insurance purchase, and no hit to the owner’s premiums. God’s Eye B costs 12,000 yuan ($1,800 USD) as a one-off purchase, roughly a fifth of Tesla’s FSD package in China (64,000 yuan, $9,400 USD).
The significance here is that there are finally consequences behind the marketing. Most companies talk up how safe and capable their driver-assist systems are, but bear no real downside when those systems fail, because the driver stays legally liable. Shifting that liability from the user to the automaker is what actually earns trust, and there’s evidence it works. Last year, BYD started covering its God’s Eye autonomous parking function, and it says that guarantee pushed feature usage from 21% to 93%. Extending the same no-strings coverage to urban driving could drive similar adoption, and it pressures rivals who still make owners buy separate “intelligent driving insurance”, or worse, make claims but provide no insurance at all, to follow.
Domestic News
Tesla has self-certified its robotaxis as Level 4 autonomous in Texas, the first time the company has carried an L4 designation anywhere. The move came as Texas Senate Bill 2807 took effect on May 28, allowing companies to run commercial driverless services once they obtain DMV authorization and self-certify their vehicles to an SAE Level 4 standard. Tesla did exactly that on day one.
That being said, the certification remains a self-attestation, not an independent regulator’s finding. And the fleet is relatively tiny, with Texas DMV records showing just 42 authorized Tesla robotaxis in the state, and Robotaxi Tracker showing even less in actual service. It’s also important to note that Texas has deliberately built one of the lightest-touch AV regulatory regimes in the US, which is why both Tesla, Waymo, and other operators have concentrated early commercial efforts there.
The symbolism, however, matters more than the 42 vehicles. After years of Musk predicting full autonomy “next year,” Tesla now has a legal L4 footprint. Whether they can scale up from there in tougher regulatory environments remains the big question. After all, in California, Robotaxi is still technically a limo service, not an autonomous vehicle service.
Waymo opened its next-generation Ojai vehicle to its first public riders, a purpose-built robotaxi that debuts the 6th-generation Waymo Driver. Select riders in Waymo’s Trusted Testers program in San Francisco, Phoenix, and Los Angeles will get free rides first, with Denver, Las Vegas, and San Diego to follow before a wider opening later this year. The Ojai (built on a minivan platform developed by Chinese company Zeekr) is designed rider-first: elevator-style doors with invisible B pillar, a flat floor, separate screens for each passenger, and accessibility features like surfaces embedding braille and screen-reader support.
Waymo’s 6th-gen Driver is now also rated for even snowier conditions, and the company says it’s scaling its Mesa, Arizona factory toward tens of thousands of units a year. In total, the Waymo Driver has now served over 20 million fully autonomous trips across 11+ cities.
Separately, Waymo confirmed it has begun mapping Northern Virginia, running vehicles with safety drivers in Alexandria and soon Arlington. There’s no commercial service planned yet. The mapping is groundwork in case Virginia authorizes autonomous ride-hail, which is not expected before 2028.
Nuro showed one of its Lucid robotaxi engineering vehicles navigating downtown Mountain View, as the Nuro-Lucid-Uber program ramps up road testing. Nuro recently received a driverless testing permit in California, and I’ve been seeing these vehicles testing on the roads myself in various parts of the San Francisco Bay Area, though testing for now still runs with a safety operator in the driver’s seat. The broader deal, which was expanded in April, commits Uber to deploy at least 35,000 Nuro-powered Lucid vehicles over six years across dozens of markets, with commercial service starting in the Bay Area later in 2026.
Mobileye unveiled two AI tools, Meteor and Genario, built to attack the autonomous driving “long tail” of rare edge cases that decide whether AVs stay geofenced or go everywhere. Meteor is a multi-agent data-mining engine that hunts through millions of hours of driving footage for reproducible failures (think partially occluded pedestrians or ambiguous road users), generates hypotheses for why they happen, and surfaces high-value training examples automatically. Genario then takes those validated failures and generates synthetic, photo-realistic variations across weather, lighting, and road layouts to expand coverage. The idea is that more data is necessary but not sufficient, since most driving data is ordinary and scaling it alone hits diminishing returns, something other AV companies have also stated. CTO Prof. Shai Shalev-Shwartz will present the work publicly for the first time at CVPR on June 3, as part of the Workshop on Autonomous Driving. There’s also research commentary from Amnon Shashua worth reading.
David Moss has done it again, completing the first coast-to-coast drive across Canada on FSD with zero interventions, running 6,051 km from Vancouver to Halifax in 4 days and 21 hours. Moss, Devin Olsen, and Spencer Scott made the trip on FSD v14.3.3 through Rocky Mountain passes, the prairies, construction zones, bad weather, and wildlife, reportedly without a single disengagement, including parking at Supercharger stops. FSD chief Ashok Elluswamy gave the team a shout-out.
International News
Bliq.ai has won approval to run fully driverless vehicles on public roads in Estonia, the first authorization of its kind in an EU member state. The Berlin and Tallinn-based startup runs what it believes is Europe’s largest fully driverless fleet (a dozen-plus vehicles) under remote supervision. Bliq retrofits Ioniq 5s with a fast-to-integrate sensor and compute stack. Its pitch is that driverless tech shouldn’t be limited to robotaxi fleets, and Germany is its next target market.
Pony.ai raised its 2026 robotaxi fleet target to 3,500 vehicles, up 500 from the goal it set just six months ago, implying that the company actually had more scaling capacity than it originally thought. Q1 revenue jumped 145% year-over-year to $34.3 million, beating estimates, with robotaxi revenue alone up 395%. The fleet has already passed 1,700 vehicles, which implies roughly 1,800 more deployments in seven months. Pony now operates in 20+ cities and recently launched what it calls Europe’s first commercial robotaxi service in Zagreb, Croatia, with Uber and Rimac-owned Verne.
I had the opportunity to chat with Pony’s CTO Dr. Leo Wang at Ride AI 2026 about how Pony looks at scaling in China and how Chinese government regulation affects Pony’s plans, which you can watch here.
Mercedes-Benz will roll out its MB.DRIVE ASSIST PRO urban system in Germany by year-end, starting in Stuttgart and Munich, with a nationwide rollout in early 2027. To be precise, this is an SAE Level 2 system, not autonomous driving: the driver must stay attentive and ready to take over. At the press of a button it guides the car point-to-point through city traffic, handling traffic lights, lane changes, intersections, and pedestrians. It’s already live in China and is coming to the US later this year. Separate from Mercedes’ existing Level 3 Drive Pilot (a motorway-only system), this is the company’s bid to commercialize urban point-to-point assistance in Europe.
Autobrains and Uber will launch a Level 4 agentic-AI robotaxi program in Munich, built on NVIDIA’s DRIVE Hyperion platform, pending regulatory approval. Munich is the first deployment city for what the partners describe as an OEM-agnostic model meant to scale across vehicle platforms. Autobrains’ pitch is its “agentic AI”: rather than one monolithic end-to-end model handling the entire driving task, it decomposes driving into specialized agents that each reason over a specific context and run on standard sensors and automotive-grade compute. It’s the third Uber autonomy tie-up in this issue (after Nuro and Pony.ai), part of Uber’s strategy to be the marketplace layer on top of many AV stacks rather than build its own.
At the same GTC Taipei event, NVIDIA also launched Cosmos 3, an open “world foundation model” for physical AI. The model simulates physical environments and generates synthetic, physics-accurate driving scenarios, which lets developers train and evaluate models with far less real-world data and cuts training cycles from months to days. It’s the same long-tail problem Mobileye is chasing, approached from the synthetic-data side. Li Auto is named among the AV developers building on it.
Pop Culture
You know a technology has truly arrived when it shows up in kids’ insults. As one overheard SF middle schooler put it, “You’re so dumb, you can only get a job driving a Waymo.”
Alright, that’s it from me… until next week. If you enjoy this newsletter, share it with your friend, colleague, or boss. Thank you for reading; Sophia out!
[SHARE BUTTON]













The piece on the two AI tools, Meteor and Genario by Mobileye, was super interesting! Sifting through data and building scenarios to help resolve elusive problems are exactly the high-powered skillsets at which I always thought AI-based tools would excel. I also appreciate you sharing links with the recommendation that they're worth a read.