In autonomous and self-driving vehicle news are Waymo, Motional, HiRain, Arbe, CDTL, Pony.ai, Verne, WeRide, Uber, AVOMO and Hyundai.
In this Article
Waymo Does Way-More Calls Cops on Ghost Gun, Mace & MJ
On September 3, one of the Waymo’s self-driving cars pulled itself over in San Francisco’s Richmond District after detecting what Waymo described as a violation of its terms of service involving a firearm. A Waymo spokesperson confirmed to the Los Angeles Times that the incident involved one of its vehicles, and the company alerted emergency services and cooperated fully with police.
Officers arrived and detained two juvenile passengers, a boy and a girl, and later recovered a loaded, AR-style ghost gun from inside the vehicle. Ghost guns, which lack serial numbers, are notoriously difficult for law enforcement to trace, and the teens were also found with suspected marijuana and mace spray before being booked into San Francisco’s Juvenile Justice Center. It wasn’t the first time Waymo has flagged troublesome riders; back in July, the company alerted San Mateo police after two teenagers were caught drinking alcohol and firing toy guns from inside one of its cars. The incidents raise real questions about how much responsibility a driverless vehicle should carry for policing its own passengers, especially as Waymo continues to expand its footprint, having just launched paid service in San Diego, Denver, and Tampa.
Motional Wants Self-Driving Cars to Show Off
Motional just released nuReasoning, a free, reasoning-focused dataset built to teach self-driving systems not just what they’re looking at, but why they’re making the choices they make. The data comes from real fleet driving in Las Vegas, Pittsburgh, Los Angeles, Boston, and Singapore, adding up to more than 105 hours of footage across 20,000 driving scenarios, each at least 20 seconds long.
What makes this one different is the emphasis on causal logic. Rather than simply labeling a pedestrian as a pedestrian, the dataset’s 247,000 human-verified annotations dig into the tactical reasoning behind a maneuver, including why a driving system might have rejected an alternative path as unsafe. That kind of detail is exactly what’s needed to train a car to handle the rare, weird stuff, like a nighttime work zone or a deer bolting across the road at the wrong moment. Developed with UCLA’s Mobility Lab and built on the lineage of Motional’s earlier nuScenes and nuPlan datasets, nuReasoning also comes with its own natural language search tool, Omnitag, so researchers can dig through scenarios without wading through raw sensor logs. A companion challenge is already underway at the European Conference on Computer Vision, with winners to be announced at NeurIPS.
Hirain and Arbe Bring High-Res Radar to China
Hirain Technologies and Arbe Robotics have landed a production contract with a major global automaker to supply ultra-high-resolution imaging radar for a Level 3 vehicle platform sold under multiple nameplates across China, with deliveries expected to begin in the fourth quarter of 2027.
The system pairs Arbe’s radar chipset, which uses 2,304 virtual channels, with Hirain’s manufacturing and sensor packaging expertise. The result is a radar sharp enough to tell a stationary car apart from a ghost reflection or a stray signal bounce, which matters enormously for eyes-off driving, the kind of Level 3 setup where a driver can legally look away from the road. It’s also a sign of where the Chinese market is headed: automakers there are racing to make Level 3 systems that work reliably at scale, and having radar as a backup to cameras is quickly becoming standard practice rather than a nice-to-have.
CDTL Shows Off Brakes and Axles for Driverless Trucks
Over in Hannover, Germany, Hangzhou Contemporary e-Drive Technology, known as CDTL, is bringing its commercial vehicle hardware to IAA Transportation 2026. The company’s pitch is straightforward: if you want a truck or bus to drive itself safely, you need the physical hardware translating digital commands into wheel motion to be just as smart as the software making the decisions.
CDTL’s EA5000NP distributed e-axle and its dry electromechanical braking system skip the pneumatic and hydraulic middlemen that have traditionally slowed down braking response in heavy vehicles. Paired with proprietary software for torque distribution and anti-lock braking, the system is designed to keep a multi-axle truck stable during hard braking or automated lane-following, which becomes a much bigger deal once there’s no human ready to grab the wheel. CDTL’s tech is already running in trucks, transit buses, and autonomous rapid transit fleets in China, Europe, the U.S., Canada, and Australia, and the company will be showing it off at Booth D37 in Hall 21 through September 20.
Pony.ai and Verne Driverless Robotaxis in Zagreb
Croatia just became the latest stop on the global robotaxi tour. Pony.ai and mobility operator Verne have started letting invited passengers ride in fully driverless vehicles along a 22-kilometer route connecting Verne’s headquarters to Zagreb’s city center and Franjo Tuđman Airport. It’s a meaningful step up from having a safety driver behind the wheel, and it builds on a partnership with Uber that launched back in April and has already logged more than 200,000 operational kilometers.
The vehicles running the route use Pony.ai’s seventh-generation platform, built around an NVIDIA DRIVE AGX domain controller running the company’s safety-certified DriveOS. That hardware has already logged more than 40 million fully driverless kilometers on the streets of major Chinese cities, so Zagreb is essentially a proving ground for taking a mature system into a brand-new regulatory environment. Verne handles the fleet and passenger side locally, Uber manages the app and dispatch, and Pony.ai supplies the driving stack, a division of labor the company hopes to repeat across the 3,500 commercial vehicles it wants running in more than 20 cities worldwide by the end of the year.
Spain Grants Its 1st Nationwide Robotaxi Permit in Madrid
Spain has officially opened the door to Level 4 autonomous vehicles, granting WeRide, Uber, and AVOMO the country’s first national operating permit under its new ES-AV framework. The approval lets the trio start mapping streets and validating sensors across Greater Madrid, laying the groundwork for a commercial robotaxi service the partners hope to launch before the year is out.
Twenty of WeRide’s purpose-built GXR vehicles will handle the early phase of testing, still with a safety specialist on board for now. It’s a notable milestone beyond Spain, too, since this marks the first time the European Union has approved the GXR platform at a national level, which should make it easier for WeRide to expand into other EU countries down the road. As with other markets in the WeRide-Uber alliance, the roles are divided cleanly: WeRide brings the driving technology, AVOMO runs the depots and vehicle maintenance, and Uber handles where and how riders actually book a trip. Madrid becomes the fourth city in the alliance’s global rollout, which is aiming for 15 metropolitan markets by 2030.
Hyundai Roadmaps Self-Driving Brain
Hyundai Motor Group used its Autonomous Driving Media Day at 42dot’s headquarters to walk through a roadmap that stretches out several years. The near-term plan leans on NVIDIA, with Level 2+ vehicles arriving in the first half of 2028 and Level 2++ systems following later that year. Hyundai’s own end-to-end driving model, called Atria AI, is aimed at reaching Level 2++ capability in the second half of 2029, and the automaker plans to standardize sensor hardware across Hyundai, Kia, 42dot, and Motional around NVIDIA’s DRIVE Hyperion 10 platform in the meantime.
Central to all of it is what Hyundai calls its Data Flywheel, a system that takes real-world driving data from its roughly seven million annual vehicle deliveries and automatically turns it into training data, validation tests, and over-the-air updates. The pipeline pays special attention to moments when something goes wrong or nearly does, like sudden disengagements or hard braking events, and feeds those straight back into retraining. Hyundai’s 42dot division is also experimenting with 3D Gaussian Splatting to recreate real-world scenes for testing, and the group is preparing a Level 4 pilot in Gwangju with South Korea’s Ministry of Land, Infrastructure and Transport before the year ends. Looking further out, 42dot is researching a Vision-Language-Action model designed to help cars reason their way around tricky, unstructured situations, the kind of double-parked-car-in-a-narrow-street problem that has stumped self-driving systems for years, with real-world testing expected to expand in early 2027.