This is just one part of the autonomous and self-driving vehicle news this week, we have a definitive round-up and an updated round-up Two of CES autonomous/ai news. In other AV news are XPEENG, Seeing Machines, Waymo, Pony.ai and CA.
XPENG and Peking University Paper
XPENG, in collaboration with Peking University, has had its paper “FastDriveVLA: Efficient End-to-End Driving via Plug-and-Play Reconstruction-based Token Pruning” accepted to AAAI 2026, one of the world’s leading AI conferences, with an acceptance rate of just 17.6 percent. The research introduces FastDriveVLA, a novel visual token pruning framework for end-to-end autonomous driving Vision-Language-Action models that mimics how human drivers focus on critical visual information while ignoring irrelevant background data.
FastDriveVLA addresses a key challenge in VLA models: the heavy computational burden caused by processing thousands of visual tokens in real time. By using a reconstruction-based, adversarial foreground-background strategy, the framework selectively retains the most valuable visual tokens, significantly reducing onboard computation without sacrificing driving performance. On the nuScenes benchmark, the method achieved state-of-the-art results, cutting visual tokens from 3,249 to 812 and delivering nearly a 7.5x reduction in computational load while maintaining high planning accuracy.
The AAAI 2026 acceptance marks XPENG’s second recognition this year at a top-tier global AI conference, following its invitation to speak at CVPR WAD and the launch of its VLA 2.0 architecture at AI Day. Together, these milestones underscore XPENG’s full-stack in-house AI capabilities and its ongoing commitment to advancing toward L4 autonomous driving and large-scale deployment of physical AI in vehicles.
Seeing Machines Launches Future Mobility Group
Seeing Machines has established a dedicated Future Mobility Group to meet growing demand from the autonomous driving sector as programs transition from development to commercial deployment. The new team will support autonomous vehicle customers across the full lifecycle—from development through large-scale rollout—by embedding Seeing Machines’ next-generation driver and occupant monitoring systems into autonomy platforms. Building on the success of its Guardian solution, already deployed in more than 1,000 self-driving development vehicles, the initiative underscores the company’s focus on human-centered safety, interior sensing, and scalable solutions for robotaxi, logistics, and remotely supervised autonomous fleets worldwide.
Waymo Robotaxis Take Parking Spots in San Francisco
Residents in San Francisco’s Mission District reported that Waymo autonomous vehicles are increasingly taking up public parking spaces, often while idle and not picking up riders. The trend highlights how expanding AV fleets are interacting with urban infrastructure and is prompting discussions about parking management and regulatory response. Residents note that the vehicles are taking free parking spaces which they would like to use themselves.
Waymo 3K Vehicle Software Recall
Waymo initiated a recall of 3,067 autonomous vehicles after a software flaw led some robotaxis to drive past stopped school buses with flashing red lights and extended stop arms, violating traffic laws. The recall follows reports and an NHTSA probe into the incidents, prompting software fixes and further safety evaluations
Pony.ai Surpasses Annual Robotaxi Fleet Target
Pony.ai announced that its robotaxi fleet reached 1,159 vehicles at the end of 2025, exceeding its goal of deploying 1,000 autonomous vehicles in commercial service.
California Law Allows Issuing of Violations to Driverless Cars
A new California law (AB 1777) set to go into effect in 2026 will allow notices of autonomous vehicle noncompliance to be issued to manufacturers when driverless cars are involved in traffic violations or hinder emergency responders. The legislation also requires AVs to have two-way communication systems, addressing gaps in current enforcement capabilities.