In autonomous and self-driving vehicle news are Stellantis, Wayve, XPENG, TIER IV, Avride, Sodexo Campus, Helm.ai, Pilot & Tesla.
Stellantis and Wayve Put One AI Driver in Two Very Different Cars
Stellantis and Wayve demonstrated supervised door-to-door hands-free driving on development prototypes of the Fiat 500e electric city car and the Maserati Grecale premium SUV at the Wave by Vento event in Turin. The companies integrated the Wayve AI Driver with the Stellantis STLA AutoDrive platform in under four weeks per vehicle, a pace they say validates an embodied AI model built to generalize across different platform geometries, dynamic envelopes and sensor suites, without bespoke code or high-definition mapping.
The Level 2++ automated driving solution is scheduled for commercial availability in North America in 2028. The Turin demonstration follows initial testing on a Jeep Grand Cherokee earlier in 2026 and advances a partnership established in May. Wayve uses an end-to-end foundation model trained on real-world driving data, an approach intended to eliminate hand-coded rules, HD maps and specialized compute silicon constraints. Stellantis intends to scale the same intelligence across consumer applications and future SAE Level 4 driverless commercial deployments built on its L4-Ready platform hardware.
XPENG Brands Its Robotaxi Service XPENG YOYO
XPENG has named its mass-produced Level 4 ride-hailing business XPENG YOYO and opened public onboarding in China through invitation-code registrations and a dedicated portal. The move shifts the automaker’s autonomous mobility program from internal technical verification and closed employee pilots into commercial-scale fleet validation. The brand is meant to provide unified identity across domestic and forthcoming international markets, and the service uses vehicles that rolled off XPENG’s Guangzhou manufacturing lines earlier this year.
Each factory-integrated vehicle runs a pure-vision driving stack governed by the VLA 2.0 foundation model, operating without LiDAR or high-definition maps. The compute architecture is built around four proprietary in-house Turing AI chips that together deliver 3,000 TOPS. XPENG plans an asset-light commercial model in which it supplies vehicle platforms, autonomous software and AI hardware, while regional mobility operators handle offline fleet management.
TIER IV Takes Autoware-Based Buses to Saudi Arabia
TIER IV has been selected by Japan’s Ministry of Economy, Trade and Industry for a Global South collaborative technology grant to run autonomous bus field operational tests in Saudi Arabia. Working with a European OEM on chassis integration, the open-source software developer will adapt its Autoware-based platform and neural network perception models to local conditions, including right-hand traffic and extreme desert heat. Testing will begin in controlled environments before moving into mixed-traffic public roads.
The project aligns with the Saudi Vision 2030 goal of automating 15 percent of the Kingdom’s public transit infrastructure. It builds on a framework first established through a memorandum of understanding with the digital platform provider Elm, and TIER IV will work directly with local transport authorities and fleet operators to validate commercial business models. The grant also funds local engineering talent pipelines, including competitive STEM programs and regulatory knowledge transfer.
Avride and Sodexo Campus Bring Delivery Robots to College Quads
Avride has entered an agreement with Sodexo Campus to deploy sidewalk delivery robots at higher education institutions across the United States. The partnership creates a standardized operating framework intended to cut deployment timelines from months to weeks across Sodexo’s network of roughly 425 partner colleges and universities, extending automated meal delivery beyond fixed dining halls.
The Avride rovers use onboard sensors and machine learning models to navigate busy walkways, academic corridors and residential quads. The platform has completed more than 600,000 commercial deliveries to date and currently operates on several major campuses, including The Ohio State University, Indiana University Bloomington, the University of Arizona and Salisbury University.
Helm.ai Reports $70 Million in Commercial Contracts
Helm.ai says it has secured $70 million in commercial contracts over the past 12 months from global automotive OEMs, Tier 1 suppliers and industrial automation companies, and is tracking toward operational breakeven ahead of automotive start of production. Its software targets production vehicle platforms ranging from Level 2+ through Level 4 driving systems, as well as automated sensor data labeling and generative simulation. Commercial deployments also reach heavy industrial off-highway automation, including vision perception systems for open-pit mining equipment.
The platform is built on a proprietary unsupervised training method called Deep Teaching, which produces physical foundation models that separate understanding of the environment from policy action, a design meant to reduce training data volume and runtime compute needs. Rather than building specialized architectures for each type of machine, Helm.ai uses a single model lineage, training pipeline and validation stack across passenger software-defined vehicles, robotics platforms and off-road industrial machinery.
Witherite Law Group’s Analysis Counts Austin’s Robotaxi Incidents
Municipal tracking data from Austin, Texas, covering July 2023 through September 2026, documents more than 320 autonomous vehicle incidents. An analysis released by traffic safety advocate Amy Witherite of Witherite Law Group found that collisions account for only 4 percent of the total, and that most friction stems from interference with public safety rather than mechanical crashes.
The city’s dashboard classified 32 percent of reports as active safety concerns, 17 percent as vehicle stalls or idling nuisances, 15 percent as roadway obstruction, 13 percent as near misses and 9 percent as failures to follow law enforcement hand signals or commands. Reports rose from 21 in 2024 to roughly 135 in 2025, and more than 110 were logged in the first nine months of 2026.
The findings point to continuing challenges around remote event response protocols, manual disengagement of vehicles by emergency responders and detection of school bus stop arms. Regulatory pushback is building elsewhere as well: San Diego City Council petitions seek municipal control over the movement of empty fleet vehicles, commercial authorization is stalled in Washington, D.C., and the Minneapolis city council has voted to require human safety operators.
DrivingBench Tests Frontier AI Models on a Real Cone Course
Researchers Aditya Ramabadran, Simon Mahns and Tobias Gessler have introduced DrivingBench, a robotics evaluation platform that tests whether frontier vision-language models can drive a physical vehicle through a cone course. The testbed uses a 2022 Toyota Corolla fitted with Comma hardware and the OpenPilot actuation stack, with two onboard camera streams connected to the models through the Model Context Protocol. The models issue discrete longitudinal and lateral commands, such as set_motion and stop_now, under real-world inference latency, in a low-speed proving ground with a human safety driver able to override.
Results varied widely. GPT-6 Astra completed the full course on its second attempt in 5 minutes 22 seconds at low speed, adjusting its steering in context after an initial lateral deviation. Claude Fable 5.1 completed 45 percent of the course, Grok 4.6 completed 11 percent and GPT-5.6 Sol completed 6 percent. Inference lag of 5 to 13 seconds per turn created notable bottlenecks. The research group has released the test harness, prompt structures, course maps and telemetry traces on Hugging Face to support embodied agent research.
Pilot and Tesla Energize the First Megacharger Hub
Pilot and Tesla have commissioned their first commercial Megacharger installation, at a Pilot travel center in Ellabell, Georgia, the first site in a high-power charging network designed for heavy-duty freight. The installation has six dedicated stalls supported by Tesla V4 cabinet power electronics delivering up to 1.2 MW per dispenser. At that output, Class 8 Tesla Semi trucks can recover about 60 percent state of charge in 30 minutes.
Ellabell begins a multi-state buildout stemming from a joint initiative announced in January 2026 that targets freight corridors including I-5 and I-10 across California, Georgia, Indiana, Nevada and Texas. Twelve additional high-power travel center locations are under construction, and initial commissioning is underway at Flying J and Pilot sites in Tulare, California; Fort Stockton, Texas; and North Las Vegas, Nevada. Each site will have four to eight heavy-duty stalls, forming Megawatt Charging System infrastructure intended to speed fleet electrification on regional and long-haul routes.
FMCSA Grants Five-Year Warning Beacon Exemption for Autonomous Class 8 Fleets
The Federal Motor Carrier Safety Administration has granted a five-year regulatory exemption allowing commercial autonomous vehicle operators to deploy cab-mounted amber warning beacons in place of traditional reflective triangles during roadside stops. Effective through October 2031, the ruling covers registered SAE Level 4 carriers, including Aurora Innovation, Kodiak Robotics, Waabi, and Stack AV, effectively replacing a series of rolling three-month waivers issued since late 2025. Beacons must activate within five minutes of stopping, stand at least 100 inches high, and operate exclusively within defined operational design domains with strict exclusions on hazardous materials, passenger transport, and triple-trailer configurations.
The regulatory reversal addresses a December 2024 denial of a joint petition from Aurora and Waymo, resolving agency concerns by requiring carrier pre-notification, rigorous hardware specifications, five-day crash reporting for incidents involving active beacons, and mandatory annual malfunction logs. Federal preemption prevents states from enforcing disparate roadside warning rules on participating interstate autonomous freight operations, establishing uniform compliance as operators accelerate commercial scale.
Agency analysis cited operational stopping distance evaluations showing comparable driver reaction thresholds between cab-mounted beacons and manual triangles, noting that broader warning requirements across commercial trucking may see future rulemaking updates. The five-year deployment framework functions as an operational testbed for federal regulators, generating continuous telematics, crash, and diagnostic data across expanding commercial fleets as developers scale toward multi-hundred-truck operations.