Another robotaxi developer faces a competence test from regulators
The U.S. National Highway Traffic Safety Administration has opened an investigation into Avride, the self-driving vehicle company partnered with Uber, after identifying 16 crashes linked to the performance of the company’s autonomous system. According to the supplied report, the crashes include one minor injury and raise concerns about how the system handles lane changes, nearby vehicles, and stationary obstacles.
All of the incidents cited by regulators occurred while a safety monitor was seated in the driver’s seat, a detail that sharpens the central question of the probe: if a human supervisor was present, why did these crashes still happen?
What regulators say went wrong
The National Highway Traffic Safety Administration’s Office of Defects Investigation said the identified crashes relate to “the competence of” Avride’s self-driving system. Specifically, the report says the system appears to have struggled when changing lanes, when responding to other vehicles in the same lane, and when reacting to stationary objects.
Those are not edge-case complaints. They are basic competencies for any road-driving automation stack. A failure pattern across those categories suggests regulators are looking at foundational system behavior rather than an isolated anomaly.
Avride declined to explain, according to the source, why safety monitors did not intervene in the crashes. The company did say it had implemented targeted technical and operational mitigations based on each incident reported between December 2025 and March 2026, and that incident frequency relative to mileage had been declining even as operations expanded.
Why the Uber connection matters
Avride is best known for sidewalk delivery robots, but it has spent years developing self-driving cars and partnered with Uber in 2024. In 2025, Uber and Avride parent company Nebius agreed to strategic investments and other commitments worth up to $375 million, according to the report.
That means the NHTSA investigation lands at a commercially sensitive moment. Uber only recently began offering rides in Avride robotaxis in Dallas, Texas, where many of the reported crashes occurred. Some incidents also took place in Austin, and at least one crash involved a vehicle carrying a passenger.
For Uber, the probe is a reminder that autonomy partnerships can import not just technology upside but regulatory exposure. For Avride, it is a test of whether the company can scale beyond pilot-stage enthusiasm into a safety case strong enough to satisfy federal scrutiny.
The broader autonomous-vehicle signal
NHTSA’s standing order on automated driving requires companies to report certain crashes, and Avride noted that it had submitted the incidents as required. That reporting framework is doing what it was intended to do: creating a record that regulators can use to identify patterns before a more serious incident forces the issue.
The central problem for the robotaxi sector is that public confidence is shaped less by marketing claims than by visible operational failures. Every crash tied to lane management or object response reinforces the view that autonomy remains brittle in common traffic scenarios. Human safety monitors can mitigate that risk only if they intervene reliably and early enough, and the source offers no explanation for why that did not happen here.
What comes next
An investigation does not establish fault on its own, but it raises the bar for evidence. Avride will need to show not just that it understands each incident, but that its system-level fixes address the recurring behaviors regulators identified. In autonomous driving, assurances about improvement are common. Demonstrating robust improvement under scrutiny is much harder.
For now, the takeaway is straightforward: a company backed by Uber and growing in major Texas markets is under federal review because its self-driving system may not yet be dependable in some of the most basic situations road vehicles face every day.
This article is based on reporting by TechCrunch. Read the original article.
Originally published on techcrunch.com
