테슬라 뉴스
Tesla Discloses Two Teleoperator-Controlled Robotaxi Crashes to NHTSA Amid Heightened Scrutiny
~에 의해
Rio
~에
May 16, 2026
Quick Summary: Tesla Discloses Two Teleoperator-Controlled Robotaxi Crashes to NHTSA
-
Disclosure: Tesla unredacted 17 Robotaxi incident reports filed with NHTSA — two crashes specifically involved human teleoperators; 15 others also disclosed
-
Crash 1 (July 2025): Teleoperator repositioning a stopped vehicle — vehicle mounted curb and contacted a metal fence during a low-speed left turn; no passengers aboard; minimal damage
-
Crash 2 (January 2026): Teleoperator took control from stopped position after safety monitor requested navigational support — collided with a temporary construction barricade at ~9 mph; front-left fender and tire scrapes; no passengers aboard
-
Teleoperator limits: Authorized for low-speed operation only (under 10 mph); primary function is repositioning maneuvers; not a failure of the ADS — a planned fallback architecture
-
Tesla's stated purpose: "This capability enables Tesla to promptly move a vehicle that may be in a compromising position, thereby mitigating the need to wait for a first responder or Tesla field representative to manually recover the vehicle."
-
Full 17-incident picture: Majority of incidents involved Robotaxi being struck by other human drivers; ADS-at-fault incidents include: 2x clipped parked car mirrors; 1x struck dog (unharmed); 1x hit thin metal chain during unprotected left turn
-
Context: Cybercab entering mass production — these disclosures are part of the transparency approach that builds the regulatory trust needed for scale
Tesla has unredacted 17 Robotaxi incident reports filed with NHTSA since the Austin launch, including two crashes that occurred while vehicles were under teleoperator control. Both were low-speed, minor collisions with no passengers aboard. Here's the full breakdown of each incident, the teleoperator architecture, the complete 17-incident picture, and what it means for the Robotaxi program's trajectory.
"This capability enables Tesla to promptly move a vehicle that may be in a compromising position, thereby mitigating the need to wait for a first responder or Tesla field representative to manually recover the vehicle." — Tesla, NHTSA filing
The Two Teleoperator Crashes: Incident Details
| Element |
Crash 1 — July 2025 |
Crash 2 — January 2026 |
| Trigger for teleoperator |
Vehicle stopped autonomously and could not determine next move (disengagement event) — teleoperator assumed control to reposition |
Onboard safety monitor requested navigational support due to unforeseen or complex environment ahead — teleoperator took command from stopped position |
| What happened |
During gradual acceleration and left turn toward side of road, vehicle mounted curb and contacted a metal fence |
Teleoperator proceeded forward and collided with a temporary construction barricade at approximately 9 mph |
| Damage |
Minimal — low-speed contact with fence |
Scrapes to front-left fender and tire |
| Passengers |
None aboard |
None aboard |
| Root cause |
Teleoperator precision challenge — limited field of view and distance judgment through remote interface during low-speed repositioning |
Teleoperator perception challenge — difficulty detecting temporary construction barricade from remote location; temporary obstacles are harder to perceive via video stream |
The Teleoperator Architecture: How It Works and Why
| Element |
Detail |
| What it is |
A planned fallback architecture — not a failure of the Autonomous Driving System (ADS); designed to handle edge cases and scenarios outside the current programming of the self-driving software |
| Speed limit |
Under 10 mph — teleoperators authorized for low-speed operation only; primary function is repositioning maneuvers in awkward or challenging situations |
| Purpose |
Prevent Robotaxis from becoming obstructions; handle the "last 1%" of driving challenges that still vex advanced AI; avoid waiting for a first responder or Tesla field representative to manually recover the vehicle |
| Known challenges |
Video stream latency; limited field of view vs. being physically in the driver's seat; difficulty judging distances and speeds through a screen; temporary obstacles harder to perceive remotely — both crashes provide data to improve teleoperator training, interface, and underlying technology |
The Full 17-Incident Picture: Context Beyond the Headlines
| Incident Category |
Detail |
Fault |
| Struck by other human drivers |
Majority of the 17 incidents — Robotaxi hit by human road users; reflects the defensive driving capabilities programmed into autonomous systems and the unpredictability of human driving |
Other driver |
| Teleoperator crashes (x2) |
July 2025 (curb + fence) and January 2026 (construction barricade) — detailed above; both low-speed, no passengers, minimal damage |
Teleoperator |
| Clipped parked car mirrors (x2) |
ADS clipped side mirrors of parked cars — suggests challenges with navigating narrow spaces and precisely judging the vehicle's perimeter |
ADS |
| Struck dog (September 2025) |
Dog darted unexpectedly into the road; Robotaxi struck the dog — dog escaped unharmed per the report |
ADS (edge case) |
| Hit thin metal chain |
Vehicle making unprotected left turn into parking lot struck a thin metal chain — a perception challenge for vision-based systems across the industry; thin, low-contrast obstacles are difficult to detect |
ADS (perception) |
Competitive Context: Tesla vs. Waymo and Zoox
| Dimension |
Tesla Robotaxi |
Waymo / Zoox |
| Total reported crashes |
Lower raw number — 17 incidents total since Austin launch |
Higher raw number — but reflects longer operational history and significantly more autonomous miles accumulated |
| Operational maturity |
Newer service — Austin launch 2025; measured, geofenced rollout; safety monitors present during expansion phases |
Longer commercial history — Waymo has been running commercial services for years; more complex urban environments; more accumulated miles |
| Transparency approach |
Proactive unredaction of 17 incident reports — including teleoperator crashes; signals commitment to regulatory transparency as a trust-building strategy |
Established reporting frameworks with NHTSA and state regulators; longer track record of regulatory engagement |
| Platform trajectory |
Cybercab entering mass production — purpose-built platform with no steering wheel or pedals; the long-term vehicle for scale |
Waymo 6th gen Jaguar I-PACE fleet; Zoox purpose-built bidirectional vehicle — both purpose-built for AV operation |
Conclusion
Key Takeaways
-
The crashes: Two low-speed teleoperator-controlled incidents — July 2025 (curb + fence) and January 2026 (construction barricade); no passengers; minimal damage; both provide data to improve remote operation
-
The architecture: Teleoperator control is a planned fallback, not an ADS failure — under 10 mph; repositioning only; designed to prevent vehicles from becoming obstructions
-
The full picture: 17 incidents total — majority caused by other human drivers; ADS-at-fault incidents include parked car mirrors, dog strike (unharmed), and thin metal chain; a realistic picture of current capabilities and limitations
-
The transparency: Proactive unredaction of all 17 reports signals Tesla's commitment to regulatory transparency — essential for building the long-term trust required for a driverless future at scale
-
The trajectory: Cybercab entering mass production — these disclosures are the data foundation that informs the next generation platform's design and the regulatory framework for its deployment
The two teleoperator crashes are not evidence that Tesla's Robotaxi program is failing — they are evidence that it is operating exactly as designed, with a human fallback layer that catches edge cases the ADS cannot handle, and a transparency framework that discloses what happens when that fallback layer also encounters its limits. The data from these incidents will improve teleoperator training, interface design, and the underlying perception systems. That is how iterative development works — and it is the same methodology that produced the 3-meter splashdown accuracy of Starship Flight 10 and the reusable rockets that the industry once said were impossible.
