Tesla's Self-Driving Software Hits More Snags, On and Off the Road
Los Angeles County sheriff's deputies are investigating a crash in which a Tesla operating in self-driving mode lost control and struck something other than the plan its software had drawn up, according to officials who responded to the scene. The crash, reported this week, adds a concrete, skid-marked data point to a question that has dogged Tesla for years: how much control the company's cars actually have when "Autopilot" or "Full Self-Driving" is doing the steering.
The timing is awkward. Tesla has spent 2024 trying to convince regulators, investors, and the driving public that it is on the cusp of solved autonomy, rolling out a limited Robotaxi service meant to showcase exactly that. Instead, a former Tesla employee has offered a different kind of cautionary tale — not a crash, but a comedy of errors. The ex-employee recounted a Robotaxi ride that should have taken a few minutes across downtown instead stretching to 43 minutes, as the app's routing logic apparently talked itself into a detour with no obvious justification. No one was hurt. No metal was bent. But for a company selling the promise of software that drives better than people do, a car that gets confused about how to get three blocks is its own kind of failure.
Together, the two incidents sketch a familiar pattern for Tesla watchers: dramatic failure on one end, baffling inefficiency on the other, both emerging from the same black box of code the company insists is improving with every mile driven. The L.A. County crash is the more serious of the two, since it raises the plain question of safety — whether a car that can lose control on a public road, even briefly, belongs on that road while a human driver may or may not be positioned to correct it in time. The Robotaxi detour is less dangerous but arguably more revealing, because it happened in the exact service Tesla has staked its near-term reputation on, with no human behind the wheel to quietly override the computer's bad judgment.
For the robotics and autonomous-vehicle industry generally, these are not small matters. Tesla's approach — camera-based perception without the lidar sensors that competitors like Waymo rely on — has long been a bet that software and cameras alone can match or beat more heavily instrumented systems, at a fraction of the cost. Every crash investigated by local police, and every anecdote of a robotaxi behaving strangely, becomes ammunition for skeptics of that bet, including regulators who have already opened inquiries into Tesla's driver-assistance systems. For riders and other drivers sharing the road, the stakes are more immediate: a 43-minute detour is an inconvenience, but a car that cannot reliably hold a lane is a hazard that predates any marketing copy about the future of transportation. Tesla has not detailed what caused either incident, and the investigation into the crash continues. What's clear is that the gap between the company's stated ambitions and its software's daily performance remains wide enough to drive a very confused car through.