How Tesla's vision system "sees" a railcar

Tesla's sensor-only approach fails in edge cases and is highly unreliable. Here, it sees something distinctly different from a train.

lol, a clip of a Tesla's vision system trying to understand a train labyrinth.zone/notice/Aq5V7…

[image or embed]

— Ketan Joshi (@ketanjoshi.co) January 15, 2025 at 6:23 AM

Children understand the concept of a train and a train of railcars by the time they are very young; trains are popular toys, not unlike Tesla's FSD.

The NHTSA's Office of Defects Investigation (ODI) officially opened a "preliminary evaluation" of the technology in October 2024 after receiving several reports of Teslas wrecking after entering areas of reduced visibility. According to the NHTSA, one collision resulted in a pedestrian fatality, and at least one other accident reportedly caused an injury.

In an information request addressed to Eddie Gates, Tesla's director of field quality, the NHTSA says the investigation "will focus on the adequacy of FSD's ability to perform in reduced visibility conditions stemming from relatively common traffic occurrences such as sun glare, fog and airborne dust" following the reported crashes.

The NHTSA says these crashes could at least partly be due to the fact that the drivers didn't know the limits of the cameras' vision and what level of responsibility they were supposed to take behind the wheel of a car with active FSD.

GlassBytes

Previously:
Tesla patents 'Full Self Sanitizing' taxis
Court finds Tesla's 'Full Self Driving' to be 'corporate puffery'
Tesla engineers focused on making 'Full Self-Driving' work for Elon
Here's what full self driving autopilot 'sees'