Tesla Autopilot Crashes into Motorcycle Riders - Why?
FortNine1 HN recommenders · 2022–2024
1.1M views · explainer · dated · Uploaded 2022-09-03
FortNine looks at why Tesla Autopilot has rear-ended motorcyclists, speculating that the vision system misreads motorcycle taillights as a distant car. It argues that radar and other sensors would help.
Cited in threads on motorcycle safety, Tesla's sensor choices and autonomous vehicles as a well-made explanation of the failure mode.
9:31Around The Corner - How Differential Steering Works (1937)
US Auto Industry
54:37How Self-Driving Cars will Destroy Cities (and what to do about it)
Not Just Bikes
17:08Cities Aren't Loud: Cars Are Loud
Not Just Bikes
48:10Bay Area Vision Meeting: Unsupervised Feature Learning and Deep Learning
Google TechTalks
52:22Google Self-Driving Car Project | SXSW Interactive 2016
SXSW
41:18"Voice Driven Development: Who needs a keyboard anyway?" by Emily Shea
Strange Loop Conference
30:09Andrej Karpathy - AI for Full-Self Driving at Tesla
Matroid
2:35:53Tensorflow and deep learning - without a PhD by Martin Görner
Devoxx
4 comments from Hacker News
It may be a contributing factor to recent cases where Tesla Autopilot rear ended motorcycles at night. Thus video has a visualization: [VIDEO] One of these crashes was right by my home, where I rode to work nearly every day prepandemic.
There is a great argument for having ultrasonic sensors and radar in a recent video by Rayan from FortNine discussing two fatal accidents involving tesla autopilot [VIDEO]
I don't really have anything to back this up, so take it with a massive gain of salt, but I seem to keep seeing discussion of Teslas hitting bikes (both cyclists and motorcycles). Most of it centers around there being enough doubt about whether Autopilot was being used, or whether the driver was actually monitoring like they should have been, or whether the cyclist "should" have been in the road, etc, as to "count" as being less safe. I can only imagine that there are impassioned interpretations of the data going multiple directions. A Verge article [0] has some discussion on this with regard to a motorcyclist that follows this pattern, and FortNine also covers this in a characteristically well-made video [1]. There's also the problem of autonomous (electric) cars being too quiet when they first begin moving, but that's easy to solve. So I'm not entirely sure about their safety being "proven", but that's largely because the problem domain is enormous and the data somewhat open to interpretation. Overall "proved" seems too certain a word as of now, without narrowly defining the scope. [0] [LINK] [1] [VIDEO]
Related video "Tesla Autopilot Crashes into Motorcycle Riders - Why?"[0], summarized to: vision used by Tesla seems to process motorcycles differently, and may be incorrectly "assuming" the closer spaced brake lights on a motorcycle is actually a far away car. More details on the homicide here[1], which shows the crash happened during daylight hours and the bike resembles a sport bike. This is a different condition than my referenced video (night collisions with cruiser-style motorcycles), but I suspect similar incorrect assumptions by Tesla vision happened. [0] [VIDEO] [1] [LINK]