3 ms·
Possible reasons you might find more slow-speed minor rear-end collisions among self-driving cars than among human-driven cars on a per-mile basis: 1. There ar
by jacobolus 11y ago
Possible reasons you might find more slow-speed minor rear-end collisions among self-driving cars than among human-driven cars on a per-mile basis:
1. There are only a tiny number of self-driving cars on the road, driving a relatively small number of miles, and accidents in general are fairly rare on a per-mile basis, so there might not yet be enough data to get a reliable idea of average crash rate. The linked study is based on analysis of only 11 (!!) crashes, 8 of which were at <5 mph, with only 2 causing “possible injuries”. As the study itself says, “[...] the corresponding 95% confidence intervals overlap. Therefore, we currently cannot rule out, with a reasonable level of confidence, the possibility that the actual rate for self-driving vehicles is lower than for conventional vehicles.” (Personally I think their confidence intervals for self-driving cars are useless given the data so far, but whatever.)
2. The Google cars which account for all of the cited crashes are AFAIK doing all of their driving on city streets, where intersections, stops, and thus these types of collisions are more common than on highways.
3. I suspect most tiny bumper bumps which don’t damage either car or injure anyone are never reported to the police when both drivers are people, but self-driving cars record every case.
4. Self-driving cars have better sensors to detect road hazards, and as a result might stop in cases where a human driver would do something which risked a more serious accident. People tail-gating these cars will end up unable to react in time and bumping them, but even these 11 crashes might have been in situations where the autonomous car stopping prevented more severe car damage or injuries.
5. Driving cautiously and strictly limiting speed to 25mph might make following human drivers frustrated and more likely to tail closely. This might be exacerbated by the Google cars’ goofy appearance. (Would be interesting to see statistics for various skills/styles of human drivers, or for human-driven NEVs.) I’m not sure NEV accidents can be meaningfully compared to standard car accidents.
Overall I think this topic is worthy of study, but the data is not really all that useful yet. Keith Naughton’s Bloomberg piece calling this a “key flaw” seems like high-order FUD / clickbait.