I’m not sure how useful that analysis is. Most (all?) of those miles are in the city where most collisions are lower speed, so it’s not really a measurement of how safe they are.
The analysis also seems to go in with the assumption that accidents are other drivers’ fault unless proven otherwise. If an AV rear-ends another vehicle, it’s because the other vehicle “cut off the AV” and is therefore the other vehicle’s fault. But if another car rear-ends the AV, it must be the other vehicle’s fault as well. Is that true? Did the AV make an abrupt lane change? Did the AV initiate an unexpected hard stop for no reason?
The article describes an accident as "a driverless Waymo was trying to turn left, but another car proceeded into the intersection from the left and made contact with the left side of the Waymo AV,” implying that the other car was at fault. But that description sounds to me like the AV was trying to turn left from the center lane, like in some of the videos linked above.
I have no doubt that under standard conditions, AVs are much safer than drivers. But I think it’s premature to determine if they are overall safer.
If humans only drove in the weather conditions that self-driving cars did it would be meaningful to compare the two figures. Since humans drive in all kinds of weather and we don’t have stats for their accidents when only driving in good weather, it doesn’t seem that meaningful of a comparison.
Also, from what I understand we are only dealing with data that Waymo and Cruise are self reporting. That seems a lot like relying on the tobacco industry for reports on the safety of smoking.
I am surprised there aren’t any official statistics from the city/state … on this non-trivial topic.
I’d also NOT be surprised if the companies does some “hush-hushing” on minor incidents and accidents to keep it off the books.
Just as an example:
German’s AAA is called ADAC. And ADAC has a good statistics on what type (make, model) car they have to “rescue”. Volkswagen set up a parallel “stranded-car-rescue-service” for VW’s only of course …
so they were (are?) constantly under-reported in the official stranded statistics, and used this distortion to bolster their reputation as “VW is a dependable car”.
Just taking a step back, it’s sorta remarkable to watch these propaganda campaigns at work. We have several powerful groups with differing interests. We have the multi-billion dollar AV companies. There are the politicians, which don’t quite serve the public but like to be seen as if they were. There are various labor groups such as firefighter unions. There’s the existing pool of taxi/rideshare drivers. There’s the pool of social media types, led by those with the hottest and least-informed takes. And there’s the media, which at best thrives on clickbait, and at worst acts as a propaganda arm for one of the other groups.
Watching all of them battle it out is… well, I dunno what, but it’s sure something. Ultimately, though, it’ll be decided by the tech.
Ars Technica drive a level 3 (hands off, but only available in certain conditions) car
Conditions include < 40 MPH, dry and clear, and followable car.
Two notes:
Has been used in Germany for over a year without incident
Mercedes will assume liability (of course with disclaimers)
This is a good article, thanks for posting it. I posted upthread when this vehicle was approved in Nevada, and my comment was….
Apparently this requires slightly less attention from the driver than other cars. Honestly, though, it feels like just splitting hairs, and that someone got some bullshit past the regulators.
…but on further review, I think that’s completely unfair - progress on this will be incremental, and this is a good example of that, and it’s clearly level 3.
I have high hopes for fully autonomous vehicles. I felt comfortable driving after one year and after 10 years I think I was fully proficient. At that point I had encountered snow, black ice, planing due to water on the road, dirt roads, mud, highways, stop and go, blown tires, deer in the road, and Boston. Let’s say I drove two hours a day for those 10 years. That’s 73,000 hours of experience.
Once we have 100,000+ self driving cars on the road they will be getting that much experience every day. If they report back anomalous situations so that the software can be updated, I think the capabilities will rapidly improve Every day some car will be encountering emergency vehicles, snow, ice, mud, rain, drunk drivers, animals in the road, blown tires, failing traffic lights, Boston, and other difficult situations.