You can read the California DMV's set of autonomous vehicle accident reports.[1] Almost all the Waymo reports are "vehicle was entering intersection, detected cross traffic, stopped, was rear-ended by human driver". There's one Waymo report where someone ran a stop sign and hit them.<p>The rear-ending problem will be solved as automatic emergency braking becomes standard on cars. Already, it's shipping on almost all high-end cars and 70-80% of midrange cars. There's a US auto industry goal of it being standard by 2022. That's probably the main feature needed for self-driving cars to coexist with human-driven cars.<p>[1] <a href="https://www.dmv.ca.gov/portal/vehicle-industry-services/autonomous-vehicles/autonomous-vehicle-collision-reports/" rel="nofollow">https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...</a>
They can use it to start fixing the roads, not the cars.<p>Just look at the cross intersection they use as an example, typical of the US. The ROAD is the death trap there. People are crossing each other at high speed, this is the BUG, and should be fixed.<p>The proper solution is putting there a roundabout, and you will have instantly less fatal accidents.<p>The US was the first to use roundabouts, but those early designs did not work well, and as a result we have those monstrosities like in the picture all around the US.<p>With electric cars that record accidents, the first bugs that we should fix are those in the roads, like those concrete barriers that have no smooth transition but a front wall out of nowhere.
I really dislike the way the data is being summarized as "100% avoided or mitigated*".<p>First, they're omitting the types of case they're failing to mitigate (rear endings in this case). Cue "60% of the time it works every time" anchorman scene.<p>Second, "mitigated" is defined to mean a 25% relative reduction in chance of serious injury. Which is good, but again smacks of "60% of the time it works every time".<p>Basically, it looks to me like someone wanted to be able to show a bold blue 100% and required the definitions they were using to be massaged to match. And apparently they got their way. This has damaged my trust in future summaries put out by Waymo.<p>A more truthful summarization style would be e.g. "X fewer serious injuries over 72 incidents", referring to expected injuries and simulated incidents of course.
This is really cool.<p>I suspect that when we have self driving cars, it's going to take off very quickly due to insurance. In particular, if you are in a non self driving car and in an accident with a self driving one, it'll be hard to not be shown as at fault. So insurance I would guess go up based on the percentage of self driving cars, and likely be really cheap for self driving cars.<p>I wonder if there's a period though where people cut off self driving cars, knowing they won't get hit. I suspect pedestrians in cities will jaywalk a lot more (and maybe that's ok).
A big caveat to this study is that, in the reconstructed simulations, once the self-driving car (SDC) deviates from what the human-driven car did, the behavior of all other agents becomes unrealistic. Often in these sims, instead of reacting to the new SDC trajectory, they replay their original behavior.<p>Ideally, once the SDC deviates, all other agents are simulated as well. A tall order, but necessary if these counterfactuals are to hold weight.
> Road safety is a major, global public health crisis. More than 1.3 million people die on the world’s roads every year, according to WHO. That’s more people than die from HIV/AIDS, and is equivalent to a passenger plane’s worth of people crashing every single hour—or one death every 30 seconds.<p>> I’ve spent over 20 years working in crash avoidance research, in the belief that improved driving technology is the key to reducing these needless deaths.<p>This is a noble goal, but I find it really damn hard to ignore the possibility that this is all just another ploy by Google to steal our data and violate our privacy. Just imagine trying to push back against a technology that saves lives for something as "frivolous" as privacy.<p>Why does it seem like the consumer can never win? The free market is supposed to be self balancing, yet I can't remember a time where it actually felt that way in the technology sector.
1) Would a random driver (random steering and throttle) also have avoided these crashes? It seems like they were fatal because several variables coincided very exactly, and really any other input would have avoided them.<p>2) What input is the Waymo driver using here? If these were not originally Waymo cars, then is it perceiving simulated video or getting raw access to ground truth vehicle positions?
This shows (sort of) that the computer does better in situations where humans are known to fail. It's entirely likely that the computers will have new and unexpected failure modes that humans would never run into.
Other humans put in place of the ones who caused the accidents would probably have a similar high avoidance rate.<p>It just shows Waymo was better than a small sample.
Figure 3 has the responder slowing unusually soon. Makes me wonder if their model is overly fit to the sample. Still, if it's genuinely that good at responding and mitigating then the future looks bright.
Another thing that would avoid fatal crashes is reducing the number of cars on the roads, making them smaller and making them slower. That's a lot cheaper than making them more complex.
What do you call that when you subconsciously suppress any tests or scenarios you know will break your code? Like you notice somethings kinda racey but it involves other components you don't control and so you don't do the parallel tests you probably should, that kinda thing.<p>Because that is exactly the bad feeling I get with Waymo never actually testing anywhere that doesn't have mm-precise maps and the all important California weather. There seems to be this implicit understanding that when you drop the thing in light snow it will just careen off the side and nobody is particularly incentivized to find that yep, that is exactly what happens.<p>(Even in these simulated scenarios here, you got what looks like a barren wasteland and exactly two cars)