Firstly, this isn't direct at you, more at the title. I think the title of "47 events" is greatly misleading. While the simulations are definitely incredibly valuable, it's misleading to use the figures from simulations in any argument for how careless Waymo is. In fact, I think the heavy usage of simulation to model realistic events should be a plus, rather than how the title makes it seem like a bad thing.
Secondly, If you read Table 1 [1] of the paper, you'll see that there was only 1 actual pedestrian event (and I guess 2 simulated events), and that was a person walking into a parked car (it's funny that they could measure the walking speed of the person at 2.7mph).
Lastly, in Table 2 they enumerate all the events that "were the fault of the other human driver". Most of them are "failure to yield", for example-- "failure to yield to a vehicle approaching from the left while making a right turn at an unsignalized intersection."
This touches on your bit on driver's responsibility. While it is true that Waymo isn't at fault, it makes me wonder how much "defensive driving" Waymo is including in their models. In the case above, without video/full context, you don't fully know whether it was an avoidable accent or not by, say, swerving to the other lane, or if that option wasn't possible due to another car. It'd be incredibly interesting to know the exact number corrective actions taken/attempted and number of probable accidents avoided.
> I think the title of "47 events" is greatly misleading
If I understand correctly, it's not the title that's misleading but the term "simulation." They're only talking about simulating what would have happened if the safety driver didn't intervene in real life, not simulations generally. A better word would be predictions -- they predict that there would have been 47 events, but safety drivers were able to prevent the majority of them.
Fair enough. I did not read the actual paper. My bad.
>that was a person walking into a parked car (it's funny that they could measure the walking speed of the person at 2.7mph).
And heh. I can't say the car's likely at fault in that case although it may still have behaved in a way the pedestrian didn't expect. And as someone else noted, some of the language I was quoting was from The Verge. The actual report is much more just the facts.
To your broader point, a human driver could probably obey every traffic law and get into tons of at least minor accidents that would be technically the other driver's fault.
Secondly, If you read Table 1 [1] of the paper, you'll see that there was only 1 actual pedestrian event (and I guess 2 simulated events), and that was a person walking into a parked car (it's funny that they could measure the walking speed of the person at 2.7mph).
Lastly, in Table 2 they enumerate all the events that "were the fault of the other human driver". Most of them are "failure to yield", for example-- "failure to yield to a vehicle approaching from the left while making a right turn at an unsignalized intersection."
This touches on your bit on driver's responsibility. While it is true that Waymo isn't at fault, it makes me wonder how much "defensive driving" Waymo is including in their models. In the case above, without video/full context, you don't fully know whether it was an avoidable accent or not by, say, swerving to the other lane, or if that option wasn't possible due to another car. It'd be incredibly interesting to know the exact number corrective actions taken/attempted and number of probable accidents avoided.
[1] https://storage.googleapis.com/sdc-prod/v1/safety-report/Way...