Waymo

[powertoold]

The roaring gas engine turns me off

 

[diplomat33]

The roaring gas engine turns me off

The newer Waymo robotaxis are electric. In fact, Krafcik says the plan is to only do EVs for all future robotaxis so that the fleet will become 100% EV.

 

[powertoold]

The newer Waymo robotaxis are electric. In fact, Krafcik says the plan is to only do EVs for all future robotaxis so that the fleet will become 100% EV.

We'll see. When things are changing so fast nowadays, plans don't mean as much Like BMW has been planning to release traffic light recognition since last year and still hasn't deployed it yet. Although they have a new blog article saying "soon":

New BMW Urban Cruise Control recognizes traffic lights

 

[bdub85]

Waymo pulls back the curtain on 6.1 million miles of self-driving car data in Phoenix

 

[diplomat33]

Waymo pulls back the curtain on 6.1 million miles of self-driving car data in Phoenix

It's great that Waymo is being so transparent.

 

[mikes_fsd]

Waymo pulls back the curtain on 6.1 million miles of self-driving car data in Phoenix

Just to be clear that is 6.1 million miles and "Waymo said that it was involved in 18 crashes and 29 near-miss collisions"
6.1 million miles and 18 actual crashes = 1 crash every 338k miles...
6.1 million miles and 47 crashes and near misses = 1 incident every 130k miles.

Also, love the clarity on actual fully driverless vehicles - maybe we will use real statistical data from now on @diplomat33
"Between January and December 2019, Waymo’s vehicles with trained safety drivers drove 6.1 million miles. In addition, from January 2019 through September 2020, its fully driverless vehicles drove 65,000 miles."

 

[diplomat33]

"In the paper, Waymo outlines how “road rule violations” of other drivers contributed to each of the eight “severe” collisions."

In other words, the crashes happened because of bad human drivers. IMO, it just highlights even more the need for autonomous driving.

Hopefully, Waymo has analyzed the data to see if they can improve the defensive driving skills of Waymo to help reduce these crashes caused by bad drivers.

 

[diplomat33]

Just to be clear that is 6.1 million miles and "Waymo said that it was involved in 18 crashes and 29 near-miss collisions"
6.1 million miles and 18 actual crashes = 1 crash every 338k miles...
6.1 million miles and 47 crashes and near misses = 1 crash every 130k miles.

Also, love the clarity on actual fully driverless vehicles - maybe we will use real statistical data from now on @diplomat33
"Between January and December 2019, Waymo’s vehicles with trained safety drivers drove 6.1 million miles. In addition, from January 2019 through September 2020, its fully driverless vehicles drove 65,000 miles."

I've always tried to use real data that was available. And now that we have more statistical data, yes, we should use it. I welcome real statistical data. I want more real data.

 

[DanCar]

Direct link to waymo pdf:
https://storage.googleapis.com/sdc-...Waymo-Public-Road-Safety-Performance-Data.pdf
Higher level page:
Waypoint - The official Waymo blog: Sharing our safety framework for fully autonomous operations

 

[diplomat33]

Just to be clear that is 6.1 million miles and "Waymo said that it was involved in 18 crashes and 29 near-miss collisions"
6.1 million miles and 18 actual crashes = 1 crash every 338k miles...
6.1 million miles and 47 crashes and near misses = 1 incident every 130k miles.

Also, love the clarity on actual fully driverless vehicles - maybe we will use real statistical data from now on @diplomat33
"Between January and December 2019, Waymo’s vehicles with trained safety drivers drove 6.1 million miles. In addition, from January 2019 through September 2020, its fully driverless vehicles drove 65,000 miles."

By the way, here is the link to Waymo's safety report and the direct data:

Safety – Waymo

 

[DanCar]

The newer Waymo robotaxis are electric. In fact, Krafcik says the plan is to only do EVs for all future robotaxis so that the fleet will become 100% EV.

Agree, future deployments will only be in 100% EVs.

 

[Daniel in SD]

One thing I'm a little confused about is that there were 27 disengagements that resulted in accidents in simulation. How many of those resulted in accidents in real life? Did the safety driver avoid all those accidents? It sounds like it.

In total, the Waymo vehicle was involved in 20 events involving contact with another object and experienced 27 disengagements that resulted in contact in post-disengagement simulation, for a total of 47 events (actual and simulated). In two of the actual events (which occurred after disengagement), post-disengage simulation revealed that the event would have been moderately more severe had the trained operator not disengaged. Therefore, these two events are treated in this paper according to their more severe simulated outcomes, yielding a total of 18 actual outcomes and 29 simulated outcomes.

It sounds like defensive driving is the most difficult problem they're dealing with. A crash every 130k miles doesn't sound that great even if they're always the other driver's fault. On other hand I'm not sure how you avoid people walking into the side of the car. I want to see the video of that! Very curious about the simulated collisions too. Looks like the vehicle was going to stop in front of the cyclist and skateboarder but the safety driver avoided it?

 

[Bitdepth]

By the way, here is the link to Waymo's safety report and the direct data:

Safety – Waymo

Wow. That's a treasure trove of information.

 

[masterxel]

The Waymo article and report mention that "nearly all" collisions were caused by the other driver. I can't find where they talk about which collisions that was not the case for.

https://storage.googleapis.com/sdc-...Waymo-Public-Road-Safety-Performance-Data.pdf
"Nearly all the events involved one or more road rule violations or other errors by a human driver or road user..."
"Nearly all events summarized above involved one or more road rule violations or other driving performance deviations by another road user."
"Nearly all the actual and simulated events involved one or more road rule violations or other incautious behavior by another agent..."

 

[DanCar]

Yeah the simulation in a safety report sounds very bizarre to me. I can imagine a bunch of gnarly situations that few people would avoid an accident in. Why report those?

 

[diplomat33]

Yeah the simulation in a safety report sounds very bizarre to me. I can imagine a bunch of gnarly situations that few people would avoid an accident in. Why report those?

Probably just to be as transparent as possible. Waymo is disclosing the near misses that might have caused real accidents if the safety driver had not intervened. That way we can see what might have happened if the Waymo car had been fully driverless in that situation.

 

[Daniel in SD]

Yeah the simulation in a safety report sounds very bizarre to me. I can imagine a bunch of gnarly situations that few people would avoid an accident in. Why report those?

I think it's because the safety driver did avoid the accident. Humans are very good at driving!
I think the accidents that the safety driver did not avoid are counted as real accidents the same as if the system were engaged. I assume none of the accidents that occurred after disengagement would have been avoided by the car otherwise they would have reported that too.

 

[diplomat33]

One thing I'm a little confused about is that there were 27 disengagements that resulted in accidents in simulation. How many of those resulted in accidents in real life? Did the safety driver avoid all those accidents? It sounds like it.

Cases that resulted in a real crash are reported as real crashes. The simulated crashes are all cases that the safety driver was able to avoid. So yes, the safety driver was able to avoid all 27 potential accidents.

It sounds like defensive driving is the most difficult problem they're dealing with. A crash every 130k miles doesn't sound that great even if they're always the other driver's fault. On other hand I'm not sure how you avoid people walking into the side of the car.

Yes, solving defensive driving seems to be Waymo's most difficult problem right now. They clearly have autonomous driving that can drive really well, the problem is the other bad drivers that violate the rules of the road. Waymo needs to refine their driving policy to better anticipate cases where another driver might make a mistake.

I do think that this gives us a glimpse into how we might approach safety for autonomous driving. We cannot eliminate all accidents, especially with human drivers sharing the road with autonomous vehicles. But we can make the autonomous car drive in a way that does not cause accidents and also avoids as many not at fault accidents as possible.

I trust that Waymo engineers are pouring over all this data and working out how they can make the Waymo a better defensive driver. When they do, Waymo will be one step closer to being able to deploy robotaxis in larger numbers, and in more areas.

 

[diplomat33]

@powertoold @Bladerskb This thread is about Waymo. Can we please not clutter the Waymo thread with a discussion that is off topic?

Thanks.

 

[Daniel in SD]

The Waymo article and report mention that "nearly all" collisions were caused by the other driver. I can't find where they talk about which collisions that was not the case for.

https://storage.googleapis.com/sdc-...Waymo-Public-Road-Safety-Performance-Data.pdf
"Nearly all the events involved one or more road rule violations or other errors by a human driver or road user..."
"Nearly all events summarized above involved one or more road rule violations or other driving performance deviations by another road user."
"Nearly all the actual and simulated events involved one or more road rule violations or other incautious behavior by another agent..."

It is the responsibility of drivers to avoid accidents even if they're not at fault. If it doesn't do that better than a human then it may be less safe than a human. They probably didn't include it because they don't think the exact numbers are relevant.
I'm curious how many accidents were precipitated by erratic (non-human like), though most likely strictly legal, behavior of the Waymo vehicle.
I'm also curious how the simulated accident rate compares to human drivers in the area. It's possible that even with a higher accident rate it could still be safer than a human because it avoids the most severe types of accidents.