Waymo

[masterxel]

I'm curious how many accidents were precipitated by erratic (non-human like), though most likely strictly legal, behavior of the Waymo vehicle.

Had this same thought as well especially when they describe the rear ends during a left hand turn.

"The remaining rear end struck collision involved a deceleration to a near stop by the Waymo Driver while making a left turn in an intersection with a following vehicle that was traveling at a speed and following distance that did not allow for the following driver to successfully respond to the Waymo Driver’s braking. The simulated collision impact was estimated to be 16 mph, and this event is categorized as S1 severity."

 

[ThomasD]

That many miles in one city just to get the car to self drive in that city. Only a limited number of people are allowed to use it. How long before Waymo can transport people throughout the entire state? What about crossing into other states not using an interstate to travel on?

 

[diplomat33]

That many miles in one city just to get the car to self drive in that city. Only a limited number of people are allowed to use it. How long before Waymo can transport people throughout the entire state? What about crossing into other states not using an interstate to travel on?

Waymo is not working to do self-driving in just that one city. That is just the city where Waymo has launched a ride-hailing service for the public. But Waymo is testing autonomous driving in 25 metro areas, in 10 States now. And Waymo has done over 20M autonomous miles so far across the US. So Waymo is testing autonomous driving in a lot of areas in order to achieve autonomous driving that can work everywhere.

Here is the updated map of where Waymo is testing autonomous driving. As you can see Waymo already has autonomous driving that works in many areas across the US.

 

[diplomat33]

Here is the table from Waymo's paper that summarizes the crash data:

 

[Daniel in SD]

That many miles in one city just to get the car to self drive in that city. Only a limited number of people are allowed to use it. How long before Waymo can transport people throughout the entire state? What about crossing into other states not using an interstate to travel on?

6 million miles is really not that many miles. It seems like it's barely enough to evaluate the safety of the system. I'm sure they'll expand to other cities but I'm sure it will remain a local taxi service not intercity transport.
One thing that doesn't add up is that they claim that 99.9% of disengagement counterfactual simulations do not show an accident. But they had 29 simulated accidents which would imply 29,000 disengagements which would be a rate of 1 every 200 miles. That seems awfully high! And very inconsistent with their reporting in California.

 

[DanCar]

6 million miles is really not that many miles. It seems like it's barely enough to evaluate the safety of the system. I'm sure they'll expand to other cities but I'm sure it will remain a local taxi service not intercity transport.
One thing that doesn't add up is that they claim that 99.9% of disengagement counterfactual simulations do not show an accident. But they had 29 simulated accidents which would imply 29,000 disengagements which would be a rate of 1 every 200 miles. That seems awfully high! And very inconsistent with their reporting in California.

They have a sophisticated simulation environment. Did they say simulations are with safety drivers?

 

[Daniel in SD]

They have a sophisticated simulation environment. Did they say simulations are with safety drivers?

Simulations in this paper are what would have happened had the safety driver not disengaged the system. In 27 of the disengagements there would have been a collision and in 2 there would be a more severe collision than there actually was.
What would a simulation with a safety driver mean?

 

[diplomat33]

I thought the conclusion section of Waymo's paper had two interesting observations:

"The frequency of challenging events that were induced by incautious behaviors of other drivers serves as a clear reminder of the challenges in collision avoidance so long as AVs share roadways with human drivers. Statistics regarding the high percentage of human collisions that are attributed to human error may lead to inflated expectations of the potential immediate safety benefits of AVs. AVs will share roads with human drivers for the foreseeable future, and significant numbers of collisions due to human driver errors that are simply unavoidable should be expected during this period."

"Due to the typology of those collisions initiated by other actors as well as the Waymo Driver’s proficiency in avoiding certain collision modes, the data presented shows a significant shift in the relative distributions of collision types as compared to national crash statistics for human drivers. For example, the Waymo Driver experienced zero actual or simulated collision-relevant contacts in the NHTSA “road departure, fixed object, rollover” single-vehicle collision typology (27% of all US roadway fatalities). In an additional example, while rear-end collisions are one of the most common collision modes for human drivers, the Waymo Driver only recorded a single front-to-rear striking collision (simulated) and that event involved an agent cutting in and immediately braking (consistent with antagonistic motive)."

Source: https://storage.googleapis.com/sdc-...Waymo-Public-Road-Safety-Performance-Data.pdf

 

[Daniel in SD]

AVs will share roads with human drivers for the foreseeable future, and significant numbers of collisions due to human driver errors that are simply unavoidable should be expected during this period.

Except that the human safety driver avoided the majority of collisions!
I do suspect that overall it is safer than the average human because of the type of accidents it gets into but I'm not convinced it gets into fewer accidents. It will be interesting to see how liability will work when people argue, correctly, that a human driver would have been able to avoid an accident where the plaintiff is at fault.

 

[Doggydogworld]

One thing that doesn't add up is that they claim that 99.9% of disengagement counterfactual simulations do not show an accident. But they had 29 simulated accidents which would imply 29,000 disengagements which would be a rate of 1 every 200 miles. That seems awfully high! And very inconsistent with their reporting in California.

Waymo One riders used to say most disengagements were for convenience, e.g. the van was being too cautious or wasn't going to drop them off where they wanted. Waymo couldn't annoy customers too much or they'd stop using the service and providing valuable feedback. I think they also intervened at times to avoid overly annoying other drivers. This footnote from page 2 explains the difference vs. California reporting:

1In this paper, “disengagement” refers to any event in which the AV’s automated driving mode is disengaged. This is broader than the definition used in certain California state regulations, where the term more narrowly refers to certain safety-relevant disengagements.

Except that the human safety driver avoided the majority of collisions!

Trained safety driver who is tested for alertness before driving, monitored continually and given "in car tasks" to maintain alertness.

I'm surprised disengagements didn't cause any accidents. I recall two such incidents a couple years ago in the CA reports. One involved a motorcycle behind the van attempting some kind of passing maneuver. The van altered course, but the safety driver who was unaware of the motorcycle disengaged and the motorcycle rear-ended the van. Simulation showed no collision had the safety driver not disengaged. The other was a safety driver who unintentionally disengaged while sleeping (!!) and ended up in the median.

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.

Brad Templeton claims US national average is 40-60 collisions in 6.1m (the article has other good info, too, if you ignore the editorializing).

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?
View attachment 603819

I bet the pedestrian was texting while walking. The bicycle was probably braking, the 2.2 mph simulated impact speed is very slow. Simply moving the van's path a bit to the left as it stopped could have avoided that one. The skateboarder is trickier, maybe the safety driver kept the van rolling instead of letting it stop in the guy's path?

 

[Daniel in SD]

Brad Templeton claims US national average is 40-60 collisions in 6.1m (the article has other good info, too, if you ignore the editorializing).

Good editorial and I agree that the data looks good enough that they should deploy. My only quibble would be the claim of "superhuman" safety. To me superhuman means beyond human capability and Waymo isn't even close to that, I'm sure there many human drivers far superior to Waymo. They're probably better than average human safety.

 

[diplomat33]

Brad Templeton claims US national average is 40-60 collisions in 6.1m (the article has other good info, too, if you ignore the editorializing).

Thanks for sharing that article. It was a great read.

Good editorial and I agree that the data looks good enough that they should deploy. My only quibble would be the claim of "superhuman" safety. To me superhuman means beyond human capability and Waymo isn't even close to that, I'm sure there many human drivers far superior to Waymo. They're probably better than average human safety.

I agree that Waymo should deploy now in more areas and scale up more aggressively. It is the best way to get more data, needed to improve the Waymo Driver in more complex driving situations. I agree with you that the Waymo Driver is not superhuman but it does seem to have a safety rating that is "good enough" to deploy in wider areas. Of course, Waymo does have more work to do. As I mentioned before, I think Waymo needs to improve their defensive driving skills to try to avoid these accidents where the other driver violated the road rules. But overall, I am quite pleased with Waymo's safety data.

I thought this quote was interesting:

"It’s a quirk of humanity that “people don’t like being killed by robots.” We would rather be killed by drunks. We expect perfection from machines that they can’t deliver, and which we don’t expect from other human drivers."

It is indeed bizarre that we seem "ok" with bad drivers. We even tolerate 40,000 deaths per year from car accidents. Yet, there are articles criticizing autonomous cars as not ready if they get into just one fender bender or minor rear end collision.

 

[Doggydogworld]

Thanks for sharing that article. It was a great read.



I agree that Waymo should deploy now in more areas and scale up more aggressively. It is the best way to get more data, needed to improve the Waymo Driver in more complex driving situations. I agree with you that the Waymo Driver is not superhuman but it does seem to have a safety rating that is "good enough" to deploy in wider areas. Of course, Waymo does have more work to do. As I mentioned before, I think Waymo needs to improve their defensive driving skills to try to avoid these accidents where the other driver violated the road rules. But overall, I am quite pleased with Waymo's safety data.

I thought this quote was interesting:

"It’s a quirk of humanity that “people don’t like being killed by robots.” We would rather be killed by drunks. We expect perfection from machines that they can’t deliver, and which we don’t expect from other human drivers."

It is indeed bizarre that we seem "ok" with bad drivers. We even tolerate 40,000 deaths per year from car accidents. Yet, there are articles criticizing autonomous cars as not ready if they get into just one fender bender or minor rear end collision.

I've been made at Waymo's timid approach for two years now. Of course I'm not the one with billions at risk. They should be safer today than in 2019 when almost all those 47 incidents happened, though. They've also (mostly) addressed a lot of issues with rider experience, e.g. smoothness, hyper-caution and pickup points.

I still think business model is their bigger issue. They want to replace car ownership on a large scale. That's where the big money is. Otherwise it's mostly airport taxi and designated driver replacement, not giant markets. It's hard to convince most in the US to give up their cars, though.

 

[Daniel in SD]

I've been made at Waymo's timid approach for two years now. Of course I'm not the one with billions at risk. They should be safer today than in 2019 when almost all those 47 incidents happened, though. They've also (mostly) addressed a lot of issues with rider experience, e.g. smoothness, hyper-caution and pickup points.

I still think business model is their bigger issue. They want to replace car ownership on a large scale. That's where the big money is. Otherwise it's mostly airport taxi and designated driver replacement, not giant markets. It's hard to convince most in the US to give up their cars, though.

Uber's gross bookings were $65 billion in 2019 so it's not a small market. The system is quite expensive and also likely has a short service life (they're not going to want to maintain decade old vehicles) so they need to focus on high utilization use cases. It seems like trucking could be huge business for them.
Don't get me wrong, I don't think I'd invest at their current valuation!

 

[diplomat33]

I've been made at Waymo's timid approach for two years now. Of course I'm not the one with billions at risk. They should be safer today than in 2019 when almost all those 47 incidents happened, though. They've also (mostly) addressed a lot of issues with rider experience, e.g. smoothness, hyper-caution and pickup points.

Yeah, I get frustrated with Waymo's slow pace too. But I do understand it. I think it is the right approach. I am just impatient for reliable driverless autonomous driving to become widely available.

Hopefully, the release of the accident report is a good sign that Waymo feels more confident in their safety record and therefore is preparing to accelerate deployment of their autonomous driving.

 

[diplomat33]

 

[ThomasD]

Can Waymo or Tesla recognize a cop directing traffic

 

[ThomasD]

Does Waymo and Tesla FSD know the rules of a stopped school bus

 

[diplomat33]

Can Waymo or Tesla recognize a cop directing traffic

Waymo: yes.

Tesla: no.

Does Waymo and Tesla FSD know the rules of a stopped school bus

Waymo: yes

Tesla: no.

 

[Doggydogworld]

Waymo: yes.

Oh yeah? How about this guy?