Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Autonomous Car Progress

This site may earn commission on affiliate links.
How do you interpret what the Cruise AV did in the video?

It looks to me like it entered the intersection when it should have known that there was no way it would make it through. (i.e. it shouldn't have entered the intersection even if it was green.) Then as soon as it was clear it completed passing through even though it was red.

But the clip doesn't really start early enough to know exactly what happened.
 
How do you interpret what the Cruise AV did in the video?


It's a hard perspective to evaluate, but you can see at 26 seconds the pedestrian in the crosswalk walks behind the Cruise vehicle, indicating it was past the crosswalk and somewhat in the intersection.

Best course of action would have been to back up, if there was space. Second best would be to stay put. Under no circumstances does it make sense to proceed while cross-traffic is flowing.
 
It looks to me like it entered the intersection when it should have known that there was no way it would make it through. (i.e. it shouldn't have entered the intersection even if it was green.) Then as soon as it was clear it completed passing through even though it was red.

But the clip doesn't really start early enough to know exactly what happened.
It is mostly blocked by other cars in the video, but I think it went into the intersection because there was a line of cars ahead that didn't wait either. The car immediately ahead of it was able to clear the intersection after it turned red.

That said, as another mentioned, it definitely is not a safe thing to do. If it moved prior to the cross traffic moving (as did the car before it), it was legally in the intersection with the right of way and there is nothing wrong.

After the cross traffic started streaming, however, it no longer had the right of way, although technically it wasn't running the red light (as it was already in the intersection prior to light turning red).
 
It is mostly blocked by other cars in the video, but I think it went into the intersection because there was a line of cars ahead that didn't wait either. The car immediately ahead of it was able to clear the intersection after it turned red.

That said, as another mentioned, it definitely is not a safe thing to do. If it moved prior to the cross traffic moving (as did the car before it), it was legally in the intersection with the right of way and there is nothing wrong.

After the cross traffic started streaming, however, it no longer had the right of way, although technically it wasn't running the red light (as it was already in the intersection prior to light turning red).
You are correct. But entering an intersection that is blocked by a stopped car is very poor because you can get stuck there, as happened to the Cruise. Then, the Cruise compounds things by darting in front of a bus moving at high speed.

I wonder if either of these unsafe practices gets reported?
 
Mobileye Keynote by Prof. Shai Shalev-Shwartz at AEK´23:

Some target KPI:s regarding reliability:
Screenshot 2023-07-12 at 11.23.34.png
 
  • Love
Reactions: diplomat33
Mobileye Keynote by Prof. Shai Shalev-Shwartz at AEK´23:

Some target KPI:s regarding reliability:
View attachment 955520

Thank you so much!! I was impatiently waiting for this keynote to be available. Watching it now. There seems to be lots of good info. He is doing a great job of explaining the Mobileye thinking and approach. I really appreciate that Mobileye details their MTBF goals, why they picked those MTBF, differences between system failures and accidents, how failures can lead to accidents etc... You can tell they have given their approach a lot of thought and are willing to share it with the public. I like that.
 
  • Like
Reactions: spacecoin
Great data, but I feel there will be huge difference in sympathy and jury awards based on "death by human error" or "death by computer or programming/sensor error".

Yes, that could be. But that is why manufacturers need to be able to defend their system's safety and reliability. I think that is why Mobileye lays out their approach like they do: show you did your due diligence to eliminate all "reproducible errors", you built a redundant system to minimize perception errors, built a driving policy that minimizes planning errors, you did proper internal and external validation, you can prove your system has a MTBF of 10M hours which is statistically safer than humans. If you do all those things, you can make a good case for why your system is "safe enough" and you did everything reasonable to minimize failures. No AV will be 100% safe but hopefully, the jury will agree that you took reasonable steps to make your system as safe as possible.

This is why I keep bringing up liability. Who is willing to take liability when the driver is not required to pay 100% attention.

It could be complicated and it could depend on the specifics of the accident. The developer of the FSD software could be liable if the accident was caused by a specific failure of the software that they wrote. The carmaker could be liable if they were found to be negligent somehow in assembling the FSD hardware or software that contributed to the accident.

But this is why manufacturers are being very cautious with deploying "self-driving" systems. They want to make sure that they got liability covered in case of an accident. For example, we see Mercedes launch their L3 system where the human does not need to supervise under certain conditions but they severely restricted the system (it can only be used on certain mapped highways, with a lead car, good weather, only at speeds at 40 mph or less). The restrictions are designed to limit risk.

Another solution to liability is the insurance model, similar to car insurance now. The idea is that companies involved in self-driving would buy special insurance specifically to cover liability in case the AV has an accident. I believe Waymo is doing this. When the AV has an accident, the insurance would pay. And AV companies would have a built-in incentive to make their AVs safer since it would make their insurance cheaper. I think the insurance model will likely win since AV companies will not want to pay out of pocket for AV accidents.
 
Last edited:
His agency learns about incidents via contacts from local law enforcement, fire department officials or from 911 calls. An average of three 911 calls per day are made because of AV-related problems, he told Automotive News.
"There appears to be a strong correlation between the huge uptick in 911 calls that we're receiving and Cruise's transition into fully driverless and passenger service," he said, noting he's requested data from AV companies to help the city better assess their impact on city streets. Those requests have gone unfulfilled.

Source: Self-driving setbacks: AV companies bruised in California showdowns

A few thoughts:

1) I think the average of three 911 calls per day relating to AV problems should be put in some context. We don't know how many of those 911 calls were for actually serious issues or just people complaining about a minor issue. Some of the calls could be prank calls too.

2) The uptick in 911 calls that seems to correlate with when Cruise when driverless seems to support the idea that Cruise was premature in going driverless.

3) I also think Waymo has been unfairly treated because they are getting lumped into this and getting their permit delayed even though it seems that Cruise is to blame for most of the issues. This is not to say that Waymo is blameless of course. But I don't think it is fair to just lump both together and treat Waymo like they are super unreliable when most of the "stalls" are Cruise's fault. I think both companies should be looked at separately.
 
Source: Self-driving setbacks: AV companies bruised in California showdowns

A few thoughts:

1) I think the average of three 911 calls per day relating to AV problems should be put in some context. We don't know how many of those 911 calls were for actually serious issues or just people complaining about a minor issue. Some of the calls could be prank calls too.

2) The uptick in 911 calls that seems to correlate with when Cruise when driverless seems to support the idea that Cruise was premature in going driverless.

3) I also think Waymo has been unfairly treated because they are getting lumped into this and getting their permit delayed even though it seems that Cruise is to blame for most of the issues. This is not to say that Waymo is blameless of course. But I don't think it is fair to just lump both together and treat Waymo like they are super unreliable when most of the "stalls" are Cruise's fault. I think both companies should be looked at separately.
Yes both should be looked at separately, which is why the complaints are in separate letters (not all lumped into one).

Yes, as per statistics to May that SF wrote in their protest letters, Cruise makes up a huge bulk of the reported complaints (they went from about 20 per month last year to 60 per month in recent months), but the trend that worried SF about Waymo was that reported incidents from Waymo went from almost non-existent (single digits cumulative) to averaging about 30 a month when they switched to driverless.

That shows both platforms are not necessarily fully ready for a further driverless expansion and that having a safety driver still made a big difference in reported incidents. As for the severity of the incidents, I don't think necessarily it has to be serious accidents (which is mostly what previous metrics focused on). Just the cars halting in traffic still has an impact that the city hopes to reduce to zero.

Of course on this point, SF is requesting more data to see how the incidents break down and as you linked yourself previously, both companies have previously denied such requests, saying it was an overreach. I suspect CPUC will now change their tune and require more data to evaluate their impact.
 
Yes both should be looked at separately, which is why the complaints are in separate letters (not all lumped into one).

Yes, as per statistics to May that SF wrote in their protest letters, Cruise makes up a huge bulk of the reported complaints (they went from about 20 per month last year to 60 per month in recent months), but the trend that worried SF about Waymo was that reported incidents from Waymo went from almost non-existent (single digits cumulative) to averaging about 30 a month when they switched to driverless.

That shows both platforms are not necessarily fully ready for a further driverless expansion and that having a safety driver still made a big difference in reported incidents. As for the severity of the incidents, I don't think necessarily it has to be serious accidents (which is mostly what previous metrics focused on). Just the cars halting in traffic still has an impact that the city hopes to reduce to zero.

Of course on this point, SF is requesting more data to see how the incidents break down and as you linked yourself previously, both companies have previously denied such requests, saying it was an overreach. I suspect CPUC will now change their tune and require more data to evaluate their impact.

Hopefully, the CPUC will get more data. I do think the severity of the incidents matter. There is a big difference between "AV stalled for 30 minutes in the middle of an intersection that blocked traffic and required a tech to come and move the vehicle" versus "AV stalled for 2 minutes on an empty road and then moved on its own". And the fact is that there is some NIMBYism going on too, like some lady complaining about Cruise AVs driving past her house at night and she does not like the noise. So I think we need to distinguish between the legit incidents that do need to be solved in order to go fully driverless and the incidents that are less serious.
 
Last edited:
  • Like
Reactions: Doggydogworld
It's supposed to prod drivers into paying more attention in work zones, I guess by confusing them. A state first: Lane striping on I-5 in Fife turns orange this summer to help drivers notice the work zone | WSDOT
I think the problem is people know it's a work zone, but they don't slow down regardless. Maybe this can psychologically play a factor, but if its only function is to bring attention, then I don't think it would be that effective.

Probably something like rumble strips would work better.

Also a digital sign that shows clearly the speed of the people travelling and whether active construction is happening also would help a lot. A lot of people ignore construction signs because too many are posted even when there is no construction going on. If there is a clearer indicator of it being in effect it would help.
 
I think the problem is people know it's a work zone, but they don't slow down regardless. Maybe this can psychologically play a factor, but if its only function is to bring attention, then I don't think it would be that effective.

Probably something like rumble strips would work better.

Also a digital sign that shows clearly the speed of the people travelling and whether active construction is happening also would help a lot. A lot of people ignore construction signs because too many are posted even when there is no construction going on. If there is a clearer indicator of it being in effect it would help.
The only way is with police enforcement.