Cruise

[EVNow]

Since most of these Robotaxi rides seem to be 1 person rides I would think that you would need just as many Robotaxis as private cars. These companies want to keep wait times down. Keeping wait times down means more cars. Most households have 2 cars. If you and your Spouse/partner work and have to go in different directions to work you would need 2 Robotaxis.

Difference is - after my commute, my car sits in the parking lot for 8 hours (well, when I used to commute for work - now it just sits in my garage). Robotaxi can take other rides.

Ofcourse during peak times there will be a lot of demand and companies need to figure out how to balance the wait time vs capital/operational cost. But if consumer cars can actually do robotaxi work, there will be private individuals and may be small / large companies that can fill in the void.

 

[Dracaris]

Until they figure all that stuff out, how about after hitting something they just turn on the flashers, signal home base that something unexpected has happened, and wait for help? Of course, that would probably lead to people bonking the car with a fist or giving it a shove at a red light just to make it stop and freak out, but that's still better than dragging someone or something under the car.

 

[Tam]

Until they figure all that stuff out, how about after hitting something they just turn on the flashers, signal home base that something unexpected has happened, and wait for help? Of course, that would probably lead to people bonking the car with a fist or giving it a shove at a red light just to make it stop and freak out, but that's still better than dragging someone or something under the car.

The robot might follow the California Vehicle Code 20002, which says not to impede traffic in property damage accidents:

“The driver of any vehicle involved in an accident resulting only in damage to any property, including vehicles, shall immediately stop the vehicle at the nearest location that will not impede traffic or otherwise jeopardize the safety of other motorists. Moving the vehicle in accordance with this subdivision does not affect the question of fault.”

In its blog, Cruise seems to blame California and the Fed for their robot dragging the victim:

"After a collision, Cruise AVs are designed to perform a maneuver to minimize the safety risks to the extent possible within the driving context. This is called achieving a minimal risk condition, and it’s required under California regulations and encouraged under Federal AV guidance. The specific maneuver, such as coming to an immediate stop, pulling over out of lane of travel, or pulling out of traffic after exiting an intersection, is highly dependent on the driving context as well as the Cruise AV’s driving capabilities in the moment."

 

[stopcrazypp]

The robot might follow the California Vehicle Code 20002, which says not to impede traffic in property damage accidents:

“The driver of any vehicle involved in an accident resulting only in damage to any property, including vehicles, shall immediately stop the vehicle at the nearest location that will not impede traffic or otherwise jeopardize the safety of other motorists. Moving the vehicle in accordance with this subdivision does not affect the question of fault.”

Except this was not a property damage accident, it involved hitting a pedestrian.

In its blog, Cruise seems to blame California and the Fed for their robot dragging the victim:

"After a collision, Cruise AVs are designed to perform a maneuver to minimize the safety risks to the extent possible within the driving context. This is called achieving a minimal risk condition, and it’s required under California regulations and encouraged under Federal AV guidance. The specific maneuver, such as coming to an immediate stop, pulling over out of lane of travel, or pulling out of traffic after exiting an intersection, is highly dependent on the driving context as well as the Cruise AV’s driving capabilities in the moment."

As mentioned up thread, it's misleading given the "minimal risk condition" Cruise typically chooses is simply to stop in its own lane (to the annoyance of other drivers). For some reason it decided to pull over in this case though.

 

[Tam]

Austin Cruise complaints:

.bumped into parked fire trucks

.Ignored police directing traffic

.almost cutting off an ambulance flashing its lights

.struggled to be moved out of traffic by Cruise employees on site.

.blocked traffic

.almost hit pedestrians per 311 calls.

.refused to let firefighters in to move the car per Cruise protocol because the remote Cruise employee cannot be sure a uniformed firefighter a real one or from Halloween.

Records show first responders in Austin struggling to respond to Cruise's self-driving cars

As Cruise pauses driverless vehicle operations, firefighters catch a break from robot car callouts.

www.kut.org

 

[diplomat33]

Austin Cruise complaints:

.bumped into parked fire trucks

.Ignored police directing traffic

.almost cutting off an ambulance flashing its lights

.struggled to be moved out of traffic by Cruise employees on site.

.blocked traffic

.almost hit pedestrians per 311 calls.

.refused to let firefighters in to move the car per Cruise protocol because the remote Cruise employee cannot be sure a uniformed firefighter a real one or from Halloween.

Records show first responders in Austin struggling to respond to Cruise's self-driving cars

As Cruise pauses driverless vehicle operations, firefighters catch a break from robot car callouts.

www.kut.org

I wonder when these complaints happened. Most of the complaints seem to deal with first responders. On Oct 12, Cruise tweeted that they had pushed a software update that improves response with first responders.

https://twitter.com/x/status/1712499933281173991

If the complaints happened before the update then hopefully the update worked to fix the issues but if the complaints happened after the update then it would seem Cruise still has work to do. In any case, the pause in driverless might be a blessing in disguise as it will give Cruise time to assess and fix these issues.

 

[Tam]

...In any case, the pause in driverless might be a blessing in disguise as it will give Cruise time to assess and fix these issues.

It is still testing its cars with a backup driver, though that is less of a monetary incentive.

The problem with backup drivers is that they may be too good at preventing issues caused by the robot, it's perceived as having no problems.

That meant the backup driver could have shut down the car that trapped the pedestrian under it. There would be no dragging, no cover-ups, no suspension, and no one would know there's a need to program the vehicle not to drag the pedestrian in the future.

 

[EVNow]

That meant the backup driver could have shut down the car that trapped the pedestrian under it. There would be no dragging, no cover-ups, no suspension, and no one would know there's a need to program the vehicle not to drag the pedestrian in the future.

No. Cruise should analyze that accident - with backup driver and realize they don't have a way to deal with VRU trapped below. All bad things don't need to happen before being addressed.

 

[EVNow]

I wonder when these complaints happened. Most of the complaints seem to deal with first responders. On Oct 12, Cruise tweeted that they had pushed a software update that improves response with first responders.

https://twitter.com/x/status/1712499933281173991

If the complaints happened before the update then hopefully the update worked to fix the issues but if the complaints happened after the update then it would seem Cruise still has work to do. In any case, the pause in driverless might be a blessing in disguise as it will give Cruise time to assess and fix these issues.

What are the chances Cruise had small bugs and exactly those bugs got exposed in those scenarios ?

The more likely scenario is that Cruise has a LOT of such issues and they took care of the issues that were pointed out (hopefully).

 

[HVM]

"Half of Cruise’s 400 cars were in San Francisco when the driverless operations were stopped. Those vehicles were supported by a vast operations staff, with 1.5 workers per vehicle. The workers intervened to assist the company’s vehicles every 2.5 to 5 miles, according to two people familiar with is operations."

https://www.nytimes.com/2023/11/03/technology/cruise-general-motors-self-driving-cars.html

So there go the self driving part, what are the Waymo's numbers?

 

[stopcrazypp]

"Half of Cruise’s 400 cars were in San Francisco when the driverless operations were stopped. Those vehicles were supported by a vast operations staff, with 1.5 workers per vehicle. The workers intervened to assist the company’s vehicles every 2.5 to 5 miles, according to two people familiar with is operations."

https://www.nytimes.com/2023/11/03/technology/cruise-general-motors-self-driving-cars.html

So there go the self driving part, what are the Waymo's numbers?

The only official numbers they have to report to the state are disengagement rates, which Cruise reported as 95,900 miles per disengagement in 2022. Waymo reported 17,060 miles per disengagement in 2022.

https://www.eetimes.com/waymo-cruise-dominate-av-testing/

It's hard to know what would be considered a comparable stat without knowing what "intervene" means for those 2.5 to 5 miles.

 

[kabin]

CNBC did a story on Cruise a week or so before driverless operations were forced to shut down. The short CNBC test drive wasn't good - 2 near miss accidents, oodles of honking, and one episode of road rage.

Apparently Cruise drives like a 15 year old teenage permit owner and Waymo drives like a smoth, skilled human (no near misses, honks, road rage).

A few take aways - there need to be far more standards for the driverless industry, NHTSA is asleep at the wheel, and for the most part these companies aren't transparent or honest with safety data.

 

[diplomat33]

It is still testing its cars with a backup driver, though that is less of a monetary incentive.

The problem with backup drivers is that they may be too good at preventing issues caused by the robot, it's perceived as having no problems.

That meant the backup driver could have shut down the car that trapped the pedestrian under it. There would be no dragging, no cover-ups, no suspension, and no one would know there's a need to program the vehicle not to drag the pedestrian in the future.

If so, that is a bad flaw in Cruise's testing procedure. That would be like Tesla removing driver supervision on the assumption that the system is safe enough because FSD beta has not had any accidents since the driver prevented them from happening.

 

[diplomat33]

"Half of Cruise’s 400 cars were in San Francisco when the driverless operations were stopped. Those vehicles were supported by a vast operations staff, with 1.5 workers per vehicle. The workers intervened to assist the company’s vehicles every 2.5 to 5 miles, according to two people familiar with is operations."

https://www.nytimes.com/2023/11/03/technology/cruise-general-motors-self-driving-cars.html

So there go the self driving part, what are the Waymo's numbers?

I guess we need to know what "intervene" means. If remote assistance provides help to the car without disengaging the self-driving, then it is not counted as a disengagement and it would still be self-driving. And Cruise remote assistance do not remote control the car. So it is still self-driving since the car is driving the whole time. But if they are counting times when a worker had to physically come and drive the car manually, then that would not be self-driving.

Having said that, any human "intervention" every 2.5 to 5 miles is terrible and shows why Cruise was not ready for driverless. It seems that Cruise looked at safety disengagements and since they were very good, they decided to deploy driverless, assuming remote assistance could take care of non-safety issues. But then we saw that the cars were having all kinds of stalls since there were many issues that they could not handle without human remote help. Cruise should have waited until their system was more reliable before going driverless. Now, they are paying the price.

 

[powertoold]

There's no good ending for Cruise. Their tech sucks and they seriously hurt a pedestrian and apparently lied about it.

It just goes back to what I and others have been saying for a while. It's difficult to assess the actual performance of these companies because they're trying to please their bosses and depend on marketing and deception to create hype and rationalize their crazy cash burn.

 

[EVNow]

Having said that, any human "intervention" every 2.5 to 5 miles is terrible and shows why Cruise was not ready for driverless. It seems that Cruise looked at safety disengagements and since they were very good, they decided to deploy driverless, assuming remote assistance could take care of non-safety issues. But then we saw that the cars were having all kinds of stalls since there were many issues that they could not handle without human remote help. Cruise should have waited until their system was more reliable before going driverless. Now, they are paying the price.

My take is Cruise was in competition with Waymo and tried to do driverless before they were ready. Basically they fraudulently manipulated the disengagement rates and lied about how safe they really were. These intervention rates are no better (or worse) than Tesla. 100 times worse than they need to be for driverless operations.

Would be very interesting to see what Waymo’s rates are.

 

[Doggydogworld]

"Half of Cruise’s 400 cars were in San Francisco when the driverless operations were stopped.

I believe "half" comes from the 50% reduction ordered after the fire truck accident. Before that they had close to 400 in SF plus maybe a couple dozen in Austin and a few more scattered among all the other cities they "service".

Those vehicles were supported by a vast operations staff, with 1.5 workers per vehicle.

Does this mean 300 total support workers for 200 vehicles, including maintenance, cleaning, remote monitoring, customer support, fleet support, etc.? For a 24x7 operation that'd be 65-70 on duty at any given time. More during peak times, fewer at 3am. That's too high, but far less than a fleet of 200 taxis which needs 200 drivers plus another dozen or so back office folk to dispatch cars and otherwise keep things running. So you have a labor advantage on day one and you continually reduce the ratio of remote monitors to cars over time as your software improves.

300 people on duty at all times would be a completely different story, though, and an obvious loser.

The workers intervened to assist the company’s vehicles every 2.5 to 5 miles, according to two people familiar with is operations."

A couple years ago during a "riding interview" Kyle Vogt said something like "a 15 second remote monitor interaction every 5 minutes cuts labor cost 95%". Rationale being 15 seconds is 5% of 5 minutes. Every 5 minutes seemed way too often to me, roughly once a mile in SF traffic. It showed their mindset, though -- frequent remote monitor interactions are no problem as long as they're short. Waymo had the opposite approach -- remote interactions should only happen in rare circumstances.

I think Cruise and Waymo currently run with one monitor for every 2-3 cars. Maybe Waymo is a little better now. Neither company releases data, so we can only guess.

BTW, "disengagements" are a different thing entirely. That term only applies to testing with safety drivers. Companies can pretty much define disengagements however they want and thus get whatever result they want. Some only report "safety-related disengagements", also undefined but obviously a subset of all disengagements. Apple famously had very strict criteria and reported a disengagement every few miles back when Waymo was already at 20k miles or so.

 

[Tam]

If so, that is a bad flaw in Cruise's testing procedure. That would be like Tesla removing driver supervision on the assumption that the system is safe enough because FSD beta has not had any accidents since the driver prevented them from happening.

I already know the ratio for an in-car backup driver is 1:1 but I didn't know it has 1.5 remote-controlled backup driver per driverless car.

Was there anyone doing anything remotely in real time in the dragging incident?

The correct action should be to remotely shutdown the car after hitting a pedestrian but perhaps the 1.5 remote backup drivers were trained to let the robot finish its destructive dragging behavior first?

 

[Tam]

My take is Cruise was in competition with Waymo and tried to do driverless before they were ready. Basically they fraudulently manipulated the disengagement rates and lied about how safe they really were. These intervention rates are no better (or worse) than Tesla. 100 times worse than they need to be for driverless operations.

Would be very interesting to see what Waymo’s rates are.

Disengage rates are only for Autonomous Vehicles with a backup driver.

Once the backup driver is no longer there, driverless Autonomous Vehicles no longer report the data. The public has to rely on informal reports from social media. I hope California changes this exemption privilege especially when there's no backup driver in the cars.

 

[EVNow]

Disengage rates are only for Autonomous Vehicles with a backup driver.

Once the backup driver is no longer there, driverless Autonomous Vehicles no longer report the data. The public has to rely on informal reports from social media. I hope California changes this exemption privilege especially when there's no backup driver in the cars.

They had to show low disengagement rates with backup driver to get driverless permit.