Cruise

[cwerdna]

Probe into GM's Cruise finds poor leadership, culture issues at center of accident response

The report addresses controversy that has swirled around Cruise since an Oct. 2 accident in which a pedestrian was dragged 20 feet by a Cruise robotaxi.

www.cnbc.com

 

[diplomat33]

Probe into GM's Cruise finds poor leadership, culture issues at center of accident response

The report addresses controversy that has swirled around Cruise since an Oct. 2 accident in which a pedestrian was dragged 20 feet by a Cruise robotaxi.

www.cnbc.com

Here are the principle findings:

Quinn Emanuel has reached the following principal findings and conclusions, which we are sharing verbatim here:

• By the morning of October 3, Cruise leadership knew about and discussed that the Cruise AV had moved forward after the initial pedestrian impact and, in doing so, had dragged the pedestrian for approximately 20 feet. More than 100 Cruise employees – including certain members of Cruise’s senior leadership, legal, government affairs, and systems integrity teams who briefed government officials – were informed of this information prior to Cruise’s meetings on October 3 with the San Francisco Mayor’s Office, NHTSA, DMV, and other government officials. In each of those meetings, Cruise had the intent to affirmatively disclose those material facts by playing the Full Video and letting the “video speak for itself.” Because Cruise adopted that approach, it did not verbally point out these facts. This is because Cruise assumed that by playing the Full Video of the Accident for its regulators and other government officials, they would ask questions and Cruise would provide further information about the pullover maneuver and pedestrian dragging.
• The weight of the evidence establishes that Cruise played or attempted to play the Full Video depicting the pedestrian dragging in their October 3 briefings with the regulators and other government officials. However, in three of these meetings, internet connectivity issues likely precluded or hampered them from seeing the Full Video clearly and fully. And Cruise failed to augment the Full Video by affirmatively pointing out the pullover maneuver and dragging of the pedestrian.
• On October 2 and 3, Cruise leadership was fixated on correcting the inaccurate media narrative that the Cruise AV, not the Nissan, had caused the Accident. This myopic focus led Cruise to convey the information about the Nissan hit-and-run driver having caused the Accident to the media, regulators, and other government officials, but to omit other important information about the Accident. Even after obtaining the Full Video, Cruise did not correct the public narrative but continued instead to share incomplete facts and video about the Accident with the media and the public. This conduct has caused both regulators and the media to accuse Cruise of misleading them.
• The reasons for Cruise’s failings in this instance are numerous: poor leadership, mistakes in judgment, lack of coordination, an “us versus them” mentality with regulators, and a fundamental misapprehension of Cruise’s obligations of accountability and transparency to the government and the public. Cruise must take decisive steps to address these issues in order to restore trust and credibility.
• Despite the failure to discuss the pullover maneuver or pedestrian dragging with regulators, the evidence reviewed to date does not establish that Cruise leadership or employees sought to intentionally mislead or hide from regulators the details of the October 2 Accident. Instead, they attempted to show the Full Video of the Accident in good faith, but with varying degrees of success due to technical issues.
• Finally, the DMV Suspension Order is a direct result of a proverbial self-inflicted wound by certain senior Cruise leadership and employees who appear not to have fully appreciated how a regulated business should interact with its regulators. Regulators and other government officials who enforce laws and regulations designed to protect human health and safety want and need to know all relevant facts about an accident involving a regulated product. It was a fundamentally flawed approach for Cruise or any other business to take the position that a video of an accident causing serious injury provides all necessary information to regulators and otherwise relieves them of the need to affirmatively and fully inform these regulators of all relevant facts. As one Cruise employee stated in a text message to another employee about this matter, our “leaders have failed us.”

Cruise Releases Third-Party Findings Regarding October 2

Third Party Reviews October 2 Cruise Incident | Cruise

getcruise.com

 

[Tam]

"Despite the failure to discuss the pullover maneuver or pedestrian dragging with regulators, the evidence reviewed to date does not establish that Cruise leadership or employees sought to intentionally mislead or hide from regulators the details of the October 2 Accident. Instead, they attempted to show the Full Video of the Accident in good faith, but with varying degrees of success due to technical issues."

There's no misleading or hiding, so why the punishment?

"It was a fundamentally flawed approach for Cruise or any other business to take the position that a video of an accident causing serious injury provides all necessary information to regulators and otherwise relieves them of the need to affirmatively and fully inform these regulators of all relevant facts. "

So they did show the video in its entirety

So, not misleading, not hiding; it's just missing the part of actually verbalizing: Look at the video; the car dragged the victim for 20 feet!

So, not verbalizing means not misleading, not hiding?

 

[diplomat33]

There's no misleading or hiding, so why the punishment?

So they did show the video in its entirety

So, not misleading, not hiding; it's just missing the part of actually verbalizing: Look at the video; the car dragged the victim for 20 feet!

So, not verbalizing means not misleading, not hiding?

The report says that Cruise played or attempted to play the full video but internet issues prevented the video from playing clearly or fully:

"The weight of the evidence establishes that Cruise played or attempted to play the Full Video depicting the pedestrian dragging in their October 3 briefings with the regulators and other government officials. However, in three of these meetings, internet connectivity issues likely precluded or hampered them from seeing the Full Video clearly and fully. And Cruise failed to augment the Full Video by affirmatively pointing out the pullover maneuver and dragging of the pedestrian."

And even after showing the video, Cruise failed to correct a false narrative but continued to share incomplete facts. That is why regulators accused Cruise of misleading them:

"Even after obtaining the Full Video, Cruise did not correct the public narrative but continued instead to share incomplete facts and video about the Accident with the media and the public. This conduct has caused both regulators and the media to accuse Cruise of misleading them."

Cruise's approach was that sharing the video was the same as sharing all the facts and if the regulators did not ask about it, then that is on them:

"In each of those meetings, Cruise had the intent to affirmatively disclose those material facts by playing the Full Video and letting the “video speak for itself.” Because Cruise adopted that approach, it did not verbally point out these facts. This is because Cruise assumed that by playing the Full Video of the Accident for its regulators and other government officials, they would ask questions and Cruise would provide further information about the pullover maneuver and pedestrian dragging."

So Cruise's approach seemed to be that they tried to show the video, did not show the video properly due to technical issues, and when people got the wrong idea, they conveniently did not correct it and was like "hey we tried to tell the whole truth and it's on you that you did not ask about it"

 

[diplomat33]

Brad Templeton has some more details from the technical investigation on what when wrong with the Cruise AV:

In particular, it is revealed the Cruise vehicle is stopped for less than 3/4 of a second between the time it finishing braking during the impact with the victim, and it starts moving to pull over.

Here is the graph that shows timeline and event sequence with velocity, acceleration and steering wheel angle of the Cruise AV:

Cruise had error in map that did not show that the right side of the street was a bike lane so the Cruise thought it had more room to pull over then it actually had:

In particular we learn that Cruise had an error in their maps of the street. On 5th, there are two lanes heading SE in which Cruise occupied the rightmost and the hit-and-run Nissan was in the leftmost when it hit the victim. To the right there is a bike lane, divided from the car lanes by soft bollards. Apparently Cruise’s maps did not fully identify that the right side of the street was a bike lane, so it felt there was room to pull over, though in fact it was already almost as pulled over as it could get. (During the live situation it recognized the bollards and did not attempt to cross them.)

Cruise perception made error in characterizing the collision. It thought the pedestrian hit the side of the vehicle instead of the front. As a result, it did not identify the pedestrian as going under the vehicle and it did not identify the legs because the classifier was not trained on only identifying legs alone:

Unfortunately, the car made an error in characterizing the impact with the pedestrian who was flung in from the left to lie in front of the Cruise Bolt. It decided she had hit the left side of the car, when in fact she hit the front and went under the car. Based on the assumption it was a left side impact, and under the belief there was a space to the right available, the car decided to pull over and so moved forward, but found it couldn’t actually pull over due to the bollards, so it stopped.

The side-collision error is a serious one, considering the pedestrian’s legs were visible in the wide angle left side camera the entire time after the Cruise-pedestrian impact. The car simply failed to identify the legs—not something their classifier was trained on or looking for, apparently. It is not clear if the camera with the view of the legs was present in the multi-camera view showed to the DMV over Zoom.

Cruise Releases Independent Reports On Oct 2 Pedestrian Dragging Event

Cruise has released reports from independent agencies that examined its crash with a pedestrian and the alleged cover-up of dragging. Lots of details

www.forbes.com

 

[Tam]

...it did not identify the legs because the classifier was not trained on only identifying legs alone:

All of this kind of labeling is so high-tech that if a labeler labels a leg, the machine would learn that it's a leg!

On the other hand, the machine should learn that if it cannot identify a leg (because the labeler did not label it) then it should be treated as an obstacle.

Thus, there should be no reason that all obstacles must be labeled specifically as a car, deer, leg... in order to avoid the obstacles.

All unlabeled obstacles should be treated as obstacles.

 

[stopcrazypp]

There's no misleading or hiding, so why the punishment?



So they did show the video in its entirety

So, not misleading, not hiding; it's just missing the part of actually verbalizing: Look at the video; the car dragged the victim for 20 feet!

So, not verbalizing means not misleading, not hiding?

It also says they had technical issues (which is very vague) that prevented properly viewing the full video when they showed it.

The summary also glossed over the fact they cut out the portion when they showed it to the media. That seems to show a deliberate effort to hide it.

 

[Doggydogworld]

There's no misleading or hiding, so why the punishment?

Lawyers hired by GM/Cruise prepared this report. This particular firm is skilled at presenting facts while still showing their client in the best possible light and giving them "action items to ensure no similar misunderstandings happen again". The CNBC article says:

The report outlines multiple instances in which then-CEO and co-founder Kyle Vogt, who resigned in late November, made the final calls to withhold information, specifically regarding media.

So Kyle deliberately withheld the info from the media. Full stop. He didn't quite step over that line with regulators, as that could lead to criminal charges, but they did their best to "unintentionally" withhold, hide and misdirect.

So they did show the video in its entirety

Only once out of several meetings:

..attempted to show regulators a video of the incident, according to the findings, but were only able to do so in one of several initial meetings due to connection or “video transmission issues.”

Wanna bet there were zero "video transmission issues" during the part when the Nissan first hit the lady?

 

[kabin]

Mary Barra on CNBC this morning said they learned something after that infamous Cruise accident. Even though they demonstrated the technology was safer than a human driver, humans have higher expectations for technology than for each other. Knowing that they are laying out new plans for Cruise over the next few months. They have confidence in Cruise.

Of course there's some positive spin for poor judgement, management, and ethics.

 

[willow_hiller]

Even though they demonstrated the technology was safer than a human driver, humans have higher expectations for technology than for each other.

I hope this is just some PR spin. Because if that's the lesson they learned from this incident, I'm genuinely concerned.

It's not a matter of humans having higher expectations for machines than we have for ourselves. If a human driver dragged a pedestrian under their vehicle for several yards, I would expect them to face some of the liability for that pedestrian's injuries. All the additional proof of safety in the aggregate does nothing to make up for causing easily preventable harm. They don't cancel each other out.

The real expectation is that autonomous machines will be built with reasonable guardrails that fail in ways to prevent avoidable loss of life or limb. Pedestrians being dragged under vehicles is not an uncommon occurrence. Why didn't Cruise have a plan in place for that type of failure? If their plan was to not hit pedestrians in the first place, then this kind of harm was an inevitable consequence of that line of thinking.

 

[diplomat33]

Even though they demonstrated the technology was safer than a human driver, humans have higher expectations for technology than for each other.

Frankly, it feels like Cruise's standard for "safer than a human" was too low. They were basically like "Our stats show that our cars reduce accident by x% so they are safer. So let us deploy our AVs now and get off our backs about all the incidents our AVs cause." No, the general public and regulators do not accept that standard. Reducing overall accidents is important but we also expect AVs not to drag pedestrians under them or repeatedly stall in intersections or block first responders. And we do want AV companies to be transparent about safety. I don't think that is too much to ask.

 

[DanCar]

Obviously computers and humans are different. They are going to have different failure modes. Holding computers to not make any mistakes that a human would is delaying deployment that will save lives. I don't disagree with the comments above, just playing devils advocate.

 

[Tam]

Obviously computers and humans are different. They are going to have different failure modes. Holding computers to not make any mistakes that a human would is delaying deployment that will save lives. I don't disagree with the comments above, just playing devils advocate.

That kind of thinking resulted in the collapse of Uber Automous Vehicle division and now GM Cruise.

Same with Boeing MCAS fatal crashes and now the door plug incident.

Some how we are told that safety is the enemy of progress.

I say it's the opposite like in the cases of Waymo L4 and Mercedes L3.

 

[DanCar]

That kind of thinking resulted in the collapse of Uber Automous Vehicle division

I studied uber driverless and had some insider info. I had previously interviewed there. Their CEO was very reckless. He didn't care about safety as much as the dollar. Uber hired the lowest paid drivers and had many other mistakes. You need to pay for safety. I don't think that is about getting something out fast. In summary Uber's failure was being cheap.

and now GM Cruise.

I think Cruise's collapse is temporary. They will stand strong again 6 months down the road.

Same with Boeing MCAS fatal crashes and now the door plug incident.

I don't see how those are relevant. There are alternatives to Boeing, so no point on being short on safety. MCAS failure was a failure of managing risk/reward. Similar to the deep sea oil explosion.

Some how we are told that safety is the enemy of progress.

Has to be taken on a case by case basis. Not every safety failure is related to rushing something out the door to save lives.

I say it's the opposite like in the cases of Waymo L4 and Mercedes L3.

Waymo maybe more ahead if they were more willing to take risk, and thus saving more lives. They have been working on it for 15 years. At one point they had to rewrite everything because the code had zero tolerance for risk. Heard that from a Waymo employee. Merc L3 is non existant? Do you know anyone doing this?

I'm disgusted by what happened at Cruise, perhaps more so because they seem to try to hide it, rather than being open about it. And agree they are trying to rush things. It doesn't seem like it would hurt too much to slow down. There were warning signals all over the place, like stuck cars and at fault accidents.

 

[diplomat33]

General Motors said Tuesday it is cutting spending in half on troubled robotaxi unit Cruise, or by about $1 billion in 2024, but said it remains committed to the self-driving project.

Cruise burned $1.9 billion in cash during 2023, and recorded a $2.7 billion pre-tax loss, not including $500 million in restructuring costs incurred in the fourth quarter as the unit cut staff, GM said.

"We are committed to Cruise," Barra said.

GM to cut spending by $1 billion on robotaxi unit Cruise in 2024

DETROIT (Reuters) -General Motors said Tuesday it is cutting spending by about $1 billion on troubled robotaxi unit Cruise in 2024, but said it remains committed to the self-driving project. Last week, Cruise disclosed probes by the U.S. Justice Department and the Securities and Exchange...

www.yahoo.com

Barra says she is committed now but I wonder for how long. Cutting spending in half is not encouraging.

 

[Tam]

...I don't see how those are relevant. There are alternatives to Boeing, so no point on being short on safety. MCAS failure was a failure of managing risk/reward. Similar to the deep sea oil explosion...

Boeing MCAS was touted as the miracle of progress. Bigger, better, more efficient engines were placed in the wrong location, tilting the nose upward. Conventional aviation design means the airplane should stay in its neutral position. However, the compliance would delay the progress. Thus, it invented "MCAS" to override human pilots to keep the airplane in the optimal position by compensating the nose-up position with the nose-down automatic maneuver. Automatic nose-down by MCAS was fatal in 2 crashes until the fleet was grounded.

It's just like purposely designing the leaning tower of Pisa when the conventional design demands that it should be vertical and not leaning but counting on new technology to compensate for it.

Their line of reasoning is: They know it, but they have to do it because the alternatives would be the holding back of progress, and we would not have MCAS or the Leaning Tower of Pisa!

That's the same with the Titan submersible implosion. The industry has already established safety that excludes carbon fiber-reinforced plastic hull for deep diving. Still, the CEO has to sell rides to passengers to fund the progress of the carbon fiber-reinforced plastic hull because the alternatives mean no money and no progress on carbon fiber-reinforced plastic hulls.

In summary, planes can still fly if appropriately designed. Houses are still good if they are not built to be leaning. Submersibles have been diving in deep sea fine for decades without using unproven carbon fiber-reinforced plastic hulls.

Progress can be made together with safety.

 

[kabin]

I hope this is just some PR spin. Because if that's the lesson they learned from this incident, I'm genuinely concerned.

It's not a matter of humans having higher expectations for machines than we have for ourselves. If a human driver dragged a pedestrian under their vehicle for several yards, I would expect them to face some of the liability for that pedestrian's injuries. All the additional proof of safety in the aggregate does nothing to make up for causing easily preventable harm. They don't cancel each other out.

The real expectation is that autonomous machines will be built with reasonable guardrails that fail in ways to prevent avoidable loss of life or limb. Pedestrians being dragged under vehicles is not an uncommon occurrence. Why didn't Cruise have a plan in place for that type of failure? If their plan was to not hit pedestrians in the first place, then this kind of harm was an inevitable consequence of that line of thinking.

Absolutely. Hopefully I didn't misrepresent what she said.

 

[kabin]

GM to cut spending by $1 billion on robotaxi unit Cruise in 2024

DETROIT (Reuters) -General Motors said Tuesday it is cutting spending by about $1 billion on troubled robotaxi unit Cruise in 2024, but said it remains committed to the self-driving project. Last week, Cruise disclosed probes by the U.S. Justice Department and the Securities and Exchange...

www.yahoo.com

Barra says she is committed now but I wonder for how long. Cutting spending in half is not encouraging.

Sounds like the plan was to lift the hood and closely monitor Cruise. What better way than to slow it all down with reduced spending. If they let go much of the staff then there should be much less to spend.

 

[kabin]

Frankly, it feels like Cruise's standard for "safer than a human" was too low. They were basically like "Our stats show that our cars reduce accident by x% so they are safer. So let us deploy our AVs now and get off our backs about all the incidents our AVs cause." No, the general public and regulators do not accept that standard. Reducing overall accidents is important but we also expect AVs not to drag pedestrians under them or repeatedly stall in intersections or block first responders. And we do want AV companies to be transparent about safety. I don't think that is too much to ask.

She didn't specify a multiple but I agree. It needs to be substantially higher than it was.

 

[kabin]

Obviously computers and humans are different. They are going to have different failure modes. Holding computers to not make any mistakes that a human would is delaying deployment that will save lives. I don't disagree with the comments above, just playing devils advocate.

For sure. There will always be technological failures so there's needs to be redundancy. And oversight for software releases and bug resolution. These systems should also be required to pass standard ADAS driving tests upon software releases and otherwise.