Cruise autonomous vehicles outperform humans in a comparable driving environment | Hacker News Reader

Cruise autonomous vehicles outperform humans in a comparable driving environment

twitter.com

20 pointsmhb18 comments

Here’s the report: https://getcruise.com/news/blog/2023/human-ridehail-crash-ra...

Biggest question is if the miles driven by autonomous vehicles are the same likelihood of crash as the non-autonomous miles driven in the survey. Effectively, all roads aren’t equally safe, and often times autonomous vehicles (rightly) drive on safer roads in calmer climates.

It seems that we've learned over the last few years that a big part of getting a robot to operate in an unstructured environment is contextual understanding. Example - a woman fell under a Cruise car recently in SF and the car ran over her leg. The car correctly guessed that something was wrong, but it did the wrong thing in that context. It stopped on top of her leg and pinned her under the vehicle.

Can you honestly tell me you'd have chosen to reverse your car in this scenario?

95% of people would freeze, stop the car, and get out and look if they thought they hit someone, to avoid making the problem worse.

Even if they would have reversed, it would have been the wrong choice.

Oops, just ran over someone. Let me back up and run over them again!

Some percentage would panic and floor it, spinning the tires and removing a lot of flesh

I think actually it's not desirable for the car to try and drive itself off a person trapped under the wheel

I’m not one to defend Cruise generally, but I believe emergency services told the car to not move, and in this case, the woman had previously been hit by another car. The real issue is that it didn’t see her lying on the road.

The correct solution in this case was to stop. Far better to jack up the car with first responders present, not continue to drive and run over the person with the other wheel.

One is also missing the context of the pedestrian had just been hit by another vehicle, it wasn't like this person was laying in the street for a while and this car ignored and ran over her. The woman was struck by one car, was on the hood of the car for a few moments, fell off that car and immediately into the path of the Cruise car.

In the news articles I’ve read, it says the vehicle was kept in place at the request of the police. Can you explain why that was the wrong thing to do?

https://archive.ph/kXOoM

Contrast this with all the news reports lately about Cruise vehicles being involved in traffic jams, accidents, and emergency vehicles, leading them to be forced to halve their # of cars on the road.

“Comparable driving environment” indeed.

Yeah, I think Cruise's social media need to be taken with a grain of salt.

Those data points alone don't really help to reinforce an argument; In the same time period, human drivers caused accidents and reacted poorly to traffic jams / emergency vehicles in more instances, by a few orders of magnitude I'm sure. The only difference is that this is considered normal/unavoidable, and no one can order the human driver population to halve their activities.

I'm sorry, but how can you be this media illiterate in this day and age?

You're making conclusions about things based on sensationalist news which - by definition - report on uncommon events? And extrapolate wildly based on 1 or 3 isolated incidents?

"8 billion people didn't get murdered today" isn't news.

"600,000 accident free miles driven last month" isn't news.

In situations like this, only large scale data matters.

Then you should also know quality of the data matters. Driving on empty streets and counting those as miles is like eating empty calories.

Perhaps if you'd looked at the news reports you'd have a different opinion. Cruise vehicles all simultaneously stopped working near a music festival where they couldn't get a 4G connection, causing a traffic jam as the festival ended. All their incidents speak to a lack of safe engineering practices.

In any case here's another HN thread today about Cruise being suspended. For obvious reasons. https://news.ycombinator.com/item?id=38002752

This study uses Cruise “estimates” and the comparable driving environment is the one Cruise currently operates in, which is pretty easy (barren streets after nine p.m.).

Further, the raw data at https://getcruise.com/news/blog/2023/human-ridehail-crash-ra... shows a much higher crash rate for the autonomous vehicles. It was only after massaging the data that the AVs were decided to be safer.

My question is whether they outperform in terms of severe accidents not all accidents in general.

Obviously a robot will be better at paying attention than a human all the time, but those situations seem more likely to be less severe. The true measure of safety seems to be whether they are better at avoiding fatal accidents, but comparing fatal accident numbers in a Tesla vs a comparable Mercedes seems to indicate that this is not the case.

How they deal with kids, who put traffic cones around the car?

How do you deal with intentionally harmful actors trying to smear technological progress by repeating specifically dishonestly framed questions living rent free inside your clueless defenders' brains through lite propoganda viral marketing campaigns?

I guess by being honest that the problem can be tricky to solve but back up the dialog to state that every podunk online article you read is framing human inconvenience from backed up traffic as more important to immediately solve over saving more lives :)

Reply on news.ycombinator.com