Hacker News Reader

Comment by slg | Hacker News Reader

I wonder what happens when these cars get in an accident. I don't even mean in the legal sense. What happens in the immediate wake of an accident? Is there a note on the car about what the other driver should do? Is there a way for someone back in the office to communicate to the other driver on the scene? Does the other driver need to wait around until some human shows up?

View on HN

I think the upside of this is that if you get hit by a self-driving car, due to the way the media behaves around self-driving accidents, you can be almost sure that they will deal with it professionally and pay off all damages if it's even remotely their fault.

They might not leave a note on your windshield but they would surely acknowledge the accident and as a company, would have to cooperate with any investigations. It's not some random dude that if they hit and run off you will never see them again.

Elaine Herzberg is still dead. Not much upside in that scenario.

That's kinda unfair though. She was jaywalking in the dark far from any crosswalk on a main street. On average 16 pedestrians are killed every day in the United States. While the car could have done better, it's not really clear that she would have survived if the car was driven by a human.

Please stop spreading misinformation. It was only dark in the DVR footage provided by Uber. Actual photos show it was pretty well lit. https://www.phoenixnewtimes.com/news/uber-self-driving-tempe...

> Further, the report says that as long as Vasquez met the minimum visual acuity for a driver's license under Arizona law, and with the lighting conditions at the scene, she "could have seen and recognized that a person was walking across the road from a distance of 1700 feet away with or without the Volvo’s headlights."

Thank you for the correction.

As long as we're correcting this story, note also that "far from any crosswalk" is also completely bogus. The place where the woman was trying to cross is where a paved path leads from a beach, and then just suddenly ends at this road. The only reason it was illegal to cross the road here is because they put up a sign. In the absence of the sign, every architectural feature of this landscape would lead a reasonable person to believe they could be here.

https://www.google.com/maps/@33.4365905,-111.9426898,3a,75y,...

Arizona has quite a few of these no pedestrian crossing signs, combined with the extreme length between intersections/legal crosswalks, it is very hostile to humans.

It’s also the fourth state in pedestrian fatalities per 100k pop. Coincidence?

It's pretty abhorrent that jaywalking is even considered a factor in the eyes of the law. What if you were senile or a child or blind or mentally unwell or intoxicated or running on little sleep, and just took a stumble onto the road? Suddenly your slaughter doesn't matter because you weren't walking on some white paint at the time? I think this needs to be changed.

Yeah it's all good intentions and PR talk until you see how the lawyers respond after an incident. IMHO Uber incident shows that it is extremely unlikely for any human to face criminal charges after a deadly incident with self driving car.

Sure, but that's a problem of the legal system.

Uber was still reachable by authorities, and did show up in court.

All I am saying is that self-driving companies aren't going to pull off a hit-and-run and dodge the fact that an accident did happen. So one doesn't really need to worry about how a self-driving car deals with the note-on-windshield part. More than likely their HQ would be aware of the accident remotely and have legal responsibility to call 911 immediately if necessary. And in the event that a car's algorithm pulled off a hit-and-run on minor property damage, the company would still be easily reachable by authorities and respond to a court summons.

Wouldn't they, as a company, be incentivized to fight desperately to avoid potentially setting a costly precendent?

Absolutely. See Uber's actual response to their deadly incident -- absolving themselves of liability and blaming the backup driver.

Depends. They have incentives to not allow costly precedents. However they also have incentive to ensure that when this goes mainstream there are precedents in their favor.

If they believe the "accident" is either a one-off (meaning they will fix the bug and a whole class never happens again) they will probably pay off quick just because their image is better for that.

If they decide the accident was their fault, not avoidable with better software, AND despite that they are still significantly safer than humans they probably pay out, adjust their price models to insurance against their faults. In this case they may fight, but it will be for "reasonable" settlements (whatever they decide that is, but probably based on similar human caused accidents)

My guess is they think one of the two are the case for everything by now, so the other possibilities don't need to be listed.

Of course the above is my guess. You can be sure those who know are not talking.

> I think the upside of this is that if you get hit by a self-driving car, due to the way the media behaves around self-driving accidents, you can be almost sure that they will deal with it professionally and pay off all damages if it's even remotely their fault.

Tesla and Uber have shown that this is not the case. Tesla tries to throw the driver under the bus or, if that fails, the victim, and Uber has acted especially despicably after their fatal crash.

Perhaps "traditional" car manufacturers with less of a cult behind them will take the PR route, but the behaviour of current companies involved in self-driving vehicles isn't very encouraging.

It isn't really fair to throw Tesla, Uber, and Cruise/GM in the same bucket in terms of liability. Tesla is level 2 automation, Uber is level 3, and Cruise is at least level 4. They all will have different capabilities and different expectations for any potential human driver so the breakdown of liability between the human and the car should vary.

https://www.nhtsa.gov/technology-innovation/automated-vehicl...

Do you have a citation for Cruise being at Level 4? I’m very skeptical of that. Friends of mine at Cruise have privately told me they’re still working towards Level 4.

I think the NHTSA documentation is also insufficiently precise.

I don't have any citations beyond what is in that medium post which should be enough to indicate it is level 4. Level 3 requires a driver to be available to take immediate control of the vehicle. The simple act of sending a car out on a road without a driver shows the car is level 4 (or the company is wildly irresponsible to try it and the state is wildly irresponsible for granting the permit, but hopefully we can rule these out).

I suppose, but the parent post claimed that media attention would drive companies to do the right thing for publicity reasons. The past behaviour of car companies around any accident involving their assistive technologies doesn't really show them caring about the potential negative damage. If the did, they'd had paid out already, regardless of blame.

My point is that the media attention involving crashes won't make companies take responsibility or pay to keep the bad news away, they act like all existing car companies and shift blame onto anyone but themselves instead.

Hah, watch them argue you agreed to binding arbitration by driving on the same road as their car.

That will just happen in the first cases. Then death will normalized and rationalized on being the pedestrians fault. At least that’s how it happened when the automobile was introduced.

ex-Cruise here. Cruise has an incident response team that trains for this. For a team that hasn't had any real-world accidents yet, they seemed pretty competent. They'll show up ASAP and take care of things.

After the first few lawsuits, the company will quickly fine a way to push the liability off from themselves in fine print.

Uber already successfully did this in Arizona, where one of their self-driving cars killed a pedestrian, but the only person charged was the safety driver: https://arstechnica.com/cars/2020/09/arizona-prosecutes-uber...

That woman’s job was specifically to monitor the road in case the software did not work. That’s why she’s culpable.

Yet she was put in a situation where she had to regularly take her eyes off the road to enter data as part of her job.

Uber is culpable for designing the job such that the risk of accidents significantly increased.

Except she wasn’t doing that either. She was watching Netflix on her phone.

Uber should have just put two people in the car and taken their phones for the duration of the shift. Totally foreseeable in hindsight which is awful considering someone died.

Most companies don't hire a second supervisor to watch their drivers. I don't think that's a reasonable standard for criminal negligence.

I do think civil liability is appropriate, and uber Settled for $9.7 Million. (it looks like the family is also suing the city and state.) https://www.azcentral.com/story/news/local/tempe/2019/03/19/...

She wasn't looking at the road because she was entering data about the drive into the uber app, at uber's request.

Also, even if she wasn't doing it at uber's request, this is such a predictable event. The driver was paid close to minimum wage... Regardless of how you feel about this, it's definitely the case that people who are paid minimum wage will typically give minimum effort. This was totally forseeable. The people in charge of making this decision at uber were either negligent or willfully ignorant, but there's no excuse to hire someone for near-minimum wage for this job if you truly care about safety.

Surely she could have parked to do that.

How do you know she didn't have to record something every 30 seconds? What if Uber told her not to? Maybe she had to enter in readings from the dashboard? How about if the data needed to be entered within 10 seconds of a light flashing? Etc...

Don't let the facts get in the way of the outrage machine!

She was watching The Voice on Hulu.

We know she didn't have to record something every 30 seconds or enter reading from the dashboard or enter something within 10 seconds of a light flashing.

She wasn't "entering data about the drive into the uber app, at uber's request". She was streaming the The Voice on Hulu.

Is this some weird type of disinformation campaign against uber? I don't like uber but it's weird seeing a made up story like this. Only other place I have seen these is in some worker organizing efforts (which I generally support - but the made up stuff can just circulate like crazy).

Except nobody with a sane mind would consider this an acceptable safety system. It's widely known humans are not capable of performing this job. Period. Everywhere else, where there's a need for long stretches of uninterrupted human focus, you have systems on top of systems ensuring this happens.

In swimming pools, you have lifeguards with <30 minutes shifts, overlapping coverage zones, and managers hovering behind them to ensure they do pay attention. On trains, you have dead-man switches that stop the train if the driver isn't paying attention. The military gives pilots amphetamines if they need to stay awake for long hours. Etc.

You just cannot expect a person to stay focused monitoring something for more than couple minutes. You need to have systems to help them and mitigate the unavoidable lapse in focus. At the very least, they should've taken away her phone and made her narrate what she sees on record, while making her aware it will be reviewed and compared against video recordings afterwards.

They set her up to fail, so it's wise to ask what her role really was. Convenient scapegoat, organizational surge circuit breaker, ablative armor - it's a plausible explanation.

You just cannot expect a person to stay focused monitoring something for more than couple minutes.

It seems reasonable to expect someone to at least not be watching Hulu when they are supposed to be a safety driver. This wasn't a case of "the task was just too hard for a human".

e.g. see https://internetofbusiness.com/uber-safety-driver-watching-v...

The lidar saw Herzberg 6 seconds before impact, and there are news stories detailing email evidence that Meyhofer and others at UberATG intentionally made the perception system filter sparse lidar returns at the time (likely because they were having difficulty with trees). Today, Meyhofer and others have vested tens of millions in stock following the Softbank investment. While the safety driver was a factor in the crash, there is a gross amount of greed that was also a factor.

Has Uber improved since then? I attended an UberATG recruiting event where they showed a disengagement where they almost hit a little girl after she got off a school bus. (They were showing this because they re-created the scene on their test track and wanted to show they learned something). But in the perception video of the event they showed, you can clearly see there being lidar returns reflected off the school bus mirror, the system clustering them into an object, and the Uber car slowing down for this ghost object. The girl was saved by this false positive, probably not by the safety driver hitting the brakes. Uber does some interesting SDC research, but I would not feel safe being within 100 meters of any of their cars.

Also ex-Cruise here. Yes there have been accidents, and there are accidents every month or so-- just check the public DMV reports. A lot are minor, like other cars knocking the radars off the front of the car back when Cruise cars had radars that stuck out awkwardly. Some were worse though-- I've seen one where airbags deployed in the other car (other car was going very fast and rear-ended the AV). Cruise has not had an UberATG / Elain Herzberg -magnitude accident, though.

The cars have a handbook for safety drivers and the case is handled like almost any other accident. This system was put in place following the first accident that occurred near the SF courthouse years ago (safety drivers were hired following that crash).

What happens if there's no safety driver? Remote assistance is there, and the city is not a large place if you have to recover a wayward robot. I am not a lawyer; there will probably be some regulation eventually.

Interesting. You’d definitely have some crash detection. A remote human operator could be assigned to take over when the car stops to drive around some obstacle in an unexpected situation. If a crash happened (whether or not it was remote controlled), you could have speakers/cameras/microphones allow the operator to communicate with bystanders until help arrives. Police would need to be called for a major accident, while for a fender bender you can probably just send a car with some security people to pay off the other driver and get them to sign an NDA, and drive the crashed car back to base.

Remote human operation is only possible with a reliable cellular data connection. There are still dead zones in many cities, especially in tunnels.

For a product that’s still being tested and not fully rolled out, you could block it from operating in those spots, if you can reliably map where they are.

How do you protect against sudden dead spots or spots where cellular data is degraded? It only takes one power cut, grumpy person with a cellphone jammer, or error in the cellular modem to cause a lapse in usable connectivity.

Excellent question. And what if the other party, at fault or not, is not a driver? What if they are a pedestrian or a cyclist? What if they are alone, incapacitated and need medical help?

> What if they are alone, incapacitated and need medical help?

Although self-driving is a hard problem, automatically determining that the car may have had an accident and alerting humans in your control center is a much easier problem. I imagine the control center humans would be responsible for calling 911 if there was an injured human. They would probably be able to see out all the cameras in the car to help make that determination, or if the cameras are all malfunctioning you probably want to dispatch ambulances anyway.

This technology already exists. GM already has this technology: GM Onstar.

OnStar Corporation is a subsidiary of General Motors that provides subscription-based communications, in-vehicle security, emergency services, hands-free calling, turn-by-turn navigation, and remote diagnostics systems throughout the United States, Canada, China, Mexico, Europe, Brazil, and Argentina

What if the accident took out the cell tower and the car lost communication with HQ? Or more likely, the car was totaled and lost communication?

> What if the accident took out the cell tower and the car lost communication with HQ? Or more likely, the car was totaled and lost communication?

What do you do if you are HQ in charge of an expedition team going along a pre-determined route and the team suddenly and abnormally loses communication with HQ?

What do you do if a flight suddenly disappears over the ocean on air traffic control, and there happen to be some other planes and ships in the area?

What do you do if a hiker doesn't come home and they left a note saying which trail they were going on for the day?

> Although self-driving is a hard problem, automatically determining that the car may have had an accident and alerting humans in your control center is a much easier problem

I have a sneaky feeling that detecting whether a driverless car has hit a cyclist is not an easy problem. I can imagine scenarios where the contact took place towards the rear of the car (and therefore not at the primary focus of sensor attention), the cyclist goes careening off to the left or right and the car blithely goes on. I'm not saying such a scenario is impossible with a traditional car but it's a dam sight harder to ignore.

Yes, the cyclist interaction came to my mind too. At times in slow traffic you may encounter the same cyclist sharing the same road over and over either in front, on the side, or about to catch-up to your car. Those blind spots are truly ennerving at such times...

Do driverless cars have their kind of blind spots?

How about dealing with potholes? Do they also swerve around to protect suspension or just ram through?

These cars don't have blind spots. They setup the spinny lidars to get almost 360 degree coverage and also dedicated lidars that are for nearfield detection. It's very hard, if not impossible, to sneak up on these cars. There are also techniques to drive defensively/safely if there is uncertainty about the scene. For potholes, it's a function of detection. You can't rely on mapping data, because potholes aren't something that might exist at the time of mapping, so you have use lidar/radar to detect a hole, but depending on angle, if there is another car in front, time of lidars, etc. it might be difficult to detect. My speculation is that the cars just ram through it.

Now that would be an interesting job... working in a call center, talking to the police remotely about an accident that took place in a state you aren't even in.

I guess the same thing that happens when a pole falls down on your car? You report it to the authorities. You take pictures for the insurance, and file your claims?

How's my driving? Ring 555 5555 if I hit you.

Reply on news.ycombinator.com