Hacker News Reader

Comment by chrsstrm | Hacker News Reader

The description given before the video was released painted a picture in my mind that the woman was on the median and "suddenly" entered the roadway in front of the vehicle. I pictured someone darting across the road directly in front of the car, with no way to stop in time.

This video shows a completely different scenario. The woman started on the median, but the vehicle was in the #2 lane. She wasn't visible to the naked eye but she also wasn't darting into traffic and had to cross the #1 lane before even being in the path of the vehicle. A human driver certainly would have difficulty stopping in time, but why did the sensor package not pick her up? This doesn't appear to be the close call we were told it was. To me, this seems like exactly the scenario that autonomous driving vehicles are intended to prevent.

View on HN

This is the real important question. Why didn't the sensor package detect this?

Humans not paying attention can never be solved, but hopefully we can get to the bottom of this failure and then fix it systemically.

Unfortunately given how shady Uber are, we will never be able to trust their technical analysis in the event one is ever released.

I have the very strong suspicion that the driver has turned autonomous drive mode off, perhaps to be able to watch tv or something. She seems extremely distracted by the dashboard and there is no way that sensors could not alert her of that presence. Also, apparently the car was driving at 38mph in an area where the speed limit was 35mph. Would the autonomous mode allow that?

> Also, apparently the car was driving at 38mph in an area where the speed limit was 35mph. Would the autonomous mode allow that?

This seems key. Whether in autonomous mode or not, whether someone got hit or not, this clearly should not have been happening, and as such indicates some malfunction or lapse in protocol.

If autonomous drive had been off then Uber would be the first to say so and not be blamed for this death.

Google Street View seems to show 45 mph - if the imagery is current, that is.

https://www.google.cz/maps/@33.4350531,-111.941492,3a,75y,33...

OTOH, if the issue was "human thought computer was driving, computer was actually off", this might be a completely new class of errors in driving (although it does happen in aviation, where both pilots think the other one is PF: https://aviation.stackexchange.com/questions/5091/how-are-co... )

Some solutions solve this better than others:

http://www.thedrive.com/tech/17083/the-battle-for-best-semi-...

Not to the naked eye.. this is a video. No way does it appear that dark to your eye.

Sure, if we were actually sitting in the front seat we would have a different perspective. Want to convince me it was unavoidable? Release the IR footage too.

Have to agree with you. This is the automatic exposure control of the camera. my 1440p dash cam does the same thing. In its video it looks like everything outside of the cone of the headlights is pitch dark, but it is actually not. I have seen deer standing by the side of the road and later checked the dash cam, which did not pick them up at all.

I think this seems like exactly the scenario that autonomous driving vehicles are intended to prevent. may be optimistic.

Lots of people are hoping they will prevent such scenarios, the motivation for fielding them is more related to making lots of money.

I disagree, if you look at any of the Waymo videos you can see they have great recognition of every car, pedestrian, and obstacle within 100 meters of the car, and most of that data (point cloud lidar/radar at the very least) would work in pitch black without even headlights. Thus it's extremely reasonable to expect that autonomous vehicles will easily prevent scenarios like this.

Inclement conditions, defensive driving, etc. are much harder to work with but this should have been cake.

If you turn your headlights off on a dark country road, your sight capabilities actually go up, not down. The problem is: most roads are lit or you st least have to compete with other headlights, and it would make you harder to see.

There was also a recent lawsuit where Uber pledged to not use any lidar tech and software from Waymo anymore. Maybe they had to deactivate Lidar.

LIDAR in question on this vehicle is an HDL-64, a commercial unit from Velodyne. Lawsuit is entirely unrelated.

If you look at any demo video from the manufacturer, the product is always perfect and amazing.

Go to [0] and [1] and count how many times they use the word "safe." We're not hoping autonomous vehicles are safer, we're being sold on them as the safer alternative. So yes, this type of close call situation is exactly where these vehicles are supposed to be superior.

[0] https://waymo.com/

[1] https://www.uber.com/info/atg/

Why should I take their marketing at face value?

Of course they are selling them as safe. I still think the funding to work on them is chasing big piles of money.

Human-driven car fatalities per day: 120+

Automated-car fatalities total: 1

Seems safer, so far.

Also, nothing's going to make walking out in front of a ton of steel going 30+mph 100% safe.

That's a very misleading comparison, even as an informal, rough comparison for the sake of argument.

Such a comparison would have to take into account the amount of human-driven cars and automated cars - not just taking a picture of a single day, but the variance over time (e.g. if today there are a thousand automated cars in operation, and yesterday there were 50, that can distort the average stats); automated cars aren't driving in certain areas/times/weather conditions whereas human-driven cars are, etc.

Automated cars may be safer, but, open snark – I hope it's not calculating its sensor data in this way — close snark.

This isn't the first automated car fatality though right? Hasn't Tesla had some?

Maybe a better word than "intended" would be "expected". As in, most normal, rational, reasonable people basically familiar with the technology involved expect this to be preventable.

Being buried in lawsuits isn't a great way to make money, generally speaking. The people hoping to make lots of money at this should have a pretty high level of motivation to make it safe.

I was under exactly the same impression. From the description a few days ago, I had imagined someone on the right hand side of the road stepping out from between two parked cars.

From the video this morning I was wondering how it was possible the vehicle lidar and radar didn't pick this up. This is exactly the sort of thing I would hope these additional sensors could pick up easily.

Also from the released video, it was clear to me the "human driver" of the car was not paying attention. Looks like they are looking at their phone at lap level 90% of the time, unless there was something like a "camera/lidar" view in the dash they were looking at.

Having a human in these self driving cars is useless. There is so little for the human to do that it is hard to keep them from checking out. And once they do, they might as well not be there. Seems like "Safety Theater" to me, make the public and the Uber riders feel like there is still a human in the loop, when there really isn't.

Yes, particularly the 'theatre' part. Its expensive and useless.

More useful would be to have a central location with humans monitoring dozens of cars, like a sort of air-traffic control situation. Better chance of keeping their attention with enough going on. They'd be there in case of difficulties, know when a car 'went off the rails' or notify authorities in case of an accident.

And with Central monitoring you could easily have people switch out frequently, and run drills like they do with lifeguards. The lifeguards at our neighborhood pool change positions every 15 or 30 minutes or go on break, with new ones cycling in. The water park does something where the supervisors will periodically introduce a special ball to the water which the lifeguards have to "save".

I worry that monitoring dozens of cars would be too much information for the "Road Traffic Control" to be able to respond in the 1.5 seconds that were available in this video, especially if they have such "low dynamic range" as what we see in this video. But maybe Lidar data would have showed the RTC operator something the car didn't see?

I'm really looking forward to the findings about what radar, lidar, and sonar sensors were saying during this time.

They have a better chance with their attention actually on the road?

What is striking to me is I haven't seen anyone talking about evasive maneuvers. The car must have been able to see the person, and the road looks empty, why not swerve?

At the absolute bare minimum, hit the brakes and reduce the impact. An attentive human driver would have at least started to hit the brakes. The software should have had plenty of time (I'm estimating a full second) to do something productive.

Unlike a human the car has complete situational awareness so it should be able to safely swerve.

Keep in mind the video isnt nessisarily representitive of what you would see with the naked eye. Most cameras have very poor contrast in low light situations that are interspaced with lights like a typical street with streetlamps, your eye does significantly better.

Yeah but the car has IR/LIDAR does it not? Low light should not be a problem there.

I would expect if IR/LIDAR are working properly the car would see the pedestrian as if broad, bright daylight, so why no reaction at all?

Indeed; I'm wondering why the police were so fast to say that the woman suddenly entered to road, when the video shows this clearly isn't the case.

I bet the police chief who gave the quotation saw that video once and was startled at how suddenly the person appeared in that video and didn't check and see things like there was another lane or anything about the surroundings.

I would expect most autonomous systems to extrapolate the movements of anything in the vicinity and check whether it's going to cross paths with the vehicle. I would speculate that LIDAR (or whatever systems they use for object detection) simply failed to detect her.

Maybe it detected her but failed at predicting her path.

Bikes are mainly curved, small pieces of metal, and absorptive pieces of rubber. Their reflectors are designed for viewing them from the front or rear, not the sides. I wonder whether lidar can even differentiate a sideways bike from noise.

Even if the bike itself was too spindly to be picked up, there's no excuse for missing the solid mass of a human body.

In principle, I agree. But, I wonder if holding a bike in front of you is almost like holding a mirror that obscures the lidar ability to detect a solid mass.

edit: but she is in front of the bike, so this shouldn't matter. Does the darkness of her clothing impact any of the sensors?

To hell with the LIDAR, human beings are practically IR flashlights. Aren’t these things supposed to have multiple sensors?

All bikes sold in the United States are required to have reflectors on the wheels for reflection to the right and left. Enthusiasts often remove them.

Even if those in-wheel reflectors are removed, the wheel rims are reflective and the seat post is reflective (at minimum). And additionally there are several things that are way less reflective than a bicycle that you don't want to hit in the middle of the road. Like a deer, or a pothole, or a pedestrian.

A bicycle shouldn't need to be covered in reflectors to be safe from getting run over by a car. However anybody who cycles at night (as I do frequently) should wear reflective clothing because the road is full of reckless drivers.

I don't think this is true. I would be surprised if there were much federal regulation about bicycles.

https://www.law.cornell.edu/cfr/text/16/1512.16

wow... that's crazy.

I guess my bike has the retroreflective sidewalls... or else I was sold a bike illegally.

You can see the side reflectors on the bicycle in this video.

Reply on news.ycombinator.com