Waymo CEO on Fatal Uber Crash: Our Car Would Have Been Able to Handle It

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
John Krafcik is confident enough in his company’s self-driving car technology to declare that the Uber tragedy in Arizona wouldn’t have happened if it was a Waymo vehicle. A safety report received by federal officials suggests that Waymo does have a thorough-enough hardware and testing regimen for preventing similar incidents.

Waymo says its equipment has run through thousands of scenarios focused exclusively on avoiding collisions involving pedestrians. The vehicles are equipped with custom lidar, radar equipment and other sensors, according to Waymo. “We’ve staged people jumping out of canvas bags or porta-potties on the side of the road, skateboarders lying on their boards, and thrown stacks of paper in front of our sensors,” Waymo said in its report.
 
Thing is this company is claiming they test those type of conditions where someone walks right in front of car at night. Most likely its probably going to be something with the hardware or software not programmed right or something.
 
every time I see their name I think:

kQBNxGhz_400x400.jpg
 
Galvin said:
Thing is this company is claiming they test those type of conditions where someone walks right in front of car at night. Most likely its probably going to be something with the hardware or software not programmed right or something.
Click to expand...

Wasn't there an underpass the uber car traversed immediately before hitting the pedestrian? That might have mitigated the circumstances by masking data behind it.

Regardless, making bold unsubstantiated claims will not only change nothing in the industry, only enlarge the target on your forehead if there is ever an incident with waymo ai. It's not like we all became communists when the challenger blew up and Russia said they didn't have those issues with their rockets.
 
Galvin said:
Thing is this company is claiming they test those type of conditions where someone walks right in front of car at night. Most likely its probably going to be something with the hardware or software not programmed right or something.
Click to expand...
They actually -do- test these types of situations. I've seen Waymo testing first hand. It's funny/slightly scary to see Waymo employees on bikes riding in front of the cars on purpose, at different types of intersections and times of day, just to make sure the cars respond appropriately.

And they've been testing like that for years on closed courses.
 
Something has to be wrong with their software. The car didn't even try to slow down. The lady was walk from the left side to the right with a bike! Sure it was a tad dark, but hell what the Lidar doing? I'd love to see the software logs. Makes no sense, sure the lady was in the wrong, but Uber should have caught it.

This article also has tests performed with Uber to see that it clearly stops in testing same location.

https://www.forbes.com/sites/alanoh...ers-failure-to-avoid-pedestrian/#415bd3195cc2
 
Last edited:
Fifliffl said:


Last time an executive was this cocky about a car not hitting a pedestrian.....
Click to expand...

It was actually confirmed that the "City Safe" feature was disabled by the salesman instead of being enabled when this video was shot.
That being said. Volvo's system works and works well. Is it 100% foolproof, NO. Is anything 100% foolproof, NO. Planes still crash, rockets still blow up and they have far more technology running. Uber runs their own system and overrides Volvo's so it would be difficult to say without 100% analyzing all the data exactly where the FAIL came into play but someone darting in front of a car going 50mph, its not going to be able to stop and avoid them 100% of the time. At best maybe an evasive maneuver and that would only be possible if other avoidance systems in the car gives an all clear. So if there where other cars or obstacles beside the vehicle, it will not swerve. Tech will get better but even the future for stupidity will rest with the stupid still being killed. You can't 100% prevent tragedy or stupidity.
 
Maxx said:
https://en.wikipedia.org/wiki/Garry_Hoy
Click to expand...

I probably shouldn't have laughed half as loud as I did reading that. Poor guy had all the trust in the world in the glass, he just didn't think about how good or not it was affixed to the frame. Whatever you got is only ever as good as the weakest link.

In Uber's case it seems the camera and radar were complete shit considering that numerous independent civilians have their own evidence of just how well lit the area of the fatality occurred actually is. The pedestrian didn't actually 'come out of the shadows' as implied by Uber and the Sheriff.
https://arstechnica.com/cars/2018/0...victim-came-from-the-shadows-dont-believe-it/
 
ol1bit said:
Sure it was a tad dark, but hell what the Lidar doing? I'd love to see the software logs.
Click to expand...

It's worth noting that the Uber car apparently uses a Velodyne LIDAR.

https://www.forbes.com/sites/alanoh...ers-failure-to-avoid-pedestrian/#e61ee35cc2d9

It works by sending out infrared laser pulses and measures the amplitude and delay of the return. The signal to noise ratio is highest when the only light is coming from the infrared lasers in the LIDAR.

In short, LIDARs work best in the dark. They are weakest in the daytime. By the way, most people use them in the daytime. I've never used one but some cursory googling verified that they work best at night. You'd get even better LIDAR output in this case if you turned out the street lights and the car's headlights.

Of course, there's more sensors than just the LIDAR in the Uber car. But if any one sensor saw an object in front of the car you'd think the software would at least alert the human driver if not hit the brakes.

The Velodyne datasheet on their cheapest model said that the thing sends its data out in UDP packets on a 100Mb network connection. The data rate's not that high. Presumably the car has a recording of the raw data.
 
Dekoth-E- said:
I mean that's great, but at what point do we draw the line and say "look, if you are stupid enough to run in front of a car, you're going to die"?
Click to expand...
One day when 95% of the cars on the road are automated I could see a system that A) doesn't kill the pedestrian and B) automatically sends them a jaywalking fine if it's clear they should have waited or checked traffic prior. Maybe not even a monetary fine but community service hours that force them to be more cautious in the future.
 
As I've long noted, the software is going to be the weak link. There's simply a ton of data that needs to be integrated, and there's always going to be fringe conditions the software won't be able to handle.
 
I still wonder how well all this tech will play with each when there are many cars from different companies running around? Will sensor signals from one car interfere with the sensors on another? There aren't unlimited frequencies available for these signals. Fairly easy for there to be 50~100 cars at a busy intersection. If they are all sending out sensor signals, will they confuse each other? Its all fine that Waymo has tested one car and one pedestrian scenarios. Have they tested 50 car scenarios with a mix of pedestrians, bicycles, dogs, squirrels, flying plastic bags, etc. Better be sure before you start boasting that your car is better then your competitors.
 
But how many people will have to die before this technology is perfected. I mean, if keeping these dangerous cars off the roads saves just one life, isn't it worth it? Think of the children! /s

Now, a bit more seriously, what if it had been a child hit and killed? I think we'd be seeing lots of people calling for the immediate end to live street use of these vehicles.
 
jardows said:
Now, a bit more seriously, what if it had been a child hit and killed? I think we'd be seeing lots of people calling for the immediate end to live street use of these vehicles
Click to expand...

we would, and we would also see a ton of posts blaming the parents for bad parenting.
 
to be fair, on that video the guy came out of nowhere, any regular driver wouldn't be able to avoid him, it was like 1 second or so, if the pedestrian wouldn't, the occupant of the car would, 60 to 0 in 1 sec that's like hiting a wall.
 
^ yup

anybody here see the Renault auto-pilot? LTT did a vid on it and it is pretty impressive and a true auto-pilot but they are being careful and only doing limited testing on public roads.
 
Dekoth-E- said:
I mean that's great, but at what point do we draw the line and say "look, if you are stupid enough to run in front of a car, you're going to die"?
Click to expand...


The way I see it, two entirely separate issues, one is "fault", as you say, someone just walked out in front of a vehicle at night and it seems never saw it coming. And the other is the vehicle's failure, and that was a failure, the car should have reacted in some way, failing to avoid any object and a certain collision was a failure of that vehicle's systems, or a failure of it's decision programing, one or the other.

I think people on all sides of this discussion need to recognize this, it's not a simple issue, it's multiple issues.
 
lcpiper said:
The way I see it, two entirely separate issues, one is "fault", as you say, someone just walked out in front of a vehicle at night and it seems never saw it coming. And the other is the vehicle's failure, and that was a failure, the car should have reacted in some way, failing to avoid any object and a certain collision was a failure of that vehicle's systems, or a failure of it's decision programing, one or the other.

I think people on all sides of this discussion need to recognize this, it's not a simple issue, it's multiple issues.
Click to expand...
I've been trying to get people to see this as well but everyone wants to single out the 1 point of failure (on either side) and not own up to the rest. Hell, people are trying to say that we are against new tech because Uber failed.
Nope, not the case, Uber needs to stop testing live, and re-evaluate its entire programming to make sure it doesn't happen again and people are stupid and walk out infront of cars on higher speed limit roads.
Only way to fix stupid is to get rid of it. So I guess this case it was handled?
 
I think this sums up the accident.
The car failed to detect an object moving in a collision vector with it. Nothing else needs to be said here. This was a gross systems failure resulting in someones death.
posted by BSmith in an earlier uber thread
 
lcpiper said:
The way I see it, two entirely separate issues, one is "fault", as you say, someone just walked out in front of a vehicle at night and it seems never saw it coming. And the other is the vehicle's failure, and that was a failure, the car should have reacted in some way, failing to avoid any object and a certain collision was a failure of that vehicle's systems, or a failure of it's decision programing, one or the other.

I think people on all sides of this discussion need to recognize this, it's not a simple issue, it's multiple issues.
Click to expand...
Granted I don't disagree that the system failed. I'm just pointing out that I don't view it as a tragedy because someone paid the price for being stupid.
 
Dekoth-E- said:
Granted I don't disagree that the system failed. I'm just pointing out that I don't view it as a tragedy because someone paid the price for being stupid.
Click to expand...
I don't view it as a tragedy either but rather an unfortunate accident, but it sure sends red flags.
 
Dekoth-E- said:
Granted I don't disagree that the system failed. I'm just pointing out that I don't view it as a tragedy because someone paid the price for being stupid.
Click to expand...


Thing is, let's play a game and see if the logic holds up. Let's change something in the scenario. Let's say the car was no longer an autonomously driven vehicle, instead it's just another car on the road and just another driver.

Now what if the driver was drunk? The pedestrian was no less stupid but ..... how would that one play out in the courts?

And if the driver wasn't drunk, just distracted ...... ?

And if the lighting at the scene wasn't as bad as the video showed..... ?

The point is, as long as there is nothing negative in the driver's part of the equation then it's just another sad accident resulting in a fatality, no need for charges. But as soon as the driver is at fault in some way things start to change.

Now as I said early, no charges are likely because the pedestrian was in fact in the wrong. It's been several days now, going on a week even, charges become less likely everyday passed.

So my one point here with all the if's and maybe's is this, what's the difference in a drunk driver and a failed accident avoidance system?

You know, i fit had just been a drunk driver they would be throwing away the key right? So why not with malfunctioning and incapable systems that were supposed to be up to the task.

Unless of course you feel that it's unfair to expect that this car should have at least tried to avoid hitting the woman.
 
lcpiper said:
Thing is, let's play a game and see if the logic holds up. Let's change something in the scenario. Let's say the car was no longer an autonomously driven vehicle, instead it's just another car on the road and just another driver.

Now what if the driver was drunk? The pedestrian was no less stupid but ..... how would that one play out in the courts?

And if the driver wasn't drunk, just distracted ...... ?

And if the lighting at the scene wasn't as bad as the video showed..... ?

The point is, as long as there is nothing negative in the driver's part of the equation then it's just another sad accident resulting in a fatality, no need for charges. But as soon as the driver is at fault in some way things start to change.

Now as I said early, no charges are likely because the pedestrian was in fact in the wrong. It's been several days now, going on a week even, charges become less likely everyday passed.

So my one point here with all the if's and maybe's is this, what's the difference in a drunk driver and a failed accident avoidance system?

You know, i fit had just been a drunk driver they would be throwing away the key right? So why not with malfunctioning and incapable systems that were supposed to be up to the task.

Unless of course you feel that it's unfair to expect that this car should have at least tried to avoid hitting the woman.
Click to expand...

1) Drunk driver, going to jail because it is illegal regardless of hitting someone. Ped is still a dumbass and got what they deserved.

2) Distraction can't be proven so chances are driver gets off and has to live with guilt the rest of their lives that is a direct result of ped being a dumbass.

3) Lighting is really irrelevant from a driving POV. Car sensors shouldn't rely on light in the visible spectrum. If they do that is a pretty serious design flaw. Even in bad lighting any Ped is going to see car headlights and or should. Their actions are still stupid in this scenario.

I don't feel it is unfair to expect that this car should have at least tried to avoid hitting the woman.
 
Dekoth-E- said:
1) Drunk driver, going to jail because it is illegal regardless of hitting someone. Ped is still a dumbass and got what they deserved.

2) Distraction can't be proven so chances are driver gets off and has to live with guilt the rest of their lives that is a direct result of ped being a dumbass.

3) Lighting is really irrelevant from a driving POV. Car sensors shouldn't rely on light in the visible spectrum. If they do that is a pretty serious design flaw. Even in bad lighting any Ped is going to see car headlights and or should. Their actions are still stupid in this scenario.

I don't feel it is unfair to expect that this car should have at least tried to avoid hitting the woman.
Click to expand...


Umm, #3, remember I was talking about a regular human driver.... but no matter, it's all bullshit and don't matter. What matters is what will happen and it looks like it's going to happen just as we both think it's going to happen. No charges, and the cars come off the roads giving all these companies another opportunity to review their accident avoidance features and assess their drivers and the roles required of them.
 
Jim Kim said:
I think this sums up the accident.
The car failed to detect an object moving in a collision vector with it. Nothing else needs to be said here. This was a gross systems failure resulting in someones death.
posted by BSmith in an earlier uber thread
Click to expand...
Pretty much that. There's an expectation that the vehicles can account for obstacles. Even if the person threw themselves in front of the car and the laws of physics ensure that a collision happened, the car is expected to have detected and reacted. It didn't.

It's like walking around doing the windmill and blaming others for getting in the way of my fists.. that's not where the fault lies, and someone died in this case.

The "interesting" part from here is determining who is at fault/taking responsibility. It's not an act of God, so it's either the company, the driver watching his phone, the Timmy the intern who did the last code commit.
 
pendragon1 said:
it didn't fail it was disabled BIIIIG difference.
Click to expand...

The default Volvo stuff was disabled; Uber's sensors and software package wasn't, and should have detected this.

Forget edge cases; this is a basic case that if this tech can't deal with 100% of the time, it isn't roadworthy.

Dead Parrot said:
I still wonder how well all this tech will play with each when there are many cars from different companies running around? Will sensor signals from one car interfere with the sensors on another? There aren't unlimited frequencies available for these signals. Fairly easy for there to be 50~100 cars at a busy intersection. If they are all sending out sensor signals, will they confuse each other? Its all fine that Waymo has tested one car and one pedestrian scenarios. Have they tested 50 car scenarios with a mix of pedestrians, bicycles, dogs, squirrels, flying plastic bags, etc. Better be sure before you start boasting that your car is better then your competitors.
Click to expand...

I would hope, but have seen nowhere, that there could be some kind of standardized networking protocol for a mesh network, so that more cars being automated actually means safer driving, as they're communicating up and down the line, not requiring each individual car to rediscover what is going on.
 
You must log in or register to reply here.
Back
Top