We still don't know the reason why she crossed the road like that. Certain roads tend to be accident black-spots because of poor design or blindspots. It is entirely possible that visibility was poor and she didn't anticipate the UBER car coming so fast despite looking both ways.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Self-Driving Uber Car Kills Pedestrian in Arizona, Accident Could Have Implications for Autonomous Vehicle Testing
- Thread starter MacRumors
- Start date
- Sort by reaction score
I disagree. The video shows that she should have been clearly seen by the human backup in the car who was looking at their phone or something else. Cameras pick up light heavily so they make the road appear darker than it would for a driver in the cab.
I have avoided worse collisions on the interstate where I live as for some reason people occasionally cross the interstate like a street... Joy of living in a major city.
I can’t say definitively that the impact would have not happened, but I would have expected heavy braking at a minimum since this lady crosses a whole travel lane.
We both agree there is no reason the cars sensors should have been unable to see her cross an entire empty lane of traffic. Additionally, these cars should be equipped with eye tracking to ensure the human backup drivers are actually looking at the road. Many new cars already use this to detect drowsy drivers such as Mercedes Attention Asssit. This isn’t new and should be leveraged here.
I also don’t get why these cars aren’t required to include heat tracking alongside its radar and LiDAR systems to help keep them from missing people... Mercedes has been including this optionally in its cars for 5+ years and it should be a minimum on autonomous driving vehicles.
http://www.autolivnightvision.com/vehicles/mercedes/
I guess if you have an s-class Amg or maybach you can get these great options. But hopefully this accident will slow this entire effort down. For me, I will accept this technology when a commercial airliner has self flying autonomous technology and there’s only one pilot aboard...then you know it’s ready for prime time.
Here's my take on this ... based on my personal driving experience, the Google Earth images of the intersection, the video that was just released, and by some of the local forum members.
1. The car's sensors should have seen the pedestrian way before the point of impact. Sensors shouldn't need the scene to be lighted for them to work.
2. Based on the video, it's hard to say whether the Uber driver should have seen the pedestrian and taken over controls to stop or slow the car. Surely, the driver should not have been looking down as has been reported.
3. Based on what local forum member have said, this was a popular spot for jaywalking. I know, in my experience, that I drive differently in areas where local behavior differs from the letter of the law. For instance, there's an intersection near me that allows for "right on red", but I rarely do that because I know that the cars making the left (with the light) cut that corner all the time. So if I were to go "right on red", I would end up in an accident. So what I'm saying is that, perhaps, local drivers in the area are aware that jaywalking is prevalent here, they will tend to pay attention and slow down as they come up to the stoplight. Maybe Uber (and other AI systems) should program in local behavior into their algorithms, if they don't already do so.
4. That median is very poorly designed. If you don't want pedestrians crossing over via the median, then don't put walking paths (or sidewalks) on the frigg'n median. It's simple.
5. The DOT or NHSA should work with the AI companies to determine what kinds of sensors need to be utilized on self-driving cars. LIDAR, RADAR, infrared, etc. Also, I'm not sure what I think about mandating that the AI companies having to share data with each other, in order to speed up development of these kinds of systems ... I'll need more time to think this through.
1. The car's sensors should have seen the pedestrian way before the point of impact. Sensors shouldn't need the scene to be lighted for them to work.
2. Based on the video, it's hard to say whether the Uber driver should have seen the pedestrian and taken over controls to stop or slow the car. Surely, the driver should not have been looking down as has been reported.
3. Based on what local forum member have said, this was a popular spot for jaywalking. I know, in my experience, that I drive differently in areas where local behavior differs from the letter of the law. For instance, there's an intersection near me that allows for "right on red", but I rarely do that because I know that the cars making the left (with the light) cut that corner all the time. So if I were to go "right on red", I would end up in an accident. So what I'm saying is that, perhaps, local drivers in the area are aware that jaywalking is prevalent here, they will tend to pay attention and slow down as they come up to the stoplight. Maybe Uber (and other AI systems) should program in local behavior into their algorithms, if they don't already do so.
4. That median is very poorly designed. If you don't want pedestrians crossing over via the median, then don't put walking paths (or sidewalks) on the frigg'n median. It's simple.
5. The DOT or NHSA should work with the AI companies to determine what kinds of sensors need to be utilized on self-driving cars. LIDAR, RADAR, infrared, etc. Also, I'm not sure what I think about mandating that the AI companies having to share data with each other, in order to speed up development of these kinds of systems ... I'll need more time to think this through.
Ok I've seen the new dashcam footage. I agree with others that the human operator was definitely looking elsewhere, not concentrating on the road, just before the accident. It was only when it was too late that she finally looked up at the road. It also looks like there was definitely room for a good driver to have swerved into the other side of the road. I've avoided similar collisions before with drunks and even wild animals running into the road.
Also jaywalking is not a crime on most parts of the world, in fact it is considered normal, if autonomous vehicles are to gain mainstream worldwide acceptance they need to be safe for pedestrians. Just imagine this thing in a busy school-zone during peak hours where kids just cross the street without looking.
Also jaywalking is not a crime on most parts of the world, in fact it is considered normal, if autonomous vehicles are to gain mainstream worldwide acceptance they need to be safe for pedestrians. Just imagine this thing in a busy school-zone during peak hours where kids just cross the street without looking.
Last edited:
Lets remember that many of these sensors these vehicles are using don't work properly in rain.. sonar, laser, etc.
That's true. Which raises some concerns for their use in bad weather conditions. But in the footage released the conditions looked pretty clear, the Lidar sensor should in theory have been able to detect the pedestrian in the darkness while she was still on the other side of the road.
Here's footage of the crash below. Dash cams tend to perform very poorly in low light, they definitely don't pick up as much detail as the human eye at night time. So a lot of what he says is conjecture based on very little information.
I have viewed the video posted by miscend. The sensors had an unobstructed view and failed to pick up the woman. It does not matter why she crossed the road, if it was legal or not, it does not change the fact that the sensors missed picking her up. The driver is always in charge of the car no matter what mode it is in. The Uber driver's attention was not on the road where it belonged. A lawyer is bound to point out that reasonable person when testing an experimental system where there is risk of someone being injured or killed, is going to be especially vigilant. I believe both the driver and UBER will be ultimately found liable for this accident.
I'd love to see these cars try to find their way around in a snow storm where they can't "see" anything. Heck, put them in cities with more intense driving conditions like New Dehli, Cairo, or Hanoi.
There's so much these vehicles can't "see."
This one is particularly odd because the vehicle absolutely should have picked up on something on the road. The weather was clear, the road was clear... if only the driver had been paying attention.
Did you watch the video? The pedestrian was crossing in pitch black also with no cross walk. Arguably, if it was really that dark, even an actual driver wouldn't see the pedestrian in time.
[doublepost=1521739184][/doublepost]
Here's the jaywalking law in Arizona:
https://azbikelaw.org/jaywalking-in-arizona/
"28-793. Crossing at other than crosswalk
A. A pedestrian crossing a roadway at any point other than within a marked crosswalk or within an unmarked crosswalk at an intersection shall yield the right-of-way to all vehicles on the roadway.
B. A pedestrian crossing a roadway at a point where a pedestrian tunnel or overhead pedestrian crossing has been provided shall yield the right-of-way to all vehicles on the roadway.
C. Between adjacent intersections at which traffic control signals are in operation, pedestrians shall not cross at any place except in a marked crosswalk."
Videos make the toad appear darker than it would to the driver in the car. If the driver couldn’t see more than a few feet in front as you suggest they either should had had high beams on or pulled the car off the road. You shouldn’t out run your headlights...
The jaywalker shouldn’t have been jaywalking but this is a Albert much avoidable crash. And the fact the sensors failed to pick the person up on a clear empty road is inexcusable. These cars can “see” at night — or so they claim.
It's the camera that makes it look that dark.
Our eyes can see in the dark much better than a dashcam that's only exposed for the bright areas of the scene.
The person crossing the road is actually very near to a street lamp so there would be some light on her. Most drivers have seen people or animals in or near the road in very dark situations. In this case, the driver simply wasn't looking.
However, why didn't the car radar/lidar pick up on it?
Single pilot in that type of aircraft ... more of a regulatory hurdle than a technology hurdle.
As for LIDAR and other tech, Uber's should have them.
- from WayMo Wiki
[doublepost=1521743868][/doublepost]
It's being worked on. ARTICLE
Then again, Tempe Az is not exactly well known for snow or frequent rain. Dust-storms? Haboob? Yes.
Regulatory hurdle is the same for self driving cars. Im sure there’re is enough technology in self driving cars to keep the death rate sensible. So what if there are a few glitches. As far as commercial aviation, I wouldn’t get on a commercial airliner with one pilot, but sure let’s get this moving to inspire confidence in the self-driving car effort.
Clearly the autonomous system, driver and Uber failed. 40MPH is relatively slow so even if the sensors failed to pick up the pedestrian the driver had sufficient time to stop or swerve left if he had been paying attention. Uber also failed since they didn't review in-car video to see that the driver they hired has a habit of distracting himself with phone usage.
Regarding the video posted by miscend
I noticed the newer sensor package has 7 cameras and 1 laser along with GPS and other stuff. The older sensor package on the Ford had 20 cameras and 7 lasers. I wonder why Uber got rid of 13 cameras and 6 lasers.
I hope for Uber's sake, it wasn't a cost cutting decision.
I noticed the newer sensor package has 7 cameras and 1 laser along with GPS and other stuff. The older sensor package on the Ford had 20 cameras and 7 lasers. I wonder why Uber got rid of 13 cameras and 6 lasers.
I hope for Uber's sake, it wasn't a cost cutting decision.
Even if there were fewer vehicles, there wouldn't be fewer vehicle-hours on the road unless the number of total trips was reduced. Autonomous vehicles don't do anything to reduce transit demand.
I personally doubt the uber-rational future scenario of consumers relinquishing car ownership en masse just because they can be driven instead. People are far more emotionally attached to cars than technology enthusiasts like to believe.
[doublepost=1521765458][/doublepost][QUOTE="deanthedev, post: 25908547, member: 1097216]@VulchR is talking about the number of accidents in total. The most important number is the rate of accidents or fatalities per miles driven.[/QUOTE]
My point pertained to VulchR's standards of matching human driving performance, rather than exceeding it, being an acceptable threshold. That discussion encompasses fatalities per miles driven.
Last edited:
She crossed less than 100 meters away from a lit crosswalk, according to the police. It was unquestionably a stupid decision on her part.
[doublepost=1521765777][/doublepost]
I agree. The ped came out of the shadows, but a human could definitely react in that time, and the car should have night vision anyway. I don't think most humans would be able to avoid it, but you would have seen the car swerve or slow down at least.
Last edited:
In case not posted yet:
How Uber's Self-Driving Technology Could Have Failed In The Fatal Tempe Crash - Forbes
https://www.forbes.com/sites/jimmcpherson/2018/03/20/uber-autonomous-crash-death/#176f1a637fbe
Besides dirty sensors, some possibilities include miscategorizing the pedestrian, or not avoiding her because the car would've had to leave its lane to do so.
How Uber's Self-Driving Technology Could Have Failed In The Fatal Tempe Crash - Forbes
https://www.forbes.com/sites/jimmcpherson/2018/03/20/uber-autonomous-crash-death/#176f1a637fbe
Besides dirty sensors, some possibilities include miscategorizing the pedestrian, or not avoiding her because the car would've had to leave its lane to do so.
I made my comments Tuesday before the Dash cam video release. I made an incorrect assumption about the direction Herzberg came from. From the median the Uber's sensors (LiDAR/radar) should have had plenty of time to ID her, and track her direction of movement as clearly hazardous. Even if it failed to ID her as "people". Coming from that spot of median, and crossing one extra lane of traffic. Like wise a defensive human driver who was properly scanning the road may spotted her, and begun to break and divert left (into a bike lane), along with possibly jamming on the car horn as an audible alert to the oblivious jaywalker.
Would that have been enough? At 40 MPH? I personally don't think so. Instead of impacting on the left/passenger side there would have been an impact on the right/driver side.
While Arizona law is on the side of the Uber and the "Safety Driver", the failure to avoid (or attempt to avoid) the collision is on Uber's tech... and their driver.... Who was doing exactly what many inconsiderates do these days without an autonomous car... checking their god damn phones when they should be operating their kinetic projectiles.
This exact same accident could have happened anyways, without the robot car, because the driver wasn't paying attention or had attention diverted from the road for the critical few seconds. Which should be a reminder to us all, as drivers, our 1 and only job is operate our vehicles. If you need to use your phone, exit the roadway, park, and then use it.
While Arizona law may not require compensation to Herzberg's family (and will find her at fault), Uber morally does (also from a PR stand). Their tech and their driver both failed. At exactly the moment they were supposed react the most. Uber also needs to turn over all the data from this crash to investigators, and the full code base for the detection and reaction/decision systems.
It's starting to look like Uber needed Waymo's technology they stole.
Would that have been enough? At 40 MPH? I personally don't think so. Instead of impacting on the left/passenger side there would have been an impact on the right/driver side.
While Arizona law is on the side of the Uber and the "Safety Driver", the failure to avoid (or attempt to avoid) the collision is on Uber's tech... and their driver.... Who was doing exactly what many inconsiderates do these days without an autonomous car... checking their god damn phones when they should be operating their kinetic projectiles.
This exact same accident could have happened anyways, without the robot car, because the driver wasn't paying attention or had attention diverted from the road for the critical few seconds. Which should be a reminder to us all, as drivers, our 1 and only job is operate our vehicles. If you need to use your phone, exit the roadway, park, and then use it.
While Arizona law may not require compensation to Herzberg's family (and will find her at fault), Uber morally does (also from a PR stand). Their tech and their driver both failed. At exactly the moment they were supposed react the most. Uber also needs to turn over all the data from this crash to investigators, and the full code base for the detection and reaction/decision systems.
It's starting to look like Uber needed Waymo's technology they stole.
I rather suspect that the per passenger-mile death toll of horse (think Christopher Reeve) and carriage transportation was probably higher, but I doubt that proper statistics are available.
We've accepted the toll from manually driven automobiles because we didn't have any better alternative. We're on the cusp of that now.
We look back on how dangerous some aspects of life were as recently as 50 years ago and marvel at how we ever accepted it. In another generation, this will be another such example.
I've heard the Streets of San Francisco described a lot of ways, but simple and straight-forward has never been one of them.
Oh, it's not London, certainly, but the only thing missing from your description is snow (but lately we've had a lot of rain).
[doublepost=1521825294][/doublepost]
Airplanes have been capable of autonomously performing the entire journey from chock to chock without any human assistance for over 50 years now. They don't do that simply because the pilots are not and have never been a major cost in air travel (paying the flight crew is at most $10 of your ticket price).
In fact, there are at least three examples where not having a flight crew would have prevented a rogue pilot from intentionally crashing the plane and killing everyone aboard. And that's even before 2001. The single most deadly airline accident of all time was firmly and absolutely a scenario an autonomous system would have never fell into.
Last edited: