A step backwards for Self Driving Cars.

Page may contain affiliate links. Please see terms for details.
Of course there may well have been insufficient time for the human driver to react.
It'll be interesting to see the onboard video of the incident...
 
"There is video of the crash, which investigators are examining but not been released to the public. "It’s very clear it would have been difficult to avoid this collision in any kind of mode (autonomous or human-driven) based on how she came from the shadows right into the roadway," Moir said. Police have previously said Herzberg was not using a crosswalk."

Uber is 'likely' not at fault in the fatal self-driving car crash, local police chief says
 
I suppose it's a bit of a kick in the teeth for the whole self-driving car fans but there are some accidents that are literally unavoidable. Timing is bad but actually this could be one of those incidents that would have happened regardless.

I'm intrigued by the whole autonomous car idea, in one respect I'd love my car to drive me to and from some where without me having to worry. I'd hate to be forced into it though.
 
I used my W213 E-class to drive me 25 miles along the A34 from Winchester to Newbury yesterday. It did it flawlessly. Adjusted the speed well, kept it within the lane, let me move across the lanes safely when changing lane, coped with vehicles moving into my lange. Hugely impressed, but I missed the experience of driving by the end of it lol
 
well the video is out - lady in dark clothing crossing an unlit part of the road with her bike, I'm afraid it's the pedestrian's fault.
 
May not be so clear cut? Would a human driver have swerved at the instant the headlamps illuminated the pedestrian? Obviously when viewing the video we have the prior knowledge something is about to happen.
Can't tell if the Uber car initiated any avoiding actions, even braking? . The human 'driver' does make a move as if to grab the steering wheel

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.
 
Last edited:
Its obviously a visibility problem on an unlit bit of highway ----which poses the question why weren't the cars headlights on full beam with no oncoming traffic. As for the driver backup "the lights are on but there's nobody at home" seems apposite. Looking down at the instruments? rather than the road in front his function here presumably is not to act as the car's constant "wingman" but to act as some sort of emergency backup in case of systems failure. The self drive function did not fail as such it merely experienced a situation beyond its operational parameters I am assuming driving by camera visual imaging here and not radar which should have picked up a moving object at distance in pitch dark. From the video the woman did not just "step on to the road" as might be inferred from the description but was already half way across the highway??
 
From the letters page of a national newspaper today:
Developers of driverless cars are discovering what airliner manufacturers have also discovered, sometimes the hard way, over the past 60 years: humans are good at controlling vehicles and try very hard to avoid accidents, while computers are excellent at monitoring and never get bored.

Reverse the roles and we find that humans get bored when reduced to a monitoring role, while computers do not mind killing themselves.
There's much truth in that.
 
If the driver was paying attention to the road ahead (instead of the instrumentation) then they'd have a better chance of spotting the pedestrian, who didn't help by seeming oblivious to the lit car bearing down on them.

I suspect the vision of the road sitting inside the car is far better than the dashcam suggests. The pedestrian didn't step out from between parked vehicles, there were zero obstructions. Also the road was not completely dark as the pic shows, soon after the collision.
1ac711e0a87ca02384661f38da1ecfc4.jpg
IMO the pedestrian caused the collision, but the driver had the opportunity to take evasive action.

Sent from my G3112 using Tapatalk
 
Quote:-
Self-driving cars rely on a combination of sensors and data systems to navigate and avoid obstacles. The vehicles typically include some combination of global positioning systems (GPS), light detection and ranging (LiDAR) sensors, radar, cameras and other equipment to help detect lane markings, bicycles, other vehicles and pedestrians. Each of these systems has particular strengths and weaknesses. “One of the things that we have noticed about accidents involving self-driving cars is that they seem strange from a human perspective; for example, the vehicles do not hit the brakes prior to the collision, which is something most human drivers do,” says Bart Selman, a computer science professor at Cornell University and director of the Intelligent Information Systems Institute. “That’s because the vehicles make decisions based on what their sensors detect. If its sensors don’t detect anything, the vehicle won’t react at all.”

from

Uber Self-Driving Car Fatality Reveals the Technology's Blind Spots


WAS THE VOLVO FITTED WITH AN OPERATIONAL LIDAR SYSTEM following Uber's recent dispute with WAYMO?
Uber isn't even using the custom self-driving-car tech at the heart of a Google lawsuit
Uber and its 245 million reasons to settle with Google's Waymo
Uber v Google trade secrets case begins
 
Last edited:
This is pretty convincing that Lidar should have picked this woman up in pitch darkness assuming it was operational. How the system chose to interpret this data is another matter of course.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.
 
This story gets more bizarre by the minute - even if you take the 'driverless' car out of the equation . The deceased was pushing a bicycle across a 6 lane highway, she was 'estranged' from her family and lived in a tent in the woods. The 'driver was an ex con with a terrible driving history who had somehow become a 'woman' but spent time in jail as a male even when 'female'...I am NOT making this up . You couldn't make this up !

I hope its fake news...Mr Trump ..?
 
Having watched the video a number of times I reckon the time between the pedestrian being visible to the point of impact is north of 2 seconds. It's reasonable to assume that the footage is darker than the reality would have been, so it's likely that the pedestrian would have been visible to an alert driver for approaching 3 seconds prior to the recorded impact. With reaction times ranging from 0.7 to 3 seconds, the pedestrian could well have been hit by a "normally driven" car, but it's also quite possible that an alert driver would have started to take evasive action, which may have mitigated the outcome.

Frankly, I tend to agree with the assessments in the Grauniad article linked by grober that this was a gross failure of the self-driving technology, which doesn't need to exclusively rely upon visible-spectrum light as does the mk1 eyeball.
 
Though not obvious if a factor here, one thing to keep in mind is that the human brain uses predictions based on past experience.

Many drivers will have experienced the feeling where you look at the car in front of you and you just know there's a good chance it will change lane. Why? Because the car is edging the line. Because it is tailgating the car in front. Because it changed lanes sharply a couple of miles earlier. Because it's a WRX with a wing the size of Wales.... whatever.

Driving outside main train stations at rush hour? Pedestrians will always 'overflow' into the road, crossing at red light, often in a hurry.

The lady pushing a pram or a puchair moving along the pavement in parallel to your car? If there's a zebra crossing ahead, she might make a 90 degrees turn into the road - prams are not very manouverable, unlike pedestrians they don't turn diagonally towards the crossing but make a last minute sharp turn instead.

Football rolling into the road? You avoid the ball, while looking-out for the child that might follow it running.

Ice cream van? School bus? Etc.

You don't have to be able to predict the behavior and intentions of other road users with 100% accuracy, just reliably enough to know when to prepare for evasive action in advance.

Relying on sensor input alone is not suffienct if you want to fully immitate the behaviour of a human driver.
 
You don't have to be able to predict the behavior and intentions of other road users with 100% accuracy, just reliably enough to know when to prepare for evasive action in advance.

Relying on sensor input alone is not suffienct if you want to fully immitate the behaviour of a human driver.
Very true.
 
The internal view appears to show the driver looking down at something, look up and make an involuntary facial expression as he realised what was happening/ about to happen.

If you watch the video closely, you will see a rectangular light in the distance (the smallest of three at the bottom of a slope on the left sloped this way- \) momentarily become obscured as the pedestrian passes 'in front' of it.

Add in the fact that headlights 'bleed light' outside their refraction beam i.e. the 'solid' edge portrayed on the road surface is not the absolute limit of their range, it's quite likely the true eyes-on view would have shown more than the camera has captured.

The pedestrian was there to be seen and a driver who was paying attention may have not only have seen her in his headlights but been alerted to her presence because of the light in the distance becoming momentarily obscured.

The video demonstrates perfectly a concern of mine regarding 'driverless' cars; the system failed AND the driver wasn't paying attention, possibly due to complacency i.e. relying on the systems.
 

Users who are viewing this thread

Back
Top Bottom