Discussion
ash73 said:
It wasn't just the system that was flawed it was their whole approach to science; foolhardy bravado for the camera, no risk assessment, no understanding of the physics, not even understanding their own code etc. Guy was probably quite lucky it crashed almost immediately in a slow corner.
I was pretty shocked by how the whole thing came across to be honest. They have epic budget, some great companies on board developing the tech they can't do in house / don't know how to do (the mechatronics and hydraulics are incredible for example) but the staff/team didn't come across as people who have been involved in any sort of professional motorsport or OEM automotive development program. The "you can hold onto the the roll cage Guy" bit was properly too! I guess it's difficult to attract people to the project in some ways as the majority of really good motorsport people want to be involved in competitive motorsport and the majority of really good automotive guys want succesful projects on their CV, a bit of doubt about the viability goes a long way to putting the good guys off.
Abbott said:
You might find this an interesting insight
https://www.ted.com/talks/chris_urmson_how_a_drive...
That's really interesting, thanks for posting. https://www.ted.com/talks/chris_urmson_how_a_drive...
Kind of shoots down the "AI isn't sophisticated enough to deal with the real world" argument - the AI can:
- recognise a cyclist's arm signals
- give way to car jumping a red light at a junction
- tell the difference between a normal car and a police car
- navigate through temporary road work diversions
- deal with a woman in an electric wheelchair chasing a duck in circles in the middle of the road!
youngsyr said:
Abbott said:
You might find this an interesting insight
https://www.ted.com/talks/chris_urmson_how_a_drive...
That's really interesting, thanks for posting. https://www.ted.com/talks/chris_urmson_how_a_drive...
Kind of shoots down the "AI isn't sophisticated enough to deal with the real world" argument - the AI can:
- recognise a cyclist's arm signals
- give way to car jumping a red light at a junction
- tell the difference between a normal car and a police car
- navigate through temporary road work diversions
- deal with a woman in an electric wheelchair chasing a duck in circles in the middle of the road!
How many cyclists use arm signals? How many use a nod of a head of a facial expression to indicate their intent.
ash73 said:
StevieBee said:
youngsyr said:
Abbott said:
You might find this an interesting insight
https://www.ted.com/talks/chris_urmson_how_a_drive...
That's really interesting, thanks for posting. https://www.ted.com/talks/chris_urmson_how_a_drive...
Kind of shoots down the "AI isn't sophisticated enough to deal with the real world" argument - the AI can:
- recognise a cyclist's arm signals
- give way to car jumping a red light at a junction
- tell the difference between a normal car and a police car
- navigate through temporary road work diversions
- deal with a woman in an electric wheelchair chasing a duck in circles in the middle of the road!
How many cyclists use arm signals? How many use a nod of a head of a facial expression to indicate their intent.
Afterall, intelligence is simply identifying and responding to a situation correctly.
If you write enough rules to cover every single possibility, what's the difference between that and a human made decision?
Their video is interesting but...
How does it cope in locations that aren't already mapped? Lane detection, signs, rules & all that.
How does it cope without other vehicles providing data - a lot of what was shown was outside the detection range of the car in the demo?
How does it cope with lots of other LIDAR sensors in the same environment, or with degraded sensors?
How does it cope with running at speeds in excess of 15 or 20mph?
What's the sensor latency? For visual stuff this can be a right pain even with part-frame processing. And sensor fusion is even worse.
How deterministic are their various models?
How many entity 'intent' detections can it cope with simultaneously, and to what level does it look at how these might interact and anticipate probable events?
Does it bother with behaviour modelling of entities in the environment so it can anticipate what a specific one will do based on past observation, or does it just do a continuous rolling assessment?
How quickly can it cope with new entities appearing? Nice straight grids are one thing, but will it panic seeing a car appear as you crest a hill/bridge with a corner mixed in for extra complexity?
Or in other words yes it looks like a shiny demo, but...
How does it cope in locations that aren't already mapped? Lane detection, signs, rules & all that.
How does it cope without other vehicles providing data - a lot of what was shown was outside the detection range of the car in the demo?
How does it cope with lots of other LIDAR sensors in the same environment, or with degraded sensors?
How does it cope with running at speeds in excess of 15 or 20mph?
What's the sensor latency? For visual stuff this can be a right pain even with part-frame processing. And sensor fusion is even worse.
How deterministic are their various models?
How many entity 'intent' detections can it cope with simultaneously, and to what level does it look at how these might interact and anticipate probable events?
Does it bother with behaviour modelling of entities in the environment so it can anticipate what a specific one will do based on past observation, or does it just do a continuous rolling assessment?
How quickly can it cope with new entities appearing? Nice straight grids are one thing, but will it panic seeing a car appear as you crest a hill/bridge with a corner mixed in for extra complexity?
Or in other words yes it looks like a shiny demo, but...
Jonesy23 said:
Their video is interesting but...
How does it cope in locations that aren't already mapped? Lane detection, signs, rules & all that.
How does it cope without other vehicles providing data - a lot of what was shown was outside the detection range of the car in the demo?
How does it cope with lots of other LIDAR sensors in the same environment, or with degraded sensors?
How does it cope with running at speeds in excess of 15 or 20mph?
What's the sensor latency? For visual stuff this can be a right pain even with part-frame processing. And sensor fusion is even worse.
How deterministic are their various models?
How many entity 'intent' detections can it cope with simultaneously, and to what level does it look at how these might interact and anticipate probable events?
Does it bother with behaviour modelling of entities in the environment so it can anticipate what a specific one will do based on past observation, or does it just do a continuous rolling assessment?
How quickly can it cope with new entities appearing? Nice straight grids are one thing, but will it panic seeing a car appear as you crest a hill/bridge with a corner mixed in for extra complexity?
Or in other words yes it looks like a shiny demo, but...
Those are just questions though, just because they aren't answered in the video doesn't mean that the technology can't handle them. In other words, none of them are valid criticisms...yet. You simply don't know whether they are cause for concern or not.How does it cope in locations that aren't already mapped? Lane detection, signs, rules & all that.
How does it cope without other vehicles providing data - a lot of what was shown was outside the detection range of the car in the demo?
How does it cope with lots of other LIDAR sensors in the same environment, or with degraded sensors?
How does it cope with running at speeds in excess of 15 or 20mph?
What's the sensor latency? For visual stuff this can be a right pain even with part-frame processing. And sensor fusion is even worse.
How deterministic are their various models?
How many entity 'intent' detections can it cope with simultaneously, and to what level does it look at how these might interact and anticipate probable events?
Does it bother with behaviour modelling of entities in the environment so it can anticipate what a specific one will do based on past observation, or does it just do a continuous rolling assessment?
How quickly can it cope with new entities appearing? Nice straight grids are one thing, but will it panic seeing a car appear as you crest a hill/bridge with a corner mixed in for extra complexity?
Or in other words yes it looks like a shiny demo, but...
It's also worth pointing out that the video is two and a half years old and this technology is advancing quickly.
youngsyr said:
People will adapt their behaviour to lower the risk of being a pedestrian around autonomous cars.
You are right pedestrians will adapt their behaviour around autonomous cars.At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
PF62 said:
You are right pedestrians will adapt their behaviour around autonomous cars.
At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
And then automatically report you to the police!At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
PF62 said:
You are right pedestrians will adapt their behaviour around autonomous cars.
At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
Is that because you believe that human drivers are so incompetent that they would run you down? Or that they would run you down out of malice? Either way it sounds like a removing human drivers from the equation would be a good thing.At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
Then we just have to find a way to deal with inconsiderate s wandering around in the road.
AnotherClarkey said:
PF62 said:
You are right pedestrians will adapt their behaviour around autonomous cars.
At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
Is that because you believe that human drivers are so incompetent that they would run you down? Or that they would run you down out of malice? Either way it sounds like a removing human drivers from the equation would be a good thing.At the moment if as a pedestrian I just walk into the road in a town centre where cars are doing 20 mph or so, I stand a good chance of being run over.
With an autonomous car, how would it deal with an unexpected pedestrian in front of it? Undoubtedly programmed to stop rather than run the pedestrian down.
So knowing that I don't have to bother waiting to cross the road; just stroll across and watch all the autonomous cars slam their brakes on.
Then we just have to find a way to deal with inconsiderate s wandering around in the road.
Proof that pedestrians and road users adapt their behaviour based on their perception of risk on a road:
http://www.telegraph.co.uk/news/2017/12/17/20mph-l...
http://www.telegraph.co.uk/news/2017/12/17/20mph-l...
PositronicRay said:
chris4652009 said:
Just caught up wit the Russian adventures, wow what an interesting and education series that was.
Bravo Guy
I don't remember much about it, I think I switched off. I do recall his naïve attitude towards Russian propaganda.Bravo Guy
skinny said:
As long as he doesn't get a bunch of students from some no name uni and their bearded enthusiastic but ultimately completely lacking in resource lecturer to fit out his van with aero parts made of cardboard and straws.
at From what I've heard his Transit hasn't had any students near it this time....he said something recently on the One Show about his latest project, but was of course unable to say much at the time given the restrictions from the programme makers.
Gassing Station | TV, Film, Video Streaming & Radio | Top of Page | What's New | My Stuff