Another fatal Tesla crash
Discussion
Active cruise and self steer as two technologies have very different levels of maturity in my experience. I had active cruise on a 2012 merc e for 5 years and now on a 2018 bmw 5. On neither have i ever experienced an issue that was 'unsafe' and use it 95% of the time on mways regardless of traffic.
However the self (semi autonomous) steer on the BMW is great in <20mph jams on mways etc but a little sketchy at anything faster and/or less than perfect conditions. It does however demand you touch the wheel ever 10 secs at these speeds so you have to be constantly awake/monitoring it. So a little pointless beyond the traffic jam scenarios.. (where it is fab).
I think tesla are being very american and overly positive/confident in their tech. Bit like facebook and uber are with data privacy. A cultural difference i come across every day at work....
However the self (semi autonomous) steer on the BMW is great in <20mph jams on mways etc but a little sketchy at anything faster and/or less than perfect conditions. It does however demand you touch the wheel ever 10 secs at these speeds so you have to be constantly awake/monitoring it. So a little pointless beyond the traffic jam scenarios.. (where it is fab).
I think tesla are being very american and overly positive/confident in their tech. Bit like facebook and uber are with data privacy. A cultural difference i come across every day at work....
My Model X has AP2.5.
It isn't very good.
The self steering is a bit erratic - it drifts around a fair bit within its lane. It hates where two lanes merge or diverge (maybe the problem in this crash). The auto-lane change is a bit "80% of the time, it works every time".
The adaptive cruise control side of it is also a bit off. Pretty good most of the time, it does sometimes brake unexpectedly. At higher speeds, it is not as good as the ACC on my Rangie. At slow speeds (stop start traffic), it's excellent.
At the time I bought the car, I could have paid £7,000 for some autonomous driving gubbins. I thought it would be a waste of money as we are at least a decade from the laws being in place to allow it other than in test condition (IMHO). Tesla advocates will tell you that AP2.5 is enough to allow fully autonomous driving in the future. I find that hard to believe.
It isn't very good.
The self steering is a bit erratic - it drifts around a fair bit within its lane. It hates where two lanes merge or diverge (maybe the problem in this crash). The auto-lane change is a bit "80% of the time, it works every time".
The adaptive cruise control side of it is also a bit off. Pretty good most of the time, it does sometimes brake unexpectedly. At higher speeds, it is not as good as the ACC on my Rangie. At slow speeds (stop start traffic), it's excellent.
At the time I bought the car, I could have paid £7,000 for some autonomous driving gubbins. I thought it would be a waste of money as we are at least a decade from the laws being in place to allow it other than in test condition (IMHO). Tesla advocates will tell you that AP2.5 is enough to allow fully autonomous driving in the future. I find that hard to believe.
Interesting video https://electrek.co/2018/04/02/tesla-fatal-autopil...
b
hstewie said:
hstewie said: Interesting video https://electrek.co/2018/04/02/tesla-fatal-autopil...
Interesting another car does it tooWatch the blue lines in the autopilot screen as it locks onto the left solid white line then takes the car straight into the barrier
The human stopped it this time.
Edited by saaby93 on Monday 2nd April 15:26
Mr2Mike said:
I don't really understand why the Tesla has no avoidance capability for static objects, surely that should be a prerequisite before attempting to make it steer itself?
Primarily cost!"Cheap" and easy to process Doppler Radar has to reject all returns that have the same phase shift as that caused by the speed of the car itself. It uses the returns phase history as it's record, rather than just time-of-flight directly.
Basically, it sends out a pulse, which bounces off an object, and the speed differential between the car and that object results in the pulse being phase shifted. The system detects that phase shift when it receives the pulse back into it's receiver some microseconds later, and identifies objects moving at a speed different to that of the car.
So what i hear you all shout! But, every single return that bounces off something stationary in the field of view ahead of the car will have a phase shift proportional to the speed of the car. The radar has to blank those returns, otherwise it will be swamped by them.
It requires a combination of technologies and a complex fusion of there returns to actually spot a stationary object. Sounds simple, but it's one of those things that it surprisingly hard to do easily.
Having said all that, right now, the major telematics Teir1's are all madly working on the development of sensor suites that combine technologies to provide a board field return. Even then however, the "smarts" have to decide what to believe and when.
Max_Torque said:
Primarily cost!
"Cheap" and easy to process Doppler Radar has to reject all returns that have the same phase shift as that caused by the speed of the car itself. It uses the returns phase history as it's record, rather than just time-of-flight directly.
Basically, it sends out a pulse, which bounces off an object, and the speed differential between the car and that object results in the pulse being phase shifted. The system detects that phase shift when it receives the pulse back into it's receiver some microseconds later, and identifies objects moving at a speed different to that of the car.
So what i hear you all shout! But, every single return that bounces off something stationary in the field of view ahead of the car will have a phase shift proportional to the speed of the car. The radar has to blank those returns, otherwise it will be swamped by them.
It requires a combination of technologies and a complex fusion of there returns to actually spot a stationary object. Sounds simple, but it's one of those things that it surprisingly hard to do easily.
Having said all that, right now, the major telematics Teir1's are all madly working on the development of sensor suites that combine technologies to provide a board field return. Even then however, the "smarts" have to decide what to believe and when.
Plenty of other car manufacturers have systems that work reasonably well, albeit not perfectly, so the problem has already been solved. Teslas aren't exactly cheap cars either."Cheap" and easy to process Doppler Radar has to reject all returns that have the same phase shift as that caused by the speed of the car itself. It uses the returns phase history as it's record, rather than just time-of-flight directly.
Basically, it sends out a pulse, which bounces off an object, and the speed differential between the car and that object results in the pulse being phase shifted. The system detects that phase shift when it receives the pulse back into it's receiver some microseconds later, and identifies objects moving at a speed different to that of the car.
So what i hear you all shout! But, every single return that bounces off something stationary in the field of view ahead of the car will have a phase shift proportional to the speed of the car. The radar has to blank those returns, otherwise it will be swamped by them.
It requires a combination of technologies and a complex fusion of there returns to actually spot a stationary object. Sounds simple, but it's one of those things that it surprisingly hard to do easily.
Having said all that, right now, the major telematics Teir1's are all madly working on the development of sensor suites that combine technologies to provide a board field return. Even then however, the "smarts" have to decide what to believe and when.
Max_Torque said:
Mr2Mike said:
I don't really understand why the Tesla has no avoidance capability for static objects, surely that should be a prerequisite before attempting to make it steer itself?
Primarily cost!"Cheap" and easy to process Doppler Radar has to reject all returns that have the same phase shift as that caused by the speed of the car itself. It uses the returns phase history as it's record, rather than just time-of-flight directly.
Basically, it sends out a pulse, which bounces off an object, and the speed differential between the car and that object results in the pulse being phase shifted. The system detects that phase shift when it receives the pulse back into it's receiver some microseconds later, and identifies objects moving at a speed different to that of the car.
So what i hear you all shout! But, every single return that bounces off something stationary in the field of view ahead of the car will have a phase shift proportional to the speed of the car. The radar has to blank those returns, otherwise it will be swamped by them.
It requires a combination of technologies and a complex fusion of there returns to actually spot a stationary object. Sounds simple, but it's one of those things that it surprisingly hard to do easily.
Having said all that, right now, the major telematics Teir1's are all madly working on the development of sensor suites that combine technologies to provide a board field return. Even then however, the "smarts" have to decide what to believe and when.
Tesla are proud of their cars capabilities and still advertise the Autonomous Braking features
It does look like theyve given priority to the pilot rather than the collision warning system.
If it cant detect stationary objects or ones that have relatively zero i.e. crossing path, that rules out that side on truck a while back, the pedestrian with the bike 2 weeks ago, and now a solid crash barrier.
Surely other manufacturers have that sorted, but see the volvo videos posted earlier
If it cant detect stationary objects or ones that have relatively zero i.e. crossing path, that rules out that side on truck a while back, the pedestrian with the bike 2 weeks ago, and now a solid crash barrier.
Surely other manufacturers have that sorted, but see the volvo videos posted earlier
Mr2Mike said:
Max_Torque said:
Primarily cost!
"Cheap" and easy to process Doppler Radar has to reject all returns that have the same phase shift as that caused by the speed of the car itself. It uses the returns phase history as it's record, rather than just time-of-flight directly.
Basically, it sends out a pulse, which bounces off an object, and the speed differential between the car and that object results in the pulse being phase shifted. The system detects that phase shift when it receives the pulse back into it's receiver some microseconds later, and identifies objects moving at a speed different to that of the car.
So what i hear you all shout! But, every single return that bounces off something stationary in the field of view ahead of the car will have a phase shift proportional to the speed of the car. The radar has to blank those returns, otherwise it will be swamped by them.
It requires a combination of technologies and a complex fusion of there returns to actually spot a stationary object. Sounds simple, but it's one of those things that it surprisingly hard to do easily.
Having said all that, right now, the major telematics Teir1's are all madly working on the development of sensor suites that combine technologies to provide a board field return. Even then however, the "smarts" have to decide what to believe and when.
Plenty of other car manufacturers have systems that work reasonably well, albeit not perfectly, so the problem has already been solved. Teslas aren't exactly cheap cars either."Cheap" and easy to process Doppler Radar has to reject all returns that have the same phase shift as that caused by the speed of the car itself. It uses the returns phase history as it's record, rather than just time-of-flight directly.
Basically, it sends out a pulse, which bounces off an object, and the speed differential between the car and that object results in the pulse being phase shifted. The system detects that phase shift when it receives the pulse back into it's receiver some microseconds later, and identifies objects moving at a speed different to that of the car.
So what i hear you all shout! But, every single return that bounces off something stationary in the field of view ahead of the car will have a phase shift proportional to the speed of the car. The radar has to blank those returns, otherwise it will be swamped by them.
It requires a combination of technologies and a complex fusion of there returns to actually spot a stationary object. Sounds simple, but it's one of those things that it surprisingly hard to do easily.
Having said all that, right now, the major telematics Teir1's are all madly working on the development of sensor suites that combine technologies to provide a board field return. Even then however, the "smarts" have to decide what to believe and when.
AW111 said:
I wonder how many humans get suckered by that white line?
Apparently that barrier gets hit regularly, which is why the crumple section was missing.
Shouldn't the white line tun to the barrier? Dangerous looking road marking to me.
Enough I think.Apparently that barrier gets hit regularly, which is why the crumple section was missing.
Shouldn't the white line tun to the barrier? Dangerous looking road marking to me.
Watching the other simulated video from elsewhere its understandable why the visual lane tracking follows the inside line , because the actual lane isnt marked properly.
The radar should though see something stationary ahead and do something.
DJP31 said:
As far as I know all the adaptive cruise control systems work on this basis. None of them detect stationary objects in the way necessary to prevent this accident.
I'm talking about autonomous emergency braking systems, like the one Uber disconnected on their Volvo. Why isn't this functionality built into the autonomous driving system?Have you seen the video of the AEB system working on the Volvo truck? Very impressive.
Mr2Mike said:
Have you seen the video of the AEB system working on the Volvo truck? Very impressive.
These posted earlier?saaby93 said:
Volvo may have more experience
https://www.youtube.com/watch?v=aNi17YLnZpg
https://www.youtube.com/watch?v=_47utWAoupo
https://www.youtube.com/watch?v=aNi17YLnZpg
https://www.youtube.com/watch?v=_47utWAoupo
saaby93 said:
These posted earlier?
Volvo Trucks - Emergency braking.saaby93 said:
Volvo may have more experience
https://www.youtube.com/watch?v=aNi17YLnZpg
https://www.youtube.com/watch?v=_47utWAoupo
https://www.youtube.com/watch?v=aNi17YLnZpg
https://www.youtube.com/watch?v=_47utWAoupo
Mr2Mike said:
DJP31 said:
As far as I know all the adaptive cruise control systems work on this basis. None of them detect stationary objects in the way necessary to prevent this accident.
I'm talking about autonomous emergency braking systems, like the one Uber disconnected on their Volvo. Why isn't this functionality built into the autonomous driving system?Have you seen the video of the AEB system working on the Volvo truck? Very impressive.
The braking isn’t exactly autonomous, it’s automatic emergency braking to try and minimise, not prevent, an impact. I imagine the reason it’s not part of the autopilot set up is due to the reasons given by Max_Torque above - i.e. cost and complexity.
Mr2Mike said:
saaby93 said:
These posted earlier?
Volvo Trucks - Emergency braking.saaby93 said:
Volvo may have more experience
https://www.youtube.com/watch?v=aNi17YLnZpg
https://www.youtube.com/watch?v=_47utWAoupo
https://www.youtube.com/watch?v=aNi17YLnZpg
https://www.youtube.com/watch?v=_47utWAoupo
DJP31 said:
The only ones I’ve seen are the two posted above, which aren’t exactly great!
Check the link I posted. DJP31 said:
The braking isn’t exactly autonomous, it’s automatic emergency braking to try and minimise, not prevent, an impact. I imagine the reason it’s not part of the autopilot set up is due to the reasons given by Max_Torque above - i.e. cost and complexity.
That would be fine, certainly better than nothing. I don't believe such a system would add significantly to the cost of an autonomous car that already has all the sensors in place. Someone posted above that Tesla are claiming to have the feature anyway, so why doesn't it work?Edited by Mr2Mike on Tuesday 3rd April 14:29
Mr2Mike said:
whats even more amazing is how they have managed to reduce the breaking distance of trucks by so much Gassing Station | EV and Alternative Fuels | Top of Page | What's New | My Stuff