Computer/gps contolled car vs human
Discussion
Interesting...
...is there a stipulation in the F1 regs that a human has to be onboard?
I guess I'm also asking if the computer would be able to set a "definitive benchmark" lap time for a given car, one that can't be bettered save for improved computer response time/software improvements..ie better than Alonso, Hamilton etc?
...is there a stipulation in the F1 regs that a human has to be onboard?
I guess I'm also asking if the computer would be able to set a "definitive benchmark" lap time for a given car, one that can't be bettered save for improved computer response time/software improvements..ie better than Alonso, Hamilton etc?
Edited by DrTre on Saturday 5th February 17:55
Not yet. They ran a driverless Audi TTS up the Pikes Peak track, it took 10 minutes longer than a driven one.
http://www.pistonheads.com/news/default.asp?storyI...
http://www.pistonheads.com/news/default.asp?storyI...
There is a long long long way to go. The amount of information humans take in with eyes, ears and "seat of the pants" is enormous, and they are nowhere near that point with automated systems. I would say maybe 4 or 5 orders of magnitude at least. For example, being able to see kids and dogs running around at the side of the road. Or driving along in wet or cold weather and all of a sudden the road noise goes quiet. Or perhaps there is a Morris Traveller / Vauxhall Nova with a driver wearing a hat. You get the idea.
The human brain is the most advanced computer in the world. A computer can crunch numbers alot faster for sure but there is alot more to driving a car than that. We process sensory information at a phenomenal rate, an automated system just "computes" where as we can think and feel.
in the future I'm sure it will be possible for an automated car to lap quicker in perfect conditions but I doubt they'l be as adaptable to the "unkmown" and changing conditions like humans are.
in the future I'm sure it will be possible for an automated car to lap quicker in perfect conditions but I doubt they'l be as adaptable to the "unkmown" and changing conditions like humans are.
Life Saab Itch said:
Like Scuffers said.
A computer doesn't have a "seat of the pants" to feel what the car is doing, it can only react. A human will pre-empt what is going to happen and adjust their inputs.
Thing is though, the 'seat of the pants' is at its most basic level the sensation of lateral accelerative force and yaw in cornering, which can both be picked up by accelerometers and gyroscopes. A car doesn't have seat of the pants feel, but I've certainly had ESP intervene before I've had a chance to.A computer doesn't have a "seat of the pants" to feel what the car is doing, it can only react. A human will pre-empt what is going to happen and adjust their inputs.
Also, how does a human know what is going to happen and adjust his/her inputs? In the case of lapping at track, I personally push harder and harder until I find the limit, and try and repeat that lap after lap until I can see its raining, or I can feel the tyres going off. If there's a bit of coolant down on the track that I can't see, then I'm probably going to crash if its at the wrong spot.
If a car can sense the rain, and it feels the reduction in lateral acceleration caused by tyre wear, then there's no reason why it can't predict how much grip there is in the same way a human does.
The big question is, how would it know what the fastest way round a corner is, which is arguably the most important part of being a quick racing driver? Humans can't afford to lap a track indefinitely to try every possible permutation of line, braking profile and weight transfer, and neither than the computer.
On that basis, the speed of the computer would be limited by the basic parameters that it uses to generate its own 'fastest line', that it then plays around with like a human. This could be done in pre-processing (i.e. the computer has an all encompassing model of the track which it CAN lap indefinitely in much faster than real time to try every permutation), but this is difficult as you would need to model the grip properties of every patch of tarmac, bump and lump, crest, kerb, etc etc.
Alternatively the computer could be given the best methods for a given type of corner, but this is in effect 'cheating' as its using human knowledge, and the computer would be limited by the knowledge of the person programming.
Finally the computer could use the telemetry of other racing drivers as a start point, and experiment with them. Again, 'cheating' as its relying on human knowledge.
Ironically if you think about it, a human racing driver uses a combination of all of these facilities when devising the best way to lap a track; he'll walk the track to spot lumps and bumps (preprocessing), he'll look at other drivers' lines and telemetry where available, and finally he'll experiment via trial and error while actually lapping the track.
Basically, it's a far more philosophical question than it seems, and on the basis that a computer is using the same sort of 'learning' methods as a human equivalent but has the advantage of perfect repeatability and persistence, then I'd hedge my bets on the computer having the edge, albeit only a fractional one.
Scuffers said:
There are two ways of doing this so far as I can see...
1) you 'program' the car to the cct, specifically, as in you get it to 'replay' a programmed lap
2) you create a car than can 'learn' the cct.
the first is relatively easy, the second is another level all together
The 2nd one is done by the VW golf that was on fith gear/top gear. And also the BMW that was on Top Gear I think does the same thing although they didn't show it. Essentially you drive it around the track slowly, so it can pick up the boundary's and knows where it's supposed to be heading. Then it drives "flat out" around the course.1) you 'program' the car to the cct, specifically, as in you get it to 'replay' a programmed lap
2) you create a car than can 'learn' the cct.
the first is relatively easy, the second is another level all together
Look up the VW Golf GTi "53 plus 1"
Edited by Munter on Monday 7th February 15:57
Munter said:
Scuffers said:
There are two ways of doing this so far as I can see...
1) you 'program' the car to the cct, specifically, as in you get it to 'replay' a programmed lap
2) you create a car than can 'learn' the cct.
the first is relatively easy, the second is another level all together
The 2nd one is done by the VW golf that was on fith gear/top gear. And also the BMW that was on Top Gear I think does the same thing although they didn't show it. Essentially you drive it around the track slowly, so it can pick up the boundary's and knows where it's supposed to be heading. Then it drives "flat out" around the course.1) you 'program' the car to the cct, specifically, as in you get it to 'replay' a programmed lap
2) you create a car than can 'learn' the cct.
the first is relatively easy, the second is another level all together
Learning a track as a static data object is fine, until something happens. Which it will. It will get colder or hotter, the humidity will change, rubber marbles will get deposited and moved around. Water and oil will leak, cement dust may well get put down. Spectators run onto the track and bits of cars fall off. How many times have we seen F1 tracks where part of it is under water and another part is totally dry? Even just adding human sight enables the brain to adapt and avoid these. Even with a huge amount of processing power, we just don't have the technology to identify these hazards and understand what they mean, and this is in a relatively "simple" closed and controlled environment of a racing circuit.
There is a lot that can be done in the area of collision prediction and avoidance, as driver assistance rather than replacement.
There is a lot that can be done in the area of collision prediction and avoidance, as driver assistance rather than replacement.
Gassing Station | General Motorsport | Top of Page | What's New | My Stuff