It may be easier to go around a track than drive around a crowded city, but reaching any level of mastery is going to be pushing it. The best racing AIs so far have only reached the level of an average amateur racing driver, doing hotlaps alone in one circuit.
To my best knowledge, no-one has ever put two or more AIs racing on one track. In computer games and simulations, all the AIs either cheat (ie. drive with physics disabled on a pre-set track when they aren't close to the player) or are terribly bad, causing collisions whenever something unexpected happens (e.g. overtaking attempts by the player). I have never raced against a simulator AI that was any good (and I've tried a lot of different games and I do competitive sim racing).
If and when this series will take place, it will be pushing the state of the art forward. There's nothing like it in existance.
This will be very good practice for autonomous drivers. Multi-agent collision avoidance is a very difficult research problem and needs to be solved before there can be large-scale deployment of autonomous vehicles. I presume the cars will be broadcasting their positions (and perhaps even their intentions) to others.
no, they banned it for the same reason they banned all other neat stuff like 6 wheels, ground effect aerodynamics, unrestricted turbos resulting in 1300HP, engine capacity/fuel flow, actually ALL engine development (everyone uses same engine). Cost, smaller teams had trouble keeping up with R&D.
F1 is boring right now, everyone drives same engine, same tires. Formula E is even worse - everyone drives same car.
(Some might say it's not much of a sport anyway, but it seems popular.)
Yeah, that is another argument. Not many people understand that driving a F1 car is hugely physically challenging though and drivers need to be in top shape. Us mere mortals would be exhausted after 2 laps and could not hold our head straight for even one high speed corner.
The "easy" way for AI is to record all telemetry of a super fast human lap and then tweak it with machine learning for subsequent runs. No human is going to be 100% consistent all the time.
And your objective function would have to feature penalty for "killing yourself" (well, do you want to rebuild your car after each turn? I guess not...), so I would expect the cars will be actually pretty slow due to this initially, like Google's own.
Human motorsports face the problem that all physical records were shattered and humans already crossed the edge of their abilities (i.e. driving 250mph on the ovals is the limit before drivers pass out)[1]. So yes, there could be something interesting for a normal human to see that robotic cars suddenly could push 300, 350, 400mph etc. But what would this do to human motorsports? Relegate them to 'meh' category, basically killing the whole sport as humans would look like kids from a kindergarten comparing to robots.
Risk would come in the form of a weighted measure of replacement costs + loss of driving capabilities during the race. e.g. Probability of damage given a certain move/overtake, and if damages does occur taking into account the expected loss of driving capability and or cost.
Also interesting would be to see how AI would implicitly have to take into account game theory with many agents - e.g. If the car in front of 'me' has been highly aggressive in terms of defending against overtakes - what would 'my' best course of action be? Should I signal to him that his strategy is not a deterrence to me trying to overtake?
Also there's no need to use "super fast human lap" - you can compute optimal lap, it's been done. The challenge will be to design AI to beat your opponents - then it gets interesting.
Perhaps this could end up in a place where the focus is on the cars and the technical performance, rather than human driver skill. For those of us who are more interested in engineering challenges than the celebrity driver of the day.
While i would find autonomous racing exciting as well, it is not something the general public would be excited about.
I imagine many other parts of the current, and of the AI formula, are equally only just possible now.
I see little future for this. If I wanted AI races I would just watch them on any F1 videogame from the last 5 years.
Watching a "simulation" that doesn't even simulate real motors? Where nothing can be fed back to car makers? No thanks.
(On the other hand and in a weird twist of irony I can enjoy cs matches and certain kinds of sports events where my home country has a fair chance of winning. :)
