AI Race Car Avoids a Collision and Makes the Perfect Overtake!

[@madhurbehl]

Here’s how the Cavalier Autonomous Racing car handled this overtaking scenario: - Coming into the overtaking zone, our perception system locks on to the opponent but has a slight error in predicting their trajectory as they exit turn 4. - No autonomous vehicle is immune to the mis-prediction of another agent’s behavior, so the motion planner evaluates (based on the incorrect prediction) that an outside pass is still possible if we accelerate now. - The car accelerates to 90 mph, closing the gap for an outside pass. - The planner quickly re-evaluates as the opponent veers towards the outside, making the pass unsafe. It slows down from 91 mph to 65 mph to avoid a certain collision. - It identifies that an inside overtake is still feasible, with the opponent staying on the outside line - and then executes the inside pass safely. - During the next lap, in a different part of the track - at the exit of turn 2, the planner applies what it learned from the turn 4 encounter; anticipating the opponent’s outside line and opting for a safe inside pass.

[@dougmintz2943]

So could you have a physical driver do about 20 laps in a simulator for that car and track and then use that person’s behavior to tune the PIDs or other characteristics in the AI car? That way there is a connection to a person and it’s a synthesis vice a replacement (or at least that’s what it’s perceived to be). This would “train” the algorithm based on a person.

[@wtalkie]

When did that become a thing??

[@myyoutubeantispamaccountformys]

why? why would people waste time on this, what is the point of AI racing? I can just as well then download a racing video game and watch that. Focus on making AI driving better on the road, these are stupid expensive vehicles, stupid expensive to run, geez this is braindead stuff this

[@Wicaeed1]

Use your imagination, think of where this will be in like, 5 or 10 years

[@felipeyoshino6951]

That's because you don't look for exploring all of the benefits of using a real laboratory to experiment an AI model to its limits, instead of simulating it inside a virtual environment. It is not a matter of playing or entertainment as the guys seems to transmit in their talking, but a documentation of improvements and discoveries about the AI technology, electronic sensing, inumerous processing topologies, and so on, already used in other industrial applications. Way beyound a demo race in a videogame waiting for you to press a button.

[@amanwholikescars]

Same reason as people not interested in pursuing working in racing do FSAE. It gives Uni students something to apply what they are studying (ai driving) to something fun to do (race cars)

[@BigKandRtv]

I assume that motor sports is not the final point of implementation for this computer engineering. But it is a good platform to use for development: Existing vehicle market, enclosed (safe) testing places (tracks), built-in fan base to attract sponsors (financing) even if 2/3 of the fan base rejects AI racing...

[@magalicacciari4467]

broaden your perceptions and avoid calling research stupid.