KC Neural Racing EV

12 distance rays (wall or car, closing speed), speed, sideways slide, spin, motor power, stored energy, accelerator, brake, steering, regen and deploy positions, wheelspin, contact flags, the direction of the road 20 m and 50 m ahead, and its 4 memory values from the step before. No map coordinates, no track ID.

Accelerator, brake pedal, steering wheel, regen (brake with the motor and charge the energy store) and deploy (spend the store on extra power), plus 4 memory values it hands itself for the next step (so it can keep track of time). Braking hard uses up the tires' grip, so brake before the corner, not in it.

Electric, rear-wheel drive, one gear: no clutch, no shifting, no stalling. Full pedal pulls about 0.45 g (as much as the rear tires grip) up to 72 mph, then the motor's 180 kW (about 240 hp) takes over: 0–60 mph in 6 s, 0–100 in about 11 s. No speed limiter: flat out it levels off around 160 mph, where air drag uses up all its power. Lifting off the accelerator regenerates a little (about 0.13 g). The regen control brakes harder with the motor (up to about 0.4 g in all) and charges a 1.5 MJ energy store; deploy spends it on 120 kW more power (about 12 s from full), which pays off above about 70 mph and can take it past its usual top speed while the energy lasts. Full throttle while cornering eats into the rear tires' grip, so the tail can step out.

Hitting a wall or another car faster than about 16 mph wrecks the car, and it stays where it stopped as an obstacle.

Each generation races on 2 of the 10 training tracks by default (Settings → Tracks per generation). Every team's drivers race in groups, one driver per team, and each driver's score is its average over the tracks. Then each team learns its own way: the genetic algorithm, NEAT and MAP-Elites breed new brains from the best ones; ES and CMA-ES move one center brain toward the copies that scored better; PPO adjusts its one shared brain by backpropagation. Half the teams combine methods: evolution plus a PPO brain that joins the population, several PPO brains whose learning rates evolve, MAP-Elites driven by a CMA-ES search, novelty search (rewarding drivers that drive differently), and islands of different methods that swap their best drivers. Each team's best driver races in the recorded showcase replay.

The KC Grand Prix: Union Station, Crown Center, Grand Blvd through the Crossroads, T-Mobile Center, Power & Light, Broadway past the Kauffman Center and the straight past Liberty Memorial. Never used for training. Every 10 generations the champions race it to see whether the skill carries over.

Everything runs in the browser, offline: the races in background workers (one per core), the learning on the page, and the results saved in the browser's own storage. Replays are re-simulated from the drivers' saved brains when you watch them, so each takes about 85 kB.