Simulation
The 3D simulator is a track you can restart forever. Hitting a wall costs no repairs and ten thousand laps cost no batteries, which is why almost all the learning happens here.
Platform
Simulated training, policy learning, model deployment, then testing and improving on a real track -- a loop that keeps returning to where it started.
Three layers
Learning
PPO nudges the model's parameters using the reward each lap earned. The curves and replays stay on your machine, so you can always go back and look at them.
The physical world
The same model goes onto a TopRacer Car and drives TopRacer Map 2018. That run tells you how much of what it learned actually holds.
Why the physical stage is not optional
However detailed a simulator gets, it is still an approximation. Floor friction, changing light, the way the car slows as the battery drains -- none of that shows up in full. It takes one run on a real track to tell which parts of the model's skill are real driving and which parts are just knowing how to please a simulator. That gap is itself one of the more interesting things to study.