Reinforcement Learning for driverless Formula Student car
This was a university project with my colleague Nicole Nobili under the surveillance of Professor Marcello Restelli. The goal of this project was to archive a self-driving agent capable of completing a Formula Student driverless trackdrive competition with techniques involving RL.
The project was realized utilizing the Formula student driverless simulator and the code in this repositories.
We decided to use Soft Actor Critic and the results where incredible. Despite the car being trained only on one track (and only for 400k steps), it was able to generalized and when faced with another track it was able to complete it effortlessly. A more in depth relation is available here (only in italian). For the full video of the car click here and here.
A translation is on going.