Then you'll love stupid car sumo simulation trained with Unity MLAgents and Proximal Policy Optimization
This is an experiment I bodged together where I tried to train two competitive agents using the same Neural Network with a lot of trial and error to try and get a working policy.
