/deepracer

AWS DeepRacer Experimentation

MIT LicenseMIT

AWS DeepRacer Experimentation

This is an informal log of my exploration of AWS DeepRacer training. I hope this might help out someone just getting started or inspire some different approachs.

Reward Functions

Iteration Model Codename Strategy
1 "PurePursuit" A Pure Pursuit algorithm inspired approach
2 "RacingLine" Use waypoints and lane preference to encourage a racing line
3 "GetFast" Add an exponential speed component
4 "SelfMotivator" Simply encourage getting around the track in as few steps as possible
5 "Minimalist" How minimal can we go?
6 "re:Invent or Bust!" Can I create a universal model?

Hyperparameter Optimizaiton

todo