Skip to content

A car soccer environment inspired by Rocket League for deep reinforcement learning experiments in an adversarial self-play setting.

License

Notifications You must be signed in to change notification settings

roboserg/RoboLeague

Repository files navigation

RoboLeague

A car soccer environment inspired by Rocket League for deep reinforcement learning experiments in an adversarial self-play setting.

Table of contest

Installation

Please refer to the Unity ML Agents repository. Current version of RoboLeague uses ML-Agents Release 11

Completed features (identical to Rocket League)

  • Ball physics (ground / wall bounces, friction model, gravity, etc)
  • Car aerial control and physics
  • Acceleration and braking on the ground
  • Single jump mechanics
  • Sticky forces

Videos

Planned features

  • OpenAI gym interface integration. Unity ML already provides a wrapper
  • Car ball interaction
  • Dodges and second jump
  • Stabilizing moment near the ground
  • Lateral friction model as in the game (ratio between Vx/Vy)

Useful reading (in random order)

Licence

Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)