Skip to content

About

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

Topics

Resources

Stars

312 stars

Watchers

17 watching

Forks

Latest commit

 

History

1 Commit

Folders and files

Repository files navigation

RoboLeague ML/RL Experiments

Unity ML-Agents experiments for vehicle control, hovering, and aerial ball handling. The soccer-pitch gameplay scene and stadium assets have been removed; only ML/RL experiment scenes remain.

Setup

Open with Unity 2020.1.17f1. This project uses ML-Agents Release 11; refer to its installation instructions for training setup.

Scenes

  • Assets/Scenes/SimpleMLTest.unity — basic ML test environment.
  • Assets/ML-Agents/Experiments/CalmHover/CalmHover.unity — hovering.
  • Assets/ML-Agents/Experiments/AirDribble/AirDribble.unity — aerial ball handling; default enabled build scene.
  • Assets/ML-Agents/Experiments/MaxSpeedPlusConstantRoll/MaxSpeedPlusRoll.unity — speed and constant roll.

Training configurations and trained ONNX models remain under Assets/ML-Agents. Shared vehicles, balls, controllers, and training surfaces remain available for the experiments.

Preview

Aerial ball handling experiment

Licence

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

About

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

Topics

Resources

Stars

312 stars

Watchers

17 watching

Forks

Releases

Packages

Contributors

Languages