Skip to content

dron3flyv3r/rl-agent-plugin

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

RL Agent Plugin for Godot 4

Reinforcement learning plugin for Godot 4 with in-editor training, live metrics, curriculum learning, self-play, distributed rollout workers, and .rlmodel export for inference.

Features

  • PPO, A2C, SAC, DQN, and MCTS trainers
  • Shared policy groups across multiple agents
  • Recurrent LSTM and GRU trunk layers for PPO/A2C
  • RL editor tooling: RLDash and RL Setup dock
  • Curriculum and self-play configuration resources
  • Distributed rollout collection with headless workers
  • Binary .rlmodel export/import for deployment

Requirements

  • Godot 4.6+ with C# support
  • .NET SDK 8.0+

Install in a Godot project

Latest stable release in this line: v0.2.0-beta

Option A: copy plugin folder

  1. Copy this repository into your game project as addons/rl-agent-plugin.
  2. Open Godot -> Project Settings -> Plugins.
  3. Build once (Alt+B).
  4. Enable RL Agent Plugin.
  5. if RLDash aren't showing in the editor, restart the editor.

Option B: add as submodule

git submodule add https://github.com/dron3flyv3r/rl-agent-plugin.git addons/rl-agent-plugin
git submodule update --init --recursive
git -C addons/rl-agent-plugin fetch --tags
git -C addons/rl-agent-plugin checkout v0.2.0-beta

Optional (bleeding edge): track main instead of a release tag.

Upgrade to a newer release

If you installed as a submodule, update to a newer tagged release with:

git -C addons/rl-agent-plugin fetch --tags
git -C addons/rl-agent-plugin switch <new-release-tag>
git add addons/rl-agent-plugin

Example:

git -C addons/rl-agent-plugin fetch --tags
git -C addons/rl-agent-plugin switch v0.2.0-beta
git add addons/rl-agent-plugin

To list available tags:

git -C addons/rl-agent-plugin tag --list

Quick usage

  1. Add an RLAcademy node in your training scene.
  2. Add one or more RLAgent2D or RLAgent3D nodes.
  3. Assign RLTrainingConfig, RLRunConfig, and per-agent PolicyGroupConfig resources.
  4. Start training from toolbar (Start Training) or RL Setup dock.
  5. Monitor metrics in RLDash.
  6. Export checkpoints to .rlmodel.

Documentation

  • Docs index: docs/README.md
  • Feature catalog: docs/features.md
  • Get started: docs/get-started.md
  • Configuration reference: docs/configuration.md
  • Architecture: docs/architecture.md
  • Algorithms: docs/algorithms.md
  • Native build + validation: docs/build-native.md
  • Tuning: docs/tuning.md

Companion demo project

If you want ready-to-run demo scenes, use:

That repo contains training environments and example game scenes that consume this plugin.

License

MIT

About

An RL Agent system with in-editor training and control. Single-/Multi-agent support.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages