edbeeching/godot_rl_MultiLevelRobot
A RL environment called MultiLevelRobot for the Godot Game Engine. This environment was created with: https://github.com/edbeeching/godot_rl_agents Downloading the environment After installing Godot RL Agents, download the environment with: gdrl.env_from_hub -r edbeeching/godot_rl_MultiLevelRobot
1260
1## Multilevel Robot Environment2 3A simple minigame environment with multiple mini-levels for the robot to pass.4For levels that feature coins, all coins must be picked up before proceeding to the next level is possible.5The final level features some enemy robots to avoid.6 7### Observations:8- Current n_steps / reset_after,9- Position of the current level's goal in the robot's local reference,10- Position of the closest coin in the robot's local reference,11- Position of the closest enemy in the robot's local reference,12- Movement direction of the closest enemy,13- Robot velocity,14- Whether all coins for the current level have been collected (0 or 1)15 16### Action space:17```gdscript18func get_action_space() -> Dictionary:19 return {20 "movement" : {21 "size": 2,22 "action_type": "continuous"23 }24 }25```26 27### Rewards:28- Positive reward for picking up a coin,29- Negative reward (and episode end) on collision with enemy robot,30- Negative reward (and episode end) on robot falling down,31- Positive reward (and episode end) on robot reaching the end of the level by passing through the portal at the end of the level,32- Positive reward every time the robot gets closer to the portal than the previous minimum distance (min distance is restarted each episode).33 34### Game over / episode end conditions:35An episode ends if the robot falls, collides with an enemy robot or finishes a level by passing through the portal.36 37### Running inference with the pretrained onnx model:38After opening the project in Godot, open the training_scene and click on `Run Current Scene` or press `F6`39 40### Training:41The default scene (training_scene) can be used for training.