deadsg / ragdolltrainer

Active Ragdoll Training with Unity ML-Agents

Ragdoll Trainer Unity Project

Active ragdoll training with Unity ML-Agents (PyTorch).

Ragdoll Agent

Based on walker example

The Robot Kyle model from the Unity assets store is used for the ragdoll.

Features:

• Default Robot Kyle rig replaced with a new rig created in blender. FBX and blend file included.

• Heuristic function inlcuded to drive the joints by user input (for development testing only).

• Added stabilizer to hips and spine. The stabilizer applies torque to help ragdoll balance.

• Added "earlyTraining" bool for initial balance/walking toward target.

• Added WallsAgent prefab for navigating around obstacles (using Ray Perception Sensor 3D).

• Added StairsAgent prefab for navigating small and large steps.

• Added curiosity to yaml to improve walls and stairs training.

Training Process (where bool in parenthesis refers to "earlyTraining" setting):

• Walking: WalkerAgent (true) -> WalkerAgent (false)
• Walls: WalkerAgent (true) -> WalkerAgent (false) -> WallsAgent (false)
• Stairs: WalkerAgent (true) -> StairsAgent (true) -> StairsAgent (false)