Reward Decomposition for Robotic Explanations

High-level explanations for RL-driven robot behavior using reward decomposition and abstract action spaces.

This project studies how reward decomposition can make reinforcement learning policies more interpretable in robotic tasks. Instead of explaining low-level motor commands directly, the work uses task-relevant reward channels and abstract actions to produce higher-level explanations of agent behavior.

Related publication

  • A Closer Look at Reward Decomposition for High-Level Robotic Explanations, IEEE ICDL 2023.

TODO

  • TODO: add project image or explanation example.
  • TODO: add code link if public.