Groupoid-Based Internal State Representations for Reinforcement Learning with Local Symmetries
Groupoid-based RL discovers local, state-dependent symmetries during interaction, learning in a symmetry-reduced space and beating standard Q-learning efficiency.