CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

This research proposes CAG-WAM, a causal action-gain world action model for robot manipulation. It addresses the limitations of traditional world action models by incorporating structural causal models to evaluate how both action-induced and environment-driven changes affect future outcomes. The method defines causal action gain (CAG) to characterize future changes and integrates it into policy optimization for improved robot decision-making.