When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited

In three linesBehavior Foundation Models (BFMs) enable scalable imitation learning but fail under dynamics shifts (friction, actuation, noise). This paper formulates BFM task-inference as robust minimax optimization, enabling adaptation to worst-case dynamics perturbations without retraining. The framework outperforms standard BFM and robust offline IL baselines under dynamics shifts.

Summary generated by Claude — human-verified