Abstract: Adaptive control policies for robots often require balancing generalization from large offline datasets with efficient adaptation to specific deployment conditions. In this paper, we propose ...