Self-Repairing Recurrent Ensembles for Real-Time Recovery from Distribution Shift

Distribution shift in imitation learning refers to the problem that the agent cannot plan proper actions for a state that has not been visited during the training. This problem can be largely attributed to the inherently narrow state-action coverage provided by expert demonstrations over the full environment. In this paper, we propose a robust offline to adaptive online imitation learning framework that handles the distribution shift problem in a lifelong, multi-phase scheme. In the offline learning phase, we leverage supplementary demonstrations to broaden the state-action coverage of the policy by utilizing a discriminator to effectively train the policy with supplementary demonstrations, thereby enhancing the robustness of the policy to distribution shift. In the subsequent online inference phase, our framework detects the occurrence of distribution shift and conducts self-supervised imitation learning from online experiences to adapt the policy to the online environments. Through extensive evaluations in MuJoCo environments, we demonstrate that our method exhibits better robustness to distribution shift and better adaptation performance to online environments than the baseline algorithms, which indicates superior performance of our framework against the distribution shift.

Robots deployed in the real world rarely operate under a single fixed dynamics model: wind changes, payloads vary, batteries drain, contacts shift, and hardware wears. Yet most learning-based controllers are trained once and deployed as if learning were complete. This prevents the robot from using deployment experience to further improve task performance. In this work, we propose a continual learning framework that uses real-world experience to improve robot policies under hidden and recurring dynamics. Our method learns a condition-aware dynamics model from real state-action trajectories by combining an analytical physics prior with a neural residual for unmodeled effects. A recurrent encoder infers the current hidden condition from recent interaction, and this estimate conditions both the residual model and the policy. Policy learning is performed via differentiable simulation using diverse learned dynamics sampled from the latent model. At deployment, these sampled conditions are replaced by conditions inferred online from recent real interaction, allowing the policy to recover recurring dynamics by recognition rather than residual re-fitting. Through extensive simulation studies and real-world experiments, we demonstrate that the framework improves policy performance under diverse unobserved disturbances. On real quadrotor trajectory tracking under changing wind, the policy recovers from recurring disturbances in roughly 1s, about 5x faster than online residual re-fitting. It also reduces large-disturbance hover and tracking errors by 65.7% and 53.3% over the state-of-the-art online adaptation approaches

Robotic systems deployed beyond controlled environments encounter abrupt and unobserved dynamics shifts that induce substantial transient performance loss, even when the nominal closed loop remains locally stabilizing. We formulate rapid inference-time recovery for physical AI as a constrained residual learning problem, in which an online correction added to a frozen reinforcement learning policy is regulated to minimize recovery time while remaining inside the robustness margin of the nominal controller. We specifically focus upon the regulation of residual's corrective authority relative to an already competent frozen policy. The learned residual produces bounded action-space corrections online, and a Stability Alignment Gate regulates this corrective authority through magnitude constraints, directional coherence, performance-conditioned regulation, and adaptive gain modulation. The method requires no environment resets, policy retraining, explicit system identification, or privileged shift information. Across mid-episode changes in actuation, mass, and contact conditions, it reduces recovery time relative to frozen SAC by up to 87% on Go1, 48% on Cassie, 30% on H1, and 20% on Scout in simulation, while preserving near-nominal steady-state performance. The framework is also characterized on two physical robots, a Trossen ViperX manipulator and an AgileX Scout Mini, cutting recovery time by 19.5% on hardware without modifying onboard controller or accessing shift information.