Dynamics Are Learned, Not Told: Semi-Supervised Discovery of Latent Dynamics Geometries For Zero-Shot Policy Adaptation
2026-06-01 • Robotics
Robotics
AI summaryⓘ
The authors address how robots can keep learning even when the physical environment changes unpredictably. Instead of programming the robot to recognize specific physical changes, they teach it to understand how those changes affect outcomes of its actions. They use a special learning method to create smooth and meaningful internal representations that work well even when the changes are complex or unknown. Their approach showed better performance and stability in robotic control tasks compared to traditional methods. This suggests focusing on the structure of the robot’s experience helps it adapt more reliably.
reinforcement learningroboticsdynamics shiftslatent spacecontrastive learningMuJoCotrajectory encoderlatent topologypolicy adaptationdomain shift
Authors
Zhiming Xu, Weitao Zhou, Xianghui Pan, Nanshan Deng, Chengju Liu, Qijun Chen, Chenpeng Yao
Abstract
Real-world dynamics shifts pose a critical challenge for reinforcement learning in robotics, as policies tightly coupled to nominal environments often fail catastrophically when physical conditions change. Most existing methods rely on encoding explicitly identified physical parameters into a latent context, a parameter-centric paradigm that depends on pre-specified axes of variation and becomes brittle under unmodeled or compound dynamics changes. We revisit dynamics adaptation from an outcome-centric perspective: rather than telling policies what the dynamics are, we enable them to learn how dynamics affect interaction outcomes. Theoretically, this is grounded in a monotonic relationship between target-domain regret and the Lipschitz constant of a trajectory dynamics encoder. Practically, this constant can be upper-bounded through contrastive learning, yielding a smooth, task-relevant latent topology without privileged dynamics information. On MuJoCo benchmarks, our method consistently outperforms parameter-centric baselines under severe dynamics shifts, including unmodeled and time-varying parameters, while also improving in-distribution stability and latent interpretability. Overall, these results validate that controlling latent geometry is a principled mechanism for robust adaptation.