Dynamics Are Learned, Not Told: Semi-Supervised Discovery of Latent Dynamics Geometries For Zero-Shot Policy Adaptation
Rattachement africain : cn. Niveau de preuve : code pays fourni par la source.
Le résumé fourni par la source
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.
Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.
Le contrôle bibliographique ouvert
Où se fait cette recherche
-
Tongji University Department of Electronics and Information Engineering pays non établi dans la noticeUniversité ou école supérieure
-
Tsinghua University pays non établi dans la noticeUniversité ou école supérieure
-
School of Computer Science and Engineering pays non établi dans la noticeUniversité ou école supérieure
-
School of Vehicle and Mobility pays non établi dans la noticeUniversité ou école supérieure
-
Simple AI pays non établi dans la noticeInstitution
-
Rimbot pays non établi dans la noticeInstitution
Department of Electronics and Information Engineering — Tongji University, Tsinghua University et School of Computer Science and Engineering, avec 3 autres affiliations.
Une affiliation ne permet pas de déduire la nationalité d’un auteur.