Enforcing the consensus between Trajectory Optimization and Policy Learning for precise robot control - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année :

Enforcing the consensus between Trajectory Optimization and Policy Learning for precise robot control

(1) , (1, 2) , (1) , (1) , (1)
1
2

Résumé

Reinforcement learning (RL) and trajectory optimization (TO) present strong complementary advantages. On one hand, RL approaches are able to learn global control policies directly from data, but generally require large sample sizes to properly converge towards feasible policies. On the other hand, TO methods are able to exploit gradient-based information extracted from simulators to quickly converge towards a locally optimal control trajectory which is only valid within the vicinity of the solution. Over the past decade, several approaches have aimed to adequately combine the two classes of methods in order to obtain the best of both worlds. Following on from this line of research, we propose several improvements on top of these approaches to learn global control policies quicker, notably by leveraging sensitivity information stemming from TO methods via Sobolev learning, and augmented Lagrangian techniques to enforce the consensus between TO and policy learning. We evaluate the benefits of these improvements on various classical tasks in robotics through comparison with existing approaches in the literature.
Fichier principal
Vignette du fichier
lelidec2022policy.pdf (923.55 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03780392 , version 1 (19-09-2022)

Identifiants

  • HAL Id : hal-03780392 , version 1

Citer

Quentin Le Lidec, Wilson Jallet, Ivan Laptev, Cordelia Schmid, Justin Carpentier. Enforcing the consensus between Trajectory Optimization and Policy Learning for precise robot control. International Conference on Robotics and Automation, May 2023, London, United Kingdom. ⟨hal-03780392v1⟩
51 Consultations
10 Téléchargements

Partager

Gmail Facebook Twitter LinkedIn More