发表机构
Centro di Ricerca E. Piaggio; Università di Pisa; ETH Zürich; NVIDIA（E.皮亚焦研究中心; 比萨大学; 苏黎世联邦理工学院; 英伟达公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对非抓取型移动操纵的复杂混合动力学与接触稀疏性问题，提出多评论家RL框架下的接触引导探索策略，经四足移动操纵器实验验证可实现现实世界中的非抓取型操纵。
AI 中文摘要
非抓取型操纵为移动操纵平台提供了移动和重新布置重型或 bulky 物体的通用技能。然而，基于模型和无模型的方法均难以应对这类任务中复杂的混合动力学以及接触的稀疏性。为应对这些挑战，我们在多评论家强化学习（RL）框架内提出了一种接触引导探索策略。专用的探索评论家由密集的寻接触奖励进行训练，该奖励引导末端执行器朝向有意义的接触点；其影响会逐步衰减，以恢复任务最优策略。我们从通用抓取算法中获取候选交互点，使探索机制能够跨不同物体几何形状进行泛化。我们在多个任务上评估了该方法，包括推箱子、搬运椅子和打开洗碗机的任务。最后，我们通过在四足移动操纵器上开展大量实验验证了搬运椅子的策略，证明了可在现实世界中部署的非抓取型操纵。
英文摘要
Non-prehensile manipulation offers versatile skills for moving and rearranging heavy or bulky objects, particularly when combined with a mobile manipulation platform. However, both model-based and model-free approaches struggle with the complex hybrid dynamics and the sparsity of the contact in these tasks. To address these challenges, we propose a contact-guided exploration strategy implemented within a Multi-Critic Reinforcement Learning (RL) framework. A dedicated exploration critic is trained with a dense contact-seeking reward that guides the end-effector toward meaningful contact points; its influence is progressively decayed to recover a task-optimal policy. We obtain candidate interaction points from a general-purpose grasping algorithm, enabling the exploration mechanism to generalise across various object geometries. We evaluate the approach on multiple tasks, including box pushing, chair transportation, and a dishwasher opening task. Finally, we validate the chair transportation policy through extensive experiments on a quadrupedal mobile manipulator, demonstrating deployable non-prehensile manipulation in the real world.
Commentsproject website: https://tolomeis.github.io/contact-guided-exp Accepted in IEEE Robotics and Automation Letter (RA-L) 8 pages, 9 figures
