Contact-Guided Exploration for Non-Prehensile Locomanipulation with Multi-Critic RL
Robotics
Summary
The authors developed a new way for robots to move big or heavy objects without grabbing them, using a special learning method that focuses on how the robot's hand touches the object. They train the robot to find good places to push or slide objects by encouraging it to make meaningful contact first, then gradually let it learn the best way to do the task. Their method works on different shaped objects and tasks like pushing boxes or moving chairs. They tested their approach on a robot with four legs and a manipulator arm, showing it can handle real-world non-grabbing manipulation tasks.
Authors
Simone Tolomei, Mayank Mittal, Franco Angelini, Manolo Garabini, Paolo Salaris, Marco Hutter
Abstract
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.