Twisting Lids Off with Two Hands
This addresses a long-standing problem in robotics for tasks requiring coordinated, contact-rich manipulation with two hands, representing a novel advancement in sim-to-real transfer for bimanual systems.
The paper tackles the challenge of bimanual manipulation in robotics by developing a sim-to-real deep reinforcement learning system that enables two multi-fingered hands to twist lids off various bottle-like objects, achieving generalization to unseen objects and demonstrating dynamic, dexterous behaviors.
Manipulating objects with two multi-fingered hands has been a long-standing challenge in robotics, due to the contact-rich nature of many manipulation tasks and the complexity inherent in coordinating a high-dimensional bimanual system. In this work, we share novel insights into physical modeling, real-time perception, and reward design that enable policies trained in simulation using deep reinforcement learning (RL) to be effectively and efficiently transferred to the real world. Specifically, we consider the problem of twisting lids of various bottle-like objects with two hands, demonstrating policies with generalization capabilities across a diverse set of unseen objects as well as dynamic and dexterous behaviors. To the best of our knowledge, this is the first sim-to-real RL system that enables such capabilities on bimanual multi-fingered hands.