The paper presents a constraint-aware policy for compliant robotic manipulation, particularly in physically-constrained environments such as household operations. The policy, developed using reinforcement learning, is designed to adjust to the force imposed by the environment. The study focuses on two types of physical constraints: prismatic and revolute joints. The policy could successfully execute various compliant-manipulation operations, both in simulation and reality. This research is seen as a step towards developing a generalized household robot.
Publication date: 18 Nov 2023
Project Page: https://arxiv.org/abs/2311.11007v1
Paper: https://arxiv.org/pdf/2311.11007
