Motion Planning for Human-Robot Collaboration based on Reinforcement Learning

摘要

Robots are good at performing repetitive tasks in modern manufacturing industries. However, robot motions are mostly planned and preprogramed with a notable lack of adaptivity to task changes. Even for slightly changed tasks, the whole system must be reprogrammed by robotics experts. Therefore, it is highly desirable to have a flexible motion planning method, with which robots can adapt to certain task changes in unstructured environments, such as production systems or warehouses, with little or no intervention needed from non-expert personnel. In this paper, we propose a user-guided motion planning algorithm in combination with reinforcement learning (RL) method to enable robots to automatically generate their motion plans for new tasks by learning from a few common tasks saved as primitive actions by kinesthetic human demonstrations. Features of these primitive actions are captured through screw transformation of the end-effector during the task. A mapping method is developed to convert features of primitive actions to new task segments and further used to construct the reward function in RL. A Q-learning algorithm is applied to train the motion planning policy, following which an adaptive motion plan for the new task can be generated or a request for additional primitive actions will be returned if current primitive actions are insufficient for satisfying new task constraints. Multiple experiments conducted on common tasks and scenarios demonstrate that the proposed RL-based motion planning method is effective.