基于分层深度强化学习的机械臂控制方法研究

Research on robot arm control method based on hierarchical reinforcement learning

针对现阶段机械臂控制困难、学习效率低的问题,基于分层深度强化学习,提出了通过以非线性微分方程表示的动态运动基元的协调配合来提高机械臂的学习效率和动态适应性的方法。此外,还提出了元控制器和子控制器的概念,分别用于策略的学习与目标任务的实现,从而实现层级概念和深度强化学习的结合。通过到达指定目标的仿真实验,验证了基于动态运动基元的分层强化学习方法的有效性。

In view of the difficulty in controlling the manipulator at the present stage and the low learning efficiency.The method improving the learning efficiency and dynamic adaptability of the manipulator by the coordination of dynamic motion primitives is proposed.The concepts of meta-controllers and sub-controllers for learning strategies and achieving established goals is proposed.The simulation verifies the effectiveness of the hierarchical reinforcement learning method based on dynamic motion primitives.