基于强化学习的超冗余机器人末端位置控制

End Position Control of Hyper-redundant Robot Based on Reinforcement Learning

超冗余连续体机器人因其细长结构,在受限空间具有重大应用价值。连续体长时间使用后由于摩擦、耗损等问题导致控制精度下降,影响性能。针对连续体机器人末端位置精确控制问题,对比有模型和无模型的控制方法,并搭建仿真和实体环境进行验证。实验结果表明,无模型方法适用性高,能实现较为复杂的运动规划,但是与环境交互次数过大,无法开展实体训练。基于模型的控制算法,采用集成方式表达环境状态转移模型,在少量交互下实现末端位置误差稳定收敛,具有较好的控制效果。

The hyper-redundant continuum robot has great application value in confined spaces due to its slender structure.After the continuum is used for a long time,the control accuracy is reduced due to friction and wear problems,which affects the performance.Aiming at the problem of precise control of the end position of the continuum robot,this paper compares the control methods with model and without model,and builds simulation and physical environment for verification.Experimental results show that the model-free method has high applicability and can achieve more complex motion planning,but the number of interactions with the environment is too large to carry out physical training;the model-based control algorithm uses an integrated way to express the environment state transition model,and in a small amount of interaction.It realizes stable convergence of the end position error and has a better control effect.