基于深度学习的冗余机械臂无碰运动规划

Obstacle Avoidance Control for a Redundant Manipulator Based on Improved RNN Method

针对冗余机械臂运动规划过程中的避碰问题,提出一种基于深度学习的避碰运动规划方法。首先,建立冗余机械臂的运动学模型,借助Gilbert Johnson Keerthi算法进行碰撞检测,确定机械臂连杆与障碍物之间的安全距离;其次,将避碰运动规划问题转换成多元函数优化问题,以循环神经网络为主体框架,用天牛须搜索机制来防止算法陷入局部最优值和保证解的多样性;最后,在仿真环境下验证文中所提算法的有效性,结果表明:该算法的预测精度比改进人工势场法平均高出10.89%、算法时间节约2.44 s,其可以帮助冗余机械臂规划一条平滑、柔顺、无碰撞的运动轨迹,具有一定的工程参考价值。

Aiming at the obstacle avoidance in motion planning for a redundant manipulator,a method based on deep learning has been proposed.Firstly,the kinematical model of the redundant manipulator has been established and the obstacle is detected by Gilbert Johnson Keerthi algorithm,which can obtain the safe distance between the link of the manipulator and the obstacles.Secondly,the collision avoidance motion planning problem is converted into a multivariate function optimization problem.The method uses the recurrent neural network as the main frame,where beetle antennae olfactory search mechanism is introduced to prevent the algorithm from falling into the local optimal value and guarantee the diversity of the solution.Finally,the effectiveness of the algorithm proposed in this paper is verified in a simulation environment.The results show that the prediction accuracy of this algorithm is 10.89%higher than that of the improved artificial potential field method on average,and the algorithm time is saved by 2.44 s.The proposed method can help the redundant manipulator plan a smooth,compliant,collision-free motion trajectory,which has a certain engineering reference value.