关键词: 跟踪误差预测, LSTM, 轨迹偏差, 离线补偿, 机器人

Abstract: To enhance the motion accuracy of robots,a motion error offline compensation method based on kinematic process element sampling was proposed,which effectively reducing the trajectory deviation of robots. Considering the motion process of joints,a robot tracking error prediction model was established based on LSTM-DNN neural network. In order to enable the predictive model to effectively learn the dynamic characteristics of the robot,a trajectory dataset for model training was constructed in high-dimensional space using Latin hypercube sampling technique. The actual motion trajectory of the robot during task execution was estimated based on the tracking error prediction model. This trajectory was then compared with the theoretical trajectory,and compensation was calculated using the normal plane method to implement feedforward compensation. Experimental results show that the trajectory deviation RMSE of the tested trajectory after error compensation is reduced from 1.106 5 mm to 0.296 0 mm,effectively enhancing the robot′s motion accuracy.

Key words: tracking error prediction, LSTM, trajectory deviation, offline compensation, robotics