Joint Trajectory Prediction of Multi-Linkage Robot Based on Graph Convolutional Network

The working accuracy of multi-linkage robot is seriously affected by the errors at the joints caused by the uncertainty factors such as vibration, wear, deformation, and manufacturing clearance. In order to improve the working accuracy, the joint motion prediction including these errors is researched, which can realize the follow-up errors pre-compensation. According to the spatial correlation and time dependence between the joints of the robot, the joints can be represented as a graph. This work proposes a joint trajectory prediction method based on graph convolutional neural network (GCN) and gated recurrent unit (GRU). A real experimental dataset is built to verify the effectiveness, including the uncertain errors. The method is validated by means of Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Accuracy, and R-square. Experimental results demonstrate that the method obtains the highest performance in the joints trajectories prediction, compared with Historical Average (HA), Autoregressive Integrated Moving Average (ARIMA), and Support Vector Regression (SVR). In the case of Accuracy, the accuracy of the proposed method is 91.549%, which is 33.40%, 7.32% and 3.19% higher than that of HA, ARIMA and SVR, respectively. The method can effectively predict the joints trajectories of multi-linkage robot with uncertain error at the joint, and provide theoretical support for further error compensation, obstacle prediction, and obstacle avoidance control of robot.