Pose Prediction for Six-axis Industrial Robots Based on Spatio - Temporal Dense Graph Convolutional Networks

Pose prediction can timely determine whether the posture of industrial robots is abnormal, thereby reducing the harm caused by abnormal actions of industrial robots. The pose prediction task requires modeling the complex dynamics of industrial robots, so it is necessary to effectively understand the complex spatio-temporal joint dynamics of industrial robot postures. This process requires considering the nonlinear relationships and coupling effects between joints, making this task very challenging. This paper proposes a pose prediction method suitable for six-axis industrial robots, achieving the interaction between space and time. Finally, in the decoding layer, we use a temporal convolution (TCN) to decode the spatio-temporal features of historical frame poses to predict future frame poses. Through long-term prediction MPJPE (mean per joint position error) experiments and long-term pose prediction visualization experiments on a self-made six-axis industrial robot dataset, the average MPJPE for short-term predictions is 7.1 and the average MPJPE for more difficult long-term pose predictions is 11.4. In the long-term prediction visualization experiments, we can accurately predict the long-term future poses.