LSTM-based external torque prediction for 6-DOF robot collision detection

作者：Zhang, Tie^*; Chen, Yijie; Ge, Peizhong; Zou, Yanbiao

来源：Journal of Mechanical Science and Technology, 2023, 37(9): 4847-4855.

DOI：10.1007/s12206-023-0837-3

• LSTM
• Collision detection
• Multi-parameter
• Prediction

In this article, a robot collision detection method based on the long short-term memory (LSTM) network is proposed. The method constructed an LSTM network based external torque prediction (LSTM-ETP) model can perform multi-parameter multi-step external torque prediction since the architecture of model is described in network and time dimensions. The multi-step prediction of external torque is performed sequentially at different time nodes, which can transform the single multi-step output into a continuous prediction output on the complete time scale, and achieve real-time collision detection based on the LSTM-ETP model. The model is trained using the motion data from a 6-DOF industrial robot without collision, and the proposed method was evaluated online using additional trajectories. Experiments show that the proposed method reduces the detection delay by about 30 %& SIM;40 % compared to the previous method. This proposed method reduces the requirement for robot dynamics model accuracy while ensuring satisfactory collision detection performance.