Adaptive Admittance Control of Robot Joints Based on LSTM Network Optimization

Compliant motion control of the collaborative robot is crucial for human-robot interaction. However, traditional admittance control relies on fixed parameters, which is difficult to cope with the influence of external disturbances and dynamic human-computer interaction forces, resulting in a decrease in tracking performance. In view of the above problems, this paper proposes a joint adaptive admittance control method based on LSTM network optimization. By constructing the parameter adaptive law under Lyapunov stability, the admittance parameters are adjusted online to enhance the robustness of the system. The long-term and short-term memory network ( LSTM ) is further used to model and predict the historical interaction torque and state sequence, compensate the system hysteresis and optimize the dynamic response of the parameters. The simulation results show that the proposed method significantly improves the trajectory tracking accuracy, suppresses the parameter fluctuation and response hysteresis, and improves the safety and adaptability of human-computer interaction under the premise of maintaining low stiffness and damping levels.