Attention LSTM-based gait control study for humanoid robots

Currently, humanoid robots are being gradually applied in industrial production areas, significantly improving factory productivity. However, existing gait control methods face multiple challenges: insufficient adaptation on uneven terrain, unnatural motion patterns leading to reduced operational efficiency, and poor stability in complex environments. In order to better adapt the robot to the complex environment of a smart factory, this study proposes an innovative approach that combines an attentional Long Short-Term Memory (LSTM) framework with a Proximal Policy Optimization (PPO) algorithm to enable the robot to intelligently focus on critical moments of the gait cycle to achieve natural and efficient gait control. of the gait cycle, enabling natural and efficient gait control for humanoid robots. The experimental results show that compared with the standard LSTM, our attention LSTM model improves motion stability by 38.58%, motion naturalness by 26.32%, and symmetry of the left and right legs by 96.12% (which is only 1.80% lower than that of human body data), which provides a new way of thinking about the motion control of humanoid robots in the intelligent manufacturing environment.