Deep self-attention reinforcement learning adaptive gait planning and control for lower limb rehabilitation exoskeletons robot

Summary Aiming at the gait planning challenges of lower limb rehabilitation exoskeletons for adapting to user-specific gait characteristics, this study proposes an adaptive gait planning framework based on an improved deep deterministic policy gradient (DDPG) algorithm. We integrate self-attention mechanism into DDPG network, optimize attention parameters via cosine annealing SGD, and adopt Bayesian optimization to adjust hyperparameters adaptively. Simulation and experimental results show that the proposed algorithm outperforms traditional DDPG in gait trajectory generation, tracking accuracy and stability under variable gait parameters and unsteady human-robot interaction. This method improves the adaptive planning capability of exoskeletons, and provides a feasible scheme for personalized lower limb rehabilitation training.