Upper Limb Rehabilitation Robot Based on Long Short-Term Memory Network Enhanced Sliding Mode Control
Against the background of an aging population, addressing post-stroke upper limb hemiplegia rehabilitation needs and the high costs, inconsistent efficacy of traditional training, this paper proposes a rehabilitation robot design integrating LSTM and sliding mode control (SMC). Upper limb rehabilitation robots, as nonlinear systems, face external disturbances; traditional SMC, though robust, requires high switching gains, leading to chattering that impairs application effects. To solve this, the study first defines upper limb joint motion characteristics and safe control ranges as the kinematic basis, then introduces LSTM to capture disturbance temporal features and realize predictive modeling of time-varying disturbances. LSTM-predicted disturbances are used as feedforward compensation in the SMC law, combined with saturation functions to reduce switching gains and suppress chattering. The solution retains SMC’s robustness, improves rehabilitation precision and comfort, meets upper limb motion requirements, and provides a feasible technical approach for robot-assisted rehabilitation of post-stroke upper limb hemiplegia.