Adaptive Admittance Control for Lower Limb Rehabilitation Robots with Nominal Interaction Mismatch Compensation

Maintaining prescribed joint motion while allowing for compliant responses to changing interaction is challenging in passive lower limb rehabilitation. We propose interaction error-driven adaptive admittance control (IE-AAC), which subtracts the predicted nominal-interaction difference between reference and commanded states. A soft dead zone processes the corrected residual before bounded scheduling of virtual damping, stiffness, and interaction mapping gain. We evaluate the controller using a common reference reconstructed from simulated therapist demonstrations and a phenomenological interaction model. In 100 strictly paired Monte Carlo trials, IE-AAC reduced the equally weighted hip–knee–ankle mean reference deviation RMSE by 20.26%, 32.09%, and 28.23% relative to fixed admittance and two literature-derived variable admittance comparators (Zhang-2018 and Cao-2024), respectively. All four controllers in this comparison share mismatch compensation, so the reductions quantify differences in the implemented scheduling laws; a separate ablation evaluates compensation itself. The results include knee-specific trade-offs in tracking, smoothness, and interaction residual.