Fatigue-Aware Assistance-as-Needed Prescribed-Performance Control for Lower-Limb Exoskeleton Rehabilitation Under Uncertainty

Lower-limb exoskeletons are potential rehabilitation aids for regaining motor function after neurological and musculoskeletal disorders. However, most present controllers provide fixed or weakly adaptive help. They cannot ensure mandated tracking performance, robustness against uncertainties, or individualized aid according to the patient’s physiological condition simultaneously. This research offers a fatigue-aware Assistance-as-Needed (AAN) prescribed-performance backstepping control system, which unifies human-state estimation, adaptive assistance regulation, human-interaction torque estimation, and robust stabilization into a unified rehabilitation technique. The controller utilizes prescribed-performance control to guarantee bounded transient and steady-state tracking errors, while adaptive estimators assess patient engagement and fatigue levels to adjust robotic assistance according to the patient’s effective contribution. An adaptive human-interaction torque estimator reconstructs the voluntary joint-torque contribution of the patient, while a robust compensation term attenuates residual modeling uncertainties, unmodeled dynamics and external disturbances to preserve the closed-loop stability. A comprehensive simulation study is performed to validate the proposed framework, which includes comparative controller evaluation, robustness analysis under uncertainties and disturbances, 30-trial Monte Carlo simulations, component-wise ablation analysis, rehabilitation assessment under different patient participation levels, and fatigue accumulation–recovery experiments. The suggested controller ensures zero prescribed-performance violations with a tracking RMSE of 0.02165 rad under nominal conditions and 0.02211 rad under uncertain operating conditions. Monte Carlo simulations reveal an average root mean square error (RMSE) of 0.02120 ± 0.00208 rad with a 100% prescribed-performance success rate of all trials. The ablation study results prove the importance of the prescribed-performance transformation to guarantee limited tracking. Its absence leads to a tracking RMSE roughly twice as large, 0.04370 rad, and prescribed-performance breaches. Meanwhile, the adaptive rehabilitation trials demonstrate that the mean assistance factor drops from 0.8171 to 0.5437 with the increase in patient participation, which encourages more voluntary patient contributions while providing precise trajectory monitoring. The results reveal that the proposed controller can simultaneously achieve prescribed-performance tracking, robust disturbance attenuation, and fatigue-aware support modification in the examined simulation environment. These results demonstrate the theoretical and numerical feasibility of the framework and justify additional hardware-in-the-loop and experimental validation for customized lower-limb robotic rehabilitation.