Stroke and other neurological damages often lead to severe upper-limb impairments, creating a high demand for labor-intensive and time-consuming physical rehabilitation. While robotic systems can provide repetitive and long-duration training, traditional controllers often apply fixed levels of assistance, ignoring the varying and unpredictable performance capabilities of individual patients. To maximize the effectiveness of therapy and promote neural plasticity, this project emphasizes Adaptive and the Assist-As-Needed (AAN) approaches, where a robotic device dynamically varies the provides assistance to encourage the patient's active voluntary effort and engagement.  

The foundational control schemes utilize admittance control to generate patient-specific reference trajectories from therapist-induced movements, and impedance control to allow for collaborative trajectory tracking. 

The core of the system's real-time adaptability relies on an adaptive Neural Network (NN) algorithm that adjusts the stiffness and damping parameters of the admittance model without requiring offline training. This adaptation is driven by a novel energy-based performance index that calculates the human input power, identifying whether the patient is an active energy provider or an energy dissipator within the system's dynamic energy flow. 

Experimental validations demonstrated that the proposed adaptive algorithms successfully balance the conflicting objectives of minimizing robot intervention and ensuring task completion. During active participation modes, the system exhibited highly compliant behavior; for instance, the passivity-aware adaptive controller achieved an 83% reduction in average stiffness when human effort was detected. 

The integration of these adaptive impedance controls with VR environments proved to deliver customized, responsive therapy that accommodates heterogeneous motor capabilities and unexpected spasms. Ultimately, this framework significantly reduces the need for continuous manual intervention by therapists, optimizes healthcare resource use, and provides a highly versatile, safety-aware solution for personalized neurological rehabilitation.