Sensor-driven control strategies for post-stroke shoulder rehabilitation exoskeletons: A systematic review

Sensor-driven shoulder exoskeletons are emerging as promising tools for post-stroke rehabilitation, offering scalable, adaptive, and patient-specific motor assistance. This systematic review analyzes 32 studies published between 2015 and April 2025, focusing on the integration of sensor modalities and control strategies in upper-limb exoskeletons targeting the shoulder complex. The review categorizes sensor types, including electromyography (EMG), inertial measurement units (IMUs), force/torque sensors, and kinematic sensors, and evaluates their role in motion tracking, user-intent detection, and feedback regulation. Control strategies are classified into five main groups: force- and admittance-based interaction control, adaptive and assist-as-needed control, human-in-the-loop control, passive support and gravity compensation, and machine-learning-based predictive control. Motor-driven actuation was the most prevalent approach, often paired with advanced control architectures. While multimodal sensor fusion enhances system responsiveness and personalization, most implementations remain in early development or validation stages, with limited clinical deployment. Challenges include sensor drift, bioelectrical signal variability, system complexity, and the need for regulatory approval. The review concludes by highlighting future directions in AI-driven control, wearable sensing, and closed-loop neurorehabilitation. These trends point toward a new generation of intelligent, user-centered exoskeletons capable of delivering high-quality therapy in both clinical and home settings.