A review of intention recognition and information perception for upper-limb exoskeletons

Abstract

Upper-limb exoskeletons have gained increasing attention in neurorehabilitation, industrial assistance, and human performance augmentation. A fundamental challenge in achieving safe, intuitive, and effective human–robot interaction lies in accurately and reliably recognizing the user’s motion intention in real time. This paper presents a comprehensive review of recent advances in intention recognition and information perception technologies for upper-limb exoskeletons. The reviewed methods are systematically categorized according to sensing modalities and algorithmic strategies. From a sensing perspective, both biological signals (e.g., EMG, EEG, and MMG) and non-biological signals (e.g., IMUs, force/torque sensors, encoders, and vision-based systems) are evaluated based on responsiveness, robustness, and practicality for real-world deployment. On the algorithmic level, intention recognition approaches are classified into model-based and data-driven methods, covering discrete classification and continuous regression paradigms. In particular, the growing role of deep learning techniques in multimodal fusion, joint torque prediction, and real-time activity recognition is highlighted, along with their advantages in adaptability and performance. Meanwhile, the integration of learning-based methods with biomechanical and dynamic models is discussed as a promising hybrid strategy to enhance interpretability, stability, and safety. Finally, current challenges and future research trends are outlined, emphasizing multimodal sensor fusion, hybrid model–data-driven frameworks, and personalized real-time intention decoding. This review aims to provide a structured reference for the design and development of intelligent upper-limb exoskeleton control systems.