A Framework for Online Human-Exoskeleton Activity Classification and Gait-Lab Control: Towards Automated Gait Rehabilitation

Accurate identification of human-exoskeleton activities (sitting, standing, walking, and stair ascent/descent) is crucial for effective automated gait rehabilitation. Inaccurate activity recognition leads to suboptimal exoskeleton control and hinders rehabilitation progress. Therefore, this paper presents a framework for online human-exoskeleton activity classification, demonstrating its performance on a real exoskeleton in a gait lab environment. A key contribution is an online fine-tuning method that enhances the classifier's adaptation to unseen data, addressing the challenges of dynamic human-exoskeleton interactions. This online motion classification method achieves high-accuracy identification of human-exoskeleton activity. This information can be used to proactively control the treadmill to achieve an automated gait rehabilitation process.