Multi-modal classification of MMH activities for multi-tasking control strategies in versatile exoskeletons

Versatile exoskeletons are active occupational exoskeletons that can recognize and consequently support specific user activities. This feature makes them suitable for assisting complex activities such as manual material handling (MMH), which involves sequences of static, dynamic, and cyclical tasks that can be combined. Active exoskeleton control requires a level of activity recognition that extends beyond the simple recognition of basic activities, such as standing, bending, and locomotion. When the atomic actions associated with MMH are identified, it becomes possible to develop advanced, task-specific assistive strategies. This work proposes a hierarchical classification algorithm, organized into three sequential layers and a fourth that runs in parallel, to classify the tasks performed and their atomic actions. The algorithm blends information from EMGs and IMUs, using binary multi-modal logistic regressors and rule-based classifiers. The algorithm was trained on 10 subjects and tested on an additional 8 subjects while performing multiple actions comprising the MMH task. The algorithm can identify up to 20 atomic actions that characterize locomotion, bending, and static postures, achieving 90.8% accuracy for hip flexion and extension, 91.1% for stoop and squat techniques, and 95.6% for payload hold. Furthermore, the algorithm demonstrated robustness across tasks, loads, and sequence variations while remaining user-agnostic.