Computer Vision-based Adaptive Control for Back Exoskeleton Performance Optimization

2025-08-08

AI Summary

To address the challenges of imprecise support strategy adjustment, poor adaptability to dynamic loads, and insufficient personalization in wearable exoskeletons, this study proposes a real-time adaptive control framework grounded in computer vision. The method estimates external load via vision-based perception, integrates electromyographic (EMG) monitoring, user comfort assessment, and preference modeling to formulate a multi-objective optimization space for closed-loop, dynamic support force regulation. Crucially, it introduces the first integration of visual perception with contextual understanding of exoskeleton–user–environment interactions, substantially reducing system latency and enhancing environmental adaptability. Experimental results demonstrate a load estimation accuracy exceeding 80%, a 23% reduction in peak erector spinae muscle activation compared to static control, significant alleviation of subjective discomfort, and consistent preservation of individualized support preferences.

Technology Category

• Intelligent Robots: State Estimation
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Computer Vision: Video Understanding & Activity Analysis

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Back exoskeletons can reduce musculoskeletal strain, but their effectiveness depends on support modulation and adaptive control. This study addresses two challenges: defining optimal support strategies and developing adaptive control based on payload estimation. We introduce an optimization space based on muscle activity reduction, perceived discomfort, and user preference, constructing functions to identify optimal strategies. Experiments with 12 subjects revealed optimal operating regions, highlighting the need for dynamic modulation. Based on these insights, we developed a vision-based adaptive control pipeline that estimates payloads in real-time by enhancing exoskeleton contextual understanding, minimising latency and enabling support adaptation within the defined optimisation space. Validation with 12 more subjects showed over 80% accuracy and improvements across all metrics. Compared to static control, adaptive modulation reduced peak back muscle activation by up to 23% while preserving user preference and minimising discomfort. These findings validate the proposed framework and highlight the potential of intelligent, context-aware control in industrial exoskeletons.

Problem

• Optimizing back exoskeleton support strategies for muscle activity reduction
• Developing real-time adaptive control based on payload estimation
• Enhancing exoskeleton performance while minimizing discomfort and latency

Innovation

• Vision-based real-time payload estimation
• Dynamic support modulation optimization
• Context-aware adaptive control pipeline