Toward Context-Aware Exoskeleton Assistance: Integrating Computer Vision Payload Estimation with a Multi-Metric Optimization Space

Abstract:Back-support exoskeletons mitigate musculoskeletal strain, yet current systems rely on reactive sensing and lack context-aware assistance modulation. This paper presents a population-derived optimization framework and a predictive vision-based adaptive control strategy. First, we construct a multi-metric optimization space combining electromyography reduction, perceived discomfort, and user preference, revealing a non-linear relationship between payload and optimal assistance from experiments with 12 subjects. Second, we develop a computer vision-based adaptive control leveraging a fine-tuned vision transformer (DINOv2) and depth sensing to estimate payloads prior to lifting, eliminating actuation latency. Validation with an additional 12 subjects demonstrates robust payload estimation (82.41% accuracy). The proposed adaptive strategy reduces peak back muscle activation by up to 23% and improves average offloading by 8.15% over static baselines, without increasing discomfort. These results highlight the benefits of predictive perception and user-centric optimization for enhanced human-exoskeleton interaction.

Robotics (cs.RO)

arXiv:2508.06207 [cs.RO]

(or arXiv:2508.06207v3 [cs.RO] for this version)

https://doi.org/10.48550/arXiv.2508.06207 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.