Synergistic Functional Muscle Networks Reveal the Passivity Behavior of the Upper-Limb in Physical Human-Robot Interaction

• Published in: IEEE robotics and automation letters Vol. 9; no. 5; pp. 4679 - 4686
• Main Authors: Oliver, Suzanne, Atashzar, S. Farokh
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.05.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Biological system modeling; Biomechanics; Electromyography; Energy absorption; Estimation; Haptic interfaces; haptics; Human engineering; human- centered robotics; Human-robot interaction; Limbs; Muscles; Networks; passivity; Physical human-robot interaction (pHRI); Prediction models; Predictions; Real time; Robot sensing systems; Robots; Sensors; Telerobotics; User centered design; Wrist
• ISSN: 2377-3766; 2377-3766
• DOI: 10.1109/LRA.2024.3382496

Utilizing the intrinsic capability of the human upper limb to absorb energy during kinesthetic human-robot interaction could allow for improved haptic feedback fidelity and reduce the conservatism of control in pHRI and telerobotic systems. However, estimating this energetic signature is complex. In this letter, we quantify this capability using the biomechanical passivity index (BioPI). If estimated correctly in real-time, this can be used as a central component of a passivity-based controller during dynamic tasks. Thus, for the first time, we investigate the power of "functional muscle networks" to create a personalized computational model for real-time BioPI estimation. These muscle networks are generated based on magnitude-squared coherence between pairs of surface electromyography (sEMG) sensors to detect synergistic coupling under different co-contraction levels. Ten healthy subjects participated in the study, holding onto a robot that perturbed their wrist while an array of sixteen sEMG sensors scanned their forearm muscle activity.Muscle networks were then generated at each trial point and input to a regression to build BioPI prediction models. Results showed a strong correlation between the BioPI predicted by the proposed muscle network model and the true BioPI. High performance was maintained using only eight-sensor subnetworks and using a generalized network instead of a subject-specific network. These results allow for estimating the BioPI in real-time, which can be used in pHRI control to safely improve haptic transparency while accounting for passivity reservoirs.