Event
MyoAcross: A Cross-Individual General Control Strategy for Musculoskeletal Systems
1 reports · 1 sources
Bozhao Wang and other authors released the MyoAcross framework, which combines individualized musculoskeletal model construction, morphology-conditioned policy learning, and virtual model augmentation to enable muscle-driven reference motion tracking without per-individual fine-tuning. The authors report a completion rate of 99.24% on 31 held-out individual models, with tracking error reduced by 22.60% compared with an unconditioned multilayer perceptron. Policy reuse was also extended to 22 held-out individual models constructed from single-image shape estimation and measured height and weight; after asymmetric augmentation, it also applies to models with unilateral lower-limb shortening. Comparison with measurements from another 120 people showed that vertical ground reaction force waveforms were consistent, while the consistency between simulated muscle activation and electromyography varied by channel.
Report timeline
- MyoAcross: A Generalist Muscle-Driven Control Policy for Musculoskeletal Models Across Individuals
The study proposes the MyoAcross framework, which combines individual musculoskeletal model construction, morphology-conditioned policy learning, and virtual model augmentation to achieve muscle-driven reference-motion tracking without per-individual fine-tuning. Across 31 held-out individual models, it achieved a 99.24% completion rate and reduced tracking error by 22.60% relative to an unconditioned multilayer perceptron. Policy reuse also extended to 22 held-out individual models constructed from single-image shape estimates, height, and body mass, as well as to a model with unilateral lower-limb shortening after asymmetry augmentation.
arXiv cs.RO ·