Muscle-driven simulation shows better match to real muscle data
Beyond Kinematics: Benchmarking Simulation Fidelity for Muscle-Driven Imitation Learning
Machine LearningRobotics
Summary
This work compares two computer simulation methods that teach machines to copy human movements using muscle models. One method focuses on detailed muscle behavior, while the other is faster to run. The researchers found that both methods copied body motions well, but the detailed muscle model better matched real muscle activity measurements. This finding helps improve designs for assistive robots that interact closely with human muscles. More work is needed to combine muscle detail with fast computing for practical use.
What this means in practice
- •For robotic assistive device designers: Use physiologically realistic muscle-driven simulations to better match human muscle activation patterns for designing assistive devices.
- •For sports biomechanics teams: Improve understanding of muscle use during motion by comparing simulation muscle activations with measured EMG data for training or injury prevention.
Authors
Ayah G. Ahmad, Claire E. Borden, Maegan Tucker
Abstract
In this work, we conduct a systematic comparison of two state-of-the-art motion-imitation reinforcement learning (MIRL) pipelines, one built on SCONE/HyFyDy and one built on MuJoCo/MyoSim. HyFyDy emphasizes physiological realism through detailed musculotendon modeling, while MuJoCo prioritizes computational efficiency and scalable policy learning. While recent work has demonstrated that both pipelines reproduce human kinematics with high fidelity, it remains unclear if they accurately capture the underlying neuromuscular behavior that produced the movement. This limitation is particularly important for robotic assistive-device design and control, where outcome measures such as muscle activation patterns and metabolic cost are often used as optimization targets. To conduct a systematic comparison, our work compares both pipelines using a common set of human motion-capture and electromyography (EMG) measurements. The results find that while both pipelines produce similar kinematics with relative accuracy, the muscle activations from HyFyDy are more aligned with the experimental EMG, as supported by the average pooled (RMSE, r) values for muscle activations from HyFyDy and MuJoCo: (0.164, 0.4) and (0.344, 0.11), respectively. While we conclude that the more advanced physiological realism of HyFyDy currently makes it more suitable for musculoskeletal modeling, both require further development to bring physiological realism to GPU-parallelizable simulation environments and advance robotic assistive device design.