KINESIS: Motion Imitation for Human Musculoskeletal Locomotion
Merkourios Simos, Alberto Silvio Chiappa, Alexander Mathis
AI summary
Problem
Current physics-based humanoid control relies on unrealistic torque controllers that ignore biomechanical joint constraints and nonlinear musculotendon dynamics, while existing muscle-driven locomotion models lack scalability and quantitative validation against human physiological data.
Approach
KINESIS uses a model-free reinforcement learning framework trained on curated motion capture data, employing hard negative mining and a mixture-of-experts architecture to learn muscle activation policies that track reference motions and generalize to high-level control tasks.
Key results
- Achieves high-fidelity motion imitation on unseen locomotion trajectories
- Generates muscle activation patterns that quantitatively correlate with human EMG recordings
- Scales seamlessly to a 290-muscle musculoskeletal model without hyperparameter tuning
- Enables zero-shot text-to-control, target reaching, and penalty kick tasks
Why it matters
Provides a scalable, physiologically plausible foundation for neuroscience research and robust, muscle-driven control in bio-inspired robotics.
Abstract
How do humans move? Advances in reinforcement learning (RL) have produced impressive results in capturing human motion using physics-based humanoid control. However, torque-controlled humanoids fail to model key aspects of human motor control such as biomechanical joint constraints and non- linear, overactuated musculotendon control. We present KI- NESIS, a model-free motion imitation framework that tackles these challenges. KINESIS is trained on 1.8 hours of locomotion data and achieves strong motion imitation performance on unseen trajectories. Through a negative mining approach, KINESIS learns robust locomotion priors that we leverage to deploy the policy on several downstream tasks such as text-to-control, target point reaching, and football penalty kicks. Importantly, KINESIS learns to generate muscle activity patterns that correlate well with human EMG activity. We show that these results scale seamlessly across biomechanical model complexity, demonstrating control of up to 290 mus- cles. Overall, the physiological plausibility makes KINESIS a promising model for tackling challenging problems in human motor control. Code, videos, and benchmarks are available at https://github.com/amathislab/Kinesis.