MS-Emulator: one policy controls 700 muscles for whole-body human motion tracking
tweetsatpreet · x · 2026-10-30
Yanan Sui's group released MS-Emulator, a general whole-body controller for the human musculoskeletal system, enabling the MS-Human-700 model (700 muscle actuators) to precisely track diverse movements — a step toward foundation models of human motor behavior.
- One shared policy generalizes across diverse reference motions with coordinated whole-body control.
- Policy learning runs through imagination in a differentiable world model, refined with adversarial rewards and value-guided flow exploration.
- Reaches 4° joint-angle RMSE and 3.5 cm root-relative body RMSE in 4 hours on KIT-Locomotion, beating SuperMimic baselines.
- An interactive demo lets you switch reference motions and inspect muscle activation.
More from Embodied
- AGIBOT's OmniHand dexterous hands tackle the plan-to-execution gap in robotics — heyshrutimishra · 2026-10-30
- AGILINK's OmniHand 3 Ultra robot hand learns Suzhou embroidery in live-action demo — heyshrutimishra · 2026-10-30
- Gaussian head on flow policies: 0.02% of data, similar success, up to 2x faster inference — francoisfleuret · 2026-10-30
- NVIDIA CEO Jensen Huang: Tesla has 'the most advanced AV stack in the world' — elonmusk · 2026-10-30
- Open-source companion robot Lamp starts shipping to 20+ countries at $969 — dee_hw · 2026-10-30
- SCAMPER demo: virtual rat trots and gallops with gaits it never saw in training — tyrell_turing · 2026-10-30