FreeMusco: Motion-Free Learning of Latent Control for Morphology-Adaptive Locomotion in Musculoskeletal Characters

TL;DR

This work tackles the challenge of generating diverse, biomechanically plausible locomotion without motion-capture data. It introduces FreeMusco, a motion-free framework that jointly learns a latent control space and a muscle-actuated policy using musculoskeletal priors, a world model, and a temporally averaged loss to capture gait rhythms. The approach generalizes across Humanoid, Ostrich, and Chimanoid morphologies and supports high-level tasks such as goal navigation and path following, revealing morphology-dependent energy-efficient gaits. By randomizing target poses and energy during training and enforcing a combination of control, balancing, and biomechanical objectives, FreeMusco demonstrates that energy-aware, morphology-adaptive locomotion can emerge without demonstrations, offering a practical path for simulating movement in data-impractical characters.

Abstract

We propose FreeMusco, a motion-free framework that jointly learns latent representations and control policies for musculoskeletal characters. By leveraging the musculoskeletal model as a strong prior, our method enables energy-aware and morphology-adaptive locomotion to emerge without motion data. The framework generalizes across human, non-human, and synthetic morphologies, where distinct energy-efficient strategies naturally appear--for example, quadrupedal gaits in Chimanoid versus bipedal gaits in Humanoid. The latent space and corresponding control policy are constructed from scratch, without demonstration, and enable downstream tasks such as goal navigation and path following--representing, to our knowledge, the first motion-free method to provide such capabilities. FreeMusco learns diverse and physically plausible locomotion behaviors through model-based reinforcement learning, guided by the locomotion objective that combines control, balancing, and biomechanical terms. To better capture the periodic structure of natural gait, we introduce the temporally averaged loss formulation, which compares simulated and target states over a time window rather than on a per-frame basis. We further encourage behavioral diversity by randomizing target poses and energy levels during training, enabling locomotion to be flexibly modulated in both form and intensity at runtime. Together, these results demonstrate that versatile and adaptive locomotion control can emerge without motion capture, offering a new direction for simulating movement in characters where data collection is impractical or impossible.

Figures (12)

• Figure 1: Musculoskeletal characters used for training and simulation.
• Figure 2: Overview of the FreeMusco framework.
• Figure 3: Locomotion sequences of Humanoid (top) and Ostrich (bottom). The policies are trained under the Velocity Only configuration where the goal includes a target horizontal velocity.
• Figure 4: Locomotion sequences of Humanoid (left), Ostrich (center), and Chimanoid (right), trained under the Velocity Only configuration.
• Figure 5: Sidewise walk (top) and backward walk (bottom) sequences of Humanoid. The policies are trained under the Velocity + Direction configuration where the goal additionally includes a target facing direction.