GaitSpan is a groundbreaking framework aimed at expanding a pretrained, basic walking policy to achieve faster locomotion. It treats walking as a seed skill, providing a reusable motor structure for balance, support, body coordination, and contact transition. The expansion includes three aspects:
-
Rhythm Generation: Modulating the frozen walking policy with multiple internal clocks and learning command-conditioned combinations of canonical actions.
-
Stride Shaping: Encouraging dynamic locomotion patterns appropriate for higher commanded speeds using a physically grounded objective inspired by spring-loaded inverted pendulum dynamics.
-
Residual Adaptation: Capturing motion details not accounted for by rhythm generation or stride shaping.
GaitSpan is the first to deliver a single command-conditioned humanoid policy that spans walking, jogging, and running-like regimes covering a continuous speed range, transferring across morphologies, and deploying zero-shot on unseen sim-to-sim and real-world terrains. Compared to baselines trained with multi-experts or imitation from humans, it learns faster and achieves stronger gait performance.
Blogger's Review: The design philosophy of GaitSpan breaks traditional locomotion learning methods by treating walking as a foundational skill, significantly enhancing humanoid robots' adaptability in various environments. Its innovative rhythm generation and stride shaping methods offer new directions and possibilities for future humanoid locomotion research.