GaitSpan: Growing Humanoid Locomotion from Walking to Running

2026-07-13 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

The GaitSpan framework, published on 2026-07-13, introduces a novel approach to humanoid locomotion by expanding a pretrained basic walking policy into faster gaits like jogging and running. Unlike traditional methods that often relearn locomotion from scratch or rely on prescribed schedules, GaitSpan treats walking as a reusable "seed skill" for fundamental motor structures such as balance, support, body coordination, and contact transition. This expansion involves three key aspects: rhythm generation, which modulates the frozen walking policy with internal clocks; stride shaping, which rewards dynamic locomotion patterns using spring-loaded inverted pendulum dynamics; and residual adaptation, capturing fine motion details. GaitSpan is the first to deliver a single command-conditioned humanoid policy capable of continuous speed ranges, transferring across morphologies, and deploying zero-shot on unseen simulated and real-world terrains, outperforming multi-expert or imitation-based baselines in learning speed and gait performance.

Key takeaway

For Robotics Engineers developing versatile humanoid locomotion, GaitSpan suggests you can significantly reduce training complexity by building upon existing walking policies rather than starting from scratch. This approach allows your systems to achieve continuous speed control from walking to running, transfer across morphologies, and deploy zero-shot on new terrains. Consider integrating seed skill expansion to enhance adaptability and performance in your next-generation humanoid designs.

Key insights

Expanding a pretrained walking policy into faster gaits via a seed skill approach offers continuous, adaptable humanoid locomotion.

Principles

• Walking serves as a foundational seed skill.
• Modulate existing policies with internal clocks.
• Reward dynamic patterns using physics-inspired objectives.

Method

GaitSpan expands a frozen walking policy through rhythm generation, stride shaping based on spring-loaded inverted pendulum dynamics, and residual adaptation.

In practice

• Deploy zero-shot on unseen sim-to-sim and real-world terrains.
• Achieve continuous speed control from walking to running.
• Transfer policies across different humanoid morphologies.

Topics

• Humanoid Locomotion
• GaitSpan Framework
• Reinforcement Learning
• Policy Transfer
• Zero-shot Deployment
• SLIP Dynamics

Best for: Research Scientist, AI Scientist, Robotics Engineer

Related on AIssential

• Mind Your Steps: A General Learning Framework for Accurate Humanoid Foothold Tracking — The framework enables terrain-agnostic 3D humanoid foothold tracking by dynamically sampling goals and using a robust target representation for real-world transfer.
• Scaling Behavior Foundation Model for Humanoid Robots — Coordinating learning paradigm, behavioral data, and model architecture significantly scales Behavior Foundation Models for humanoid control.
• CTS-MoE: Implicit Terrain Adaptation via Mixture-of-Experts for Perceptive Locomotion — CTS-MoE uses perception-gated Mixture-of-Experts and multi-critics for adaptive legged locomotion on discontinuous terrains.
• LoComposition: Terrain-Adaptive Energy-Efficient Quadruped Locomotion without Gait Priors — Decoupling quadruped locomotion control into distinct mechanisms enables efficient, terrain-adaptive behavior without explicit gait priors.
• Where Should Action Generation Begin? A Learnable Source Prior for Generative Robot Policies — A learnable, proprioception-conditioned prior significantly improves generative robot policy performance and efficiency.
• TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes — Dynamic robot skills are mastered by focusing on crucial interaction windows and practicing distinct actions.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.