GIMloco: Generic Internal Model-Based Locomotion for Quadruped Robots
Zhonghuai Yan, Quan Quan
AI summary
Problem
Robust quadruped locomotion using only proprioception requires encoding long observation histories, but regression methods suffer from linear dimension growth and TCNs incur heavy computational overhead.
Approach
GIMloco maps proprioceptive observation histories into a compact, stable internal model space using a predesigned first-order integral system, which then drives state estimation, latent learning, and policy networks.
Key results
- Outperforms baselines in velocity tracking, overshoot, and response speed
- Navigates more complex terrains with improved training stability
- Reduces training time by two orders of magnitude compared to TCNs
- Validates effective sim-to-real transfer on a physical Unitree Go2 robot
Why it matters
Enables agile, sensor-efficient legged robot control with significantly faster training, advancing practical deployment in unstructured environments.
Abstract
A central challenge in robust quadruped locomo- tion, which relies solely on proprioceptive information, is how to effectively encode the history of observations. While current methods, such as regression, struggle with high-dimensional multi-time-step histories, and Temporal Convolutional Net- works (TCNs) incur computational overhead, we propose a more efficient and theoretically grounded alternative. Inspired by the Generic Internal Model (GIM) from control theory, we introduce GIMloco, which maps the history of proprioceptive observations into a compact and stable internal model space through a predesigned first-order integral system with stability and orthogonality guarantees. This encoded representation drives three downstream tasks: state estimation, latent variable learning, and control policy learning. Our experiments show that GIMloco outperforms strong baselines in velocity tracking, system overshoot, response speed. Furthermore, it can navigate more complex terrains while also demonstrating better training stability across random seeds. Crucially, our method reduces training time by two orders of magnitude compared to TCN- based approaches. Our work presents GIMloco as a robust and computationally efficient framework for locomotion based on proprioceptive information.