Quadruped locomotion provides a natural setting for understanding when model-free learning can outperform model-based control design, by exploiting data patterns to bypass the difficulty of optimizing over discrete contacts and the combinatorial explosion of mode changes. We give a principled analysis of why imitation learning with quadrupeds can be inherently effective in a small data regime, based on the structure of its limit cycles, Poincar\'e return maps, and local numerical properties of neural networks. The understanding motivates a new imitation learning method that regulates the alignment between variations in a latent space and those over the output actions. Hardware experiments confirm that a few seconds of demonstration is sufficient to train various locomotion policies from scratch entirely offline with reasonable robustness.
@article{arxiv.2603.06961,
title = {Learning Quadruped Walking from Seconds of Demonstration},
author = {Ruipeng Zhang and Hongzhan Yu and Ya-Chien Chang and Chenghao Li and Henrik I. Christensen and Sicun Gao},
journal= {arXiv preprint arXiv:2603.06961},
year = {2026}
}