Quadruped Parkour Learning: Sparsely Gated Mixture of Experts with Visual Input

Michael Ziegltrum, Jianhao Jiao, Tianhu Peng, Chengxu Zhou, Dimitrios Kanoulas · 2026

Read Quadruped Parkour Learning: Sparsely Gated Mixture of Experts with Visual Input by Michael Ziegltrum, Jianhao Jiao, Tianhu Peng, Chengxu Zhou, Dimitrios Kanoulas (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-19344.txt.

Description

Ziegltrum and colleagues test sparse-gated mixture-of-experts designs, which use only a subset of weights at run time, for vision-based quadruped parkour. Robotic parkour over steps and gaps is a hard movement benchmark, but recent policies usually rely on dense perceptron nets. The authors compare expert and standard policies with matched active weights at inference. The test shows whether sparse expert routing itself lifts agile movement, rather than merely adding size.

Catalog record

CREATORMichael Ziegltrum, Jianhao Jiao, Tianhu Peng, Chengxu Zhou, Dimitrios Kanoulas
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34857
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.19344

Browse the full library · About the Tower of Babel