Adaptive Linear Path Model-Based Diffusion

Yutaka Shimizu, Masayoshi Tomizuka · 2026

Read Adaptive Linear Path Model-Based Diffusion by Yutaka Shimizu, Masayoshi Tomizuka (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-2602-02831.txt.

Description

Yutaka Shimizu and 1 co-author address a tuning problem for robot control in a 2026 paper. Diffusion models, which build actions step by step from noise, can guide robots well. Their results, however, depend heavily on scheduling choices, meaning the settings that control each step.

The proposed LP-MBD replaces the usual schedule with a straight-line path inspired by flow matching. An adaptive version uses reinforcement learning to adjust steps and noise for each task. Studies on simulations, Brax benchmarks, and a mobile robot follow. The authors report simpler tuning, strong performance, and better robustness in real time.

Catalog record

CREATORYutaka Shimizu, Masayoshi Tomizuka
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02831

Browse the full library · About the Tower of Babel