Extendable Planning via Multiscale Diffusion
Chang Chen, Hany Hamed, Doojin Baek, Taegu Kang, Samyeul Noh, Yoshua Bengio, Sungjin Ahn · 2025
Read Extendable Planning via Multiscale Diffusion by Chang Chen, Hany Hamed, Doojin Baek, Taegu Kang, Samyeul Noh, Yoshua Bengio, Sungjin Ahn (2025) 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-2503-20102.txt.
Description
Chang Chen and co-authors tackle long-horizon planning for AI agents in a 2025 paper. Diffusion-based planners learn from sample paths and steer well over short stretches, but they falter when a task needs a path far longer than any seen in training. Longer training paths, in turn, make the models worse, a bind the authors name and set out to break.
Their answer comes in two phases. First, a stitching process builds longer paths step by step from short ones, round by round. Second, a planner that reasons across several time scales learns and acts on those long paths. Extra tricks fold the whole ladder of scales into a single model, so no fleet of planners is needed. Experiments showed strong gains, pointing to planning that scales without losing its footing.
Catalog record
| CREATOR | Chang Chen, Hany Hamed, Doojin Baek, Taegu Kang, Samyeul Noh, Yoshua Bengio, Sungjin Ahn |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 67 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-20102