MPO: Boosting LLM Agents with Meta Plan Optimization
Weimin Xiong, Yifan Song, Qingxiu Dong, Bingchan Zhao, Feifan Song, Xun Wang, Sujian Li · 2025
Read MPO: Boosting LLM Agents with Meta Plan Optimization by Weimin Xiong, Yifan Song, Qingxiu Dong, Bingchan Zhao, Feifan Song, Xun Wang, Sujian Li (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-2503-02682.txt.
Description
Weimin Xiong and co-authors present Meta Plan Optimization, a way to improve planning by language model agents. Agents often invent flawed plan steps, a fault called planning hallucination. Most fixes need retraining for each new agent, which is costly.
The framework gives agents clear high-level guidance through meta plans. Those plans are general, so they travel well across tasks. Feedback from task runs then refines the plans over time. No agent retraining is needed for the gains. Tests on two planning tasks beat prior baseline methods clearly. Analysis shows better task speed and better transfer to new settings. The design works as a plug-and-play add-on for existing agents.
Catalog record
| CREATOR | Weimin Xiong, Yifan Song, Qingxiu Dong, Bingchan Zhao, Feifan Song, Xun Wang, Sujian Li |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-02682