Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.19838 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908606681579520 |
|---|---|
| author | He, Shiqi Cui, Yue Ma, Xinyu Li, Yaliang Ding, Bolin Chowdhury, Mosharaf |
| author_facet | He, Shiqi Cui, Yue Ma, Xinyu Li, Yaliang Ding, Bolin Chowdhury, Mosharaf |
| contents | Autonomous web agents powered by large language models (LLMs) show strong potential for performing goal-oriented tasks such as information retrieval, report generation, and online transactions. These agents mark a key step toward practical embodied reasoning in open web environments. However, existing approaches remain limited in reasoning depth and efficiency: vanilla linear methods fail at multi-step reasoning and lack effective backtracking, while other search strategies are coarse-grained and computationally costly. We introduce Branch-and-Browse, a fine-grained web agent framework that unifies structured reasoning-acting, contextual memory, and efficient execution. It (i) employs explicit subtask management with tree-structured exploration for controllable multi-branch reasoning, (ii) bootstraps exploration through efficient web state replay with background reasoning, and (iii) leverages a page action memory to share explored actions within and across sessions. On the WebArena benchmark, Branch-and-Browse achieves a task success rate of 35.8\% and reduces execution time by up to 40.4\% relative to state-of-the-art methods. These results demonstrate that Branch-and-Browse is a reliable and efficient framework for LLM-based web agents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_19838 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Branch-and-Browse: Efficient and Controllable Web Exploration with Tree-Structured Reasoning and Action Memory He, Shiqi Cui, Yue Ma, Xinyu Li, Yaliang Ding, Bolin Chowdhury, Mosharaf Artificial Intelligence Computation and Language Machine Learning Autonomous web agents powered by large language models (LLMs) show strong potential for performing goal-oriented tasks such as information retrieval, report generation, and online transactions. These agents mark a key step toward practical embodied reasoning in open web environments. However, existing approaches remain limited in reasoning depth and efficiency: vanilla linear methods fail at multi-step reasoning and lack effective backtracking, while other search strategies are coarse-grained and computationally costly. We introduce Branch-and-Browse, a fine-grained web agent framework that unifies structured reasoning-acting, contextual memory, and efficient execution. It (i) employs explicit subtask management with tree-structured exploration for controllable multi-branch reasoning, (ii) bootstraps exploration through efficient web state replay with background reasoning, and (iii) leverages a page action memory to share explored actions within and across sessions. On the WebArena benchmark, Branch-and-Browse achieves a task success rate of 35.8\% and reduces execution time by up to 40.4\% relative to state-of-the-art methods. These results demonstrate that Branch-and-Browse is a reliable and efficient framework for LLM-based web agents. |
| title | Branch-and-Browse: Efficient and Controllable Web Exploration with Tree-Structured Reasoning and Action Memory |
| topic | Artificial Intelligence Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2510.19838 |