Research map / Large language models / Agents, code and math
R-WoM: Retrieval-augmented World Model For Computer-use Agents
Kai Mei, Jiang Guo, Shuaichen Chang, Mingwen Dong, Dongkyu Lee, Xing Niu, Jiarong Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace WebDreamer with tutorial-grounded R-WoM | WebDreamer · OSWorld (87-task subset) | success rate (%): 31.24 -> 38.54 | Improved |
| replace WebDreamer with R-WoM | WebDreamer (Qwen-2.5-VL-72B) · WebArena (113-template subset) | success rate (%): 24.50 -> 28.49 | Improved |
| add plain RAG over tutorials | Vanilla Claude-3.5-Sonnet · OSWorld (87-task subset) | success rate (%): 22.43 -> 22.19 | No clear effect |
| add plain RAG over tutorials | Vanilla Claude-3.7-Sonnet · OSWorld (87-task subset) | success rate (%): 28.47 -> 27.76 | No clear effect |
| add plain RAG over tutorials | Vanilla Qwen-2.5-VL-72B · OSWorld (87-task subset) | success rate (%): 26.36 -> 30.84 | Improved |
| add WebDreamer world model without grounding | Vanilla Claude-3.5-Sonnet · WebArena (113-template subset) | success rate (%): 27.74 -> 29.82 | Improved |
| add retrieved tutorials to full-procedure planning | Qwen-2.5-VL-72B · Full-procedure planning alignment | alignment (%): 50.0 -> 90.0 | Improved |
| add retrieved tutorials to full-procedure planning | Claude-3.7-Sonnet · Full-procedure planning alignment | alignment (%): 65.0 -> 95.0 | Improved |
| ground world model with oracle tutorials instead of retrieved (ablation) | R-WoM · OSWorld and WebArena | success rate | Improved |
| increase imagination horizon beyond 2 for ungrounded world model (ablation) | WebDreamer · OSWorld and WebArena | success rate | Hurt |
| increase horizon up to 3 for grounded world model (ablation) | R-WoM · OSWorld and WebArena | success rate | Improved |
| use R-WoM with synthesized self-play tutorials | Vanilla Claude-4.5-Sonnet · OSWorld (tutorial-scarce) | success rate (%): 45.83% -> 49.29% | Improved |
Closest papers by meaning
- Web Agents with World Models: Learning and Leveraging Environment Dynamics in Web Navigation (ICLR 2025 · similarity 0.92)
- Reinforcement World Model Learning for LLM-based Agents (NeurIPS 2026 · similarity 0.91)
- From Words to Actions: Unveiling the Theoretical Underpinnings of LLM-Driven Autonomous Systems (ICML 2024 · similarity 0.90)
- Generating Code World Models with Large Language Models Guided by Monte Carlo Tree Search (NeurIPS 2024 · similarity 0.90)
- Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents (NeurIPS 2026 · similarity 0.90)
- WorldCoder, a Model-Based LLM Agent: Building World Models by Writing Code and Interacting with the Environment (NeurIPS 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
