SynWorld: Virtual Scenario Synthesis for Agentic Action Knowledge Refinement

2025/04/04 by Runnan Fang, Xiaobin Wang, Fang, Runnan +21 · 1 voice · 9 citations
Computer Science · Engineering · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Multi-Agent Systems and Negotiation #Multiagent Systems (cs.MA) #Robotics and Automated Systems #Semantic Web and Ontologies #cs.AI #cs.CL #cs.CV #cs.LG #cs.MA

paper · pdf · doi:10.48550/arxiv.2504.03561

openalex publication_date 2025/04/04 · arxiv published 2025/04/04 · arxiv updated 2025/06/01 · openalex created_date 2025/10/15 · openalex updated_date 2026/07/28

Abstract

In the interaction between agents and their environments, agents expand their capabilities by planning and executing actions. However, LLM-based agents face substantial challenges when deployed in novel environments or required to navigate unconventional action spaces. To empower agents to autonomously explore environments, optimize workflows, and enhance their understanding of actions, we propose SynWorld, a framework that allows agents to synthesize possible scenarios with multi-step action invocation within the action space and perform Monte Carlo Tree Search (MCTS) exploration to effectively refine their action knowledge in the current environment. Our experiments demonstrate that SynWorld is an effective and general approach to learning action knowledge in new environments. Code is available at https://github.com/zjunlp/SynWorld.

Citations

• Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems
• Towards Trustworthy GUI Agents: A Survey
• A Survey of WebAgents: Towards Next-Generation AI Agents for Web Automation with Large Foundation Models
• A Survey of Efficient Reasoning for Large Reasoning Models: Language, Multimodality, and Beyond
• OmniThink: Expanding Knowledge Boundaries in Machine Writing through Thinking
• From Exploration to Mastery: Enabling LLMs to Master Tools via Self-Driven Interactions
• Retrieval Augmented Generation (RAG) and Beyond: A Comprehensive Survey on How to Make your LLMs use External Data More Wisely
• Symbolic Learning Enables Self-Evolving Agents
• LongRAG: Enhancing Retrieval-Augmented Generation with Long-context LLMs
• An LLM-Tool Compiler for Fused Parallel Function Calling
• The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling: A Survey
• Automating REST API Postman Test Cases Using LLM
• Towards Practical Tool Usage for Continually Learning LLMs
• A Survey on Large Language Model-Based Game Agents
• RepairAgent: An Autonomous, LLM-Based Agent for Program Repair
• AnyTool: Self-Reflective, Hierarchical Agents for Large-Scale API Calls
• Understanding the planning of LLM agents: A survey
• Small LLMs Are Weak Tool Learners: A Multi-LLM Agent
• EASYTOOL: Enhancing LLM-based Agents with Concise Tool Instruction
• Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security
• Igniting Language Intelligence: The Hitchhiker's Guide From Chain-of-Thought Reasoning to Language Agents
• LLaVA-Plus: Learning to Use Tools for Creating Multimodal Agents
• Agents: An Open-source Framework for Autonomous Language Agents
• Counterfactually Auditable Lifecycle Certification for Autonomous Agents
• AdaPlanner: Adaptive Planning from Feedback with Language Models
• Self-Refine: Iterative Refinement with Self-Feedback
• Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agents
• Distributing Accountability, Not Capability: Phase Separation and the LLM Workflow Quadrant in Autonomous AI Agent Architectures
• WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents
• ScienceWorld: Is your Agent Smarter than a 5th Grader?
• ALFWorld: Aligning Text and Embodied Environments for Interactive Learning
• HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering
• Tool-Planner: Task Planning with Clusters across Multiple Tools

Cited by

• AutoContext: Instance-Level Context Learning for LLM Agents
• Towards General Agentic Intelligence via Environment Scaling
• Memp: Exploring Agent Procedural Memory
• WebDancer: Towards Autonomous Information Seeking Agency
• Large Language Models for Planning: A Comprehensive and Systematic Survey
• SkillNet: Create, Evaluate, and Connect AI Skills

Discussions

• 5. SynWorld Equips LLM-based agents with scenario simulation and MCTS-based refinement of action knowledge arxiv.org/abs/2504.03561 [bsky, 0 points, 1 comments]

Related

• Self-directed Learning of Action Models using Exploratory Planning
• KnowAgent: Knowledge-Augmented Planning for LLM-Based Agents
• Self-Generated In-Context Examples Improve LLM Agents for Sequential Decision-Making Tasks
• AdaWorld: Learning Adaptable World Models with Latent Actions
• Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning