Adaptive Domain Modeling with Language Models: A Multi-Agent Approach to Task Planning

2025-06-24

AI Summary

This work addresses complex task planning without manually engineered environment models. We propose TAPAS, a multi-agent collaborative framework integrating large language models (LLMs) with symbolic planning. Its core contribution is a dynamic tool-calling mechanism enabling closed-loop inter-agent coordination: downstream agents can request upstream agents to modify the domain model, initial state, or goal specification in real time—adapting to new attributes and constraints without redefining domain knowledge. TAPAS incorporates ReAct-style reasoning-execution, automatic natural-language-to-PDDL plan translation, and structured tool invocation. Evaluated on standard planning benchmarks and the VirtualHome simulation platform, TAPAS achieves significant improvements in task completion rate and model generalization, demonstrating the feasibility of efficient, robust task planning under zero-shot, human-model-free conditions.

Technology Category

• Multiagent Systems: Multiagent Planning
• Planning, Routing, and Scheduling: Planning with Language Models
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

We introduce TAPAS (Task-based Adaptation and Planning using AgentS), a multi-agent framework that integrates Large Language Models (LLMs) with symbolic planning to solve complex tasks without the need for manually defined environment models. TAPAS employs specialized LLM-based agents that collaboratively generate and adapt domain models, initial states, and goal specifications as needed using structured tool-calling mechanisms. Through this tool-based interaction, downstream agents can request modifications from upstream agents, enabling adaptation to novel attributes and constraints without manual domain redefinition. A ReAct (Reason+Act)-style execution agent, coupled with natural language plan translation, bridges the gap between dynamically generated plans and real-world robot capabilities. TAPAS demonstrates strong performance in benchmark planning domains and in the VirtualHome simulated real-world environment.

Problem

• Integrates LLMs with symbolic planning for complex tasks
• Generates and adapts domain models without manual redefinition
• Bridges dynamic plans with real-world robot capabilities

Innovation

• Multi-agent framework integrating LLMs with symbolic planning
• Specialized LLM-based agents adapt domain models dynamically
• ReAct-style execution bridges plans and robot capabilities