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

TL;DR

TAPAS introduces a multi-agent framework combining LLMs and symbolic planning to adaptively generate and modify domain models for complex task planning without manual environment modeling.

Contribution

It presents a novel multi-agent system that uses structured tool-calling with LLMs for dynamic domain modeling and planning in complex environments.

Findings

Strong performance in benchmark planning domains

Effective adaptation to novel attributes and constraints

Bridges gap between plans and robot capabilities

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.