LLM-mediated Dynamic Plan Generation with a Multi-Agent Approach
Reo Abe, Akifumi Ito, Kanata Takayasu, Satoshi Kurihara
TL;DR
This paper addresses dynamic environment planning for autonomous systems by proposing a fully automatic network-generation pipeline that uses GPT-4o within the Agent Network Architecture (ANA) to fuse reactive and deliberative planning. It introduces object-based and condition-based status collection, recursive network construction, and network optimization to produce large-scale, generalizable agent networks without relying on manually integrated planning frameworks. Evaluation shows that the automatically generated networks achieve substantial coverage relative to human-crafted networks and can be expanded to larger scales, though planning performance can degrade without targeted expansion in very large networks. The work demonstrates a significant step toward scalable, adaptable planning for robotics, autonomous vehicles, and smart systems, with future work focusing on scale-control strategies and real-world validation.
Abstract
Planning methods with high adaptability to dynamic environments are crucial for the development of autonomous and versatile robots. We propose a method for leveraging a large language model (GPT-4o) to automatically generate networks capable of adapting to dynamic environments. The proposed method collects environmental "status," representing conditions and goals, and uses them to generate agents. These agents are interconnected on the basis of specific conditions, resulting in networks that combine flexibility and generality. We conducted evaluation experiments to compare the networks automatically generated with the proposed method with manually constructed ones, confirming the comprehensiveness of the proposed method's networks and their higher generality. This research marks a significant advancement toward the development of versatile planning methods applicable to robotics, autonomous vehicles, smart systems, and other complex environments.