[2Yin-B] Poster 4

[2Yin-B-40] A BDI Agent Considering Others' Intentions Based on a Multi-Context System

〇Hiraku Gondo¹, Hiroki Sakaji¹, Itsuki Noda¹ (1. Hokkaido University)

Keywords:

BDI logic,Logical Reasoning,Multi Agent

In this study, we construct a BDI (Belief–Desire–Intention) agent that estimates the intentions of others and incorporates them into its own decision-making, based on a multi-context system. While BDI logic enables formal reasoning about agents’ intentions through beliefs, desires, and intentions, its high expressive power leads to increased computational complexity. To address this issue, we adopt a multi-context system for BDI in which beliefs, desires, and intentions are separated into independent contexts, thereby providing a framework that balances tractable reasoning with formal verifiability. Specifically, an LLM is employed to generate hypotheses regarding other agents’ intentions, and these hypotheses are verified through logical inference within each context. In this way, we integrate natural language–based intention estimation with formal validation grounded in logical reasoning. Experiments are conducted in a multi-agent simulation environment involving both cooperative and adversarial interactions, demonstrating that the proposed method is effective for intention estimation and action selection.