Human and Embedded MAS Communication with KQML Using Chat-BDI

Abstract: Belief–Desire–Intention (BDI) agents have been widely adopted in Multi-Agent Systems (MAS) due to their suitability for autonomous decision-making. Recent advances have demonstrated that BDI-based MAS can be embedded in resource-constrained, geographically distributed devices, enabling autonomous operation in real-world and pervasive environments. However, while agent-to-agent communication in such systems is well supported through standardized languages and middleware, effective Human–Agent Interaction (HAI) remains a significant challenge, particularly in scenarios that require real-time supervision, transparency, and ethical human control. Existing approaches either lack support for embedded and distributed settings or expose low-level communication details that hinder usability and interactive supervision. This work presents a chat-based interaction approach that enables humans to communicate with geographically distributed BDI-based MAS. The approach allows human users to select speech-act performatives and message content through a web interface, which automatically generates valid Knowledge Query and Manipulation Language (KQML) messages and delivers them to embedded agents. By abstracting away protocol and architectural complexities, the solution enables human supervision of autonomous agents. A proof-of-concept implementation was developed and evaluated using a resource-constrained device simulator, in which a BDI-controlled Unmanned Aerial Vehicle receives and executes human instructions during flight. The results demonstrate the approach's feasibility and effectiveness in supporting human-in-the-loop supervision for embedded and geographically distributed MAS.