XAIgent: A Multi-Agentic Design Method for Explainable Human-Robot Interaction in Scientific Discovery

Le résumé fourni par la source

Robotic systems are increasingly used in scientific exploration, but their opaque decision-making processes hinder interpretability, limiting human understanding and trust in human-robot collaboration. To address this, explainability must be integrated into interactive human-robot interaction (HRI) systems, especially for non-roboticist scientists. This work presents ‘XAIgent’, a multi-agent system (MAS)-based framework designed to support interactive and explainable HRI in scientific exploration. By decomposing robotic decision-making into modular, agent-based components, XAIgent provides more interpretable and adaptive explanations than monolithic approaches. Our approach guides the integration of explainability into robotic assistants to accelerate scientific discovery. We outline key principles for designing interactive systems that allow scientists to query a robot’s actions, understand its reasoning and errors in real-time, intervene when necessary, and explore its functionalities to enhance transparency and control by demonstrating a simulated scientific environment created in Unity, accompanied by a preliminary user study evaluating the system’s effectiveness. Implementation details are provided at: https://collaborative-work-space.github.io/xaigent/.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.