Integrating Ontology with Deep Reinforcement Learning: A Formal Framework for Explainability in Robotic Applications
Auteurs
Bhattacharya S., Naudet Y.
Référence
Proceedings of the 18th ACM International Conference on Pervasive Technologies Related to Assistive Environments Petra 2025, pp. 60-63, 2025
Description
This paper presents a formal framework that integrates an ontology with Deep Reinforcement Learning (DRL) to enhance explainability in AI-driven robotic applications. By mapping DRL components to ontological concepts, our approach generates human-readable explanations for complex model decisions. We demonstrate the framework through a robotic arm pick-and-place task, illustrating improved transparency and interpretability. The integration facilitates domain knowledge incorporation while addressing the critical need for explainable AI systems. Despite challenges in computational overhead and ontological alignment, our contribution advances trustworthy AI systems that support effective human-AI collaboration.
