Immersive Explainability: Visualizing Robot Navigation Decisions through XAI Semantic Scene Projections in Virtual Reality

VR visualization of AI explanations significantly improves human understanding: The study found that displaying explainable AI (XAI) attribution scores as visual highlights on objects in a virtual reality scene substantially enhanced non-expert users' objective understanding of why robots made specific navigation decisions, compared to abstract numerical explanations alone.

Combining multiple information sources builds trust: The research demonstrated that integrating both XAI explanations and lidar sensor data visualization was more effective than either alone—participants reported better perceived predictability of robot behavior when they could see both what the robot "thought was important" and what sensors it could actually perceive.

Context matters for AI transparency: By grounding abstract AI attribution scores in recognizable scene objects (a semantic level of explanation), the VR interface made robot decision-making more intuitive for regular people. This suggests that effective AI explainability requires translating technical outputs into human-understandable contexts rather than just presenting raw numbers.

Immersive interfaces may be key for human-robot trust: The use of virtual reality to visualize robot reasoning represents a practical approach to addressing a fundamental challenge in human-robot interaction—that people struggle to trust and work safely with robots whose decision-making remains opaque to them.