Online Explanation Generation for Human-Robot Teaming

Timing matters for understanding: The paper demonstrates that spreading explanations throughout a robot's task execution—rather than providing all information at once—reduces mental workload for human teammates, as measured by NASA Task Load Index (TLX) scores in human subject studies.

Online explanations must account for dependencies: A key technical challenge identified is that different parts of an explanation often depend on each other; the paper presents methods to handle these dependencies when generating explanations in real-time rather than all at once.

Explainability should consider human cognitive limits: The research challenges the assumption that humans can understand any explanation regardless of how much information is presented, arguing that responsible AI design must account for the mental demands placed on human teammates working alongside robots.

Practical testing with humans validates the approach: The authors evaluated their methods with actual human subjects in a simulated rover task, not just synthetic testing, providing real-world evidence that their online explanation approach improves human-robot team performance.

Explanation generation is foundational to trustworthy robotics: The paper positions the ability to generate clear, well-timed explanations as "one of the key requirements of explainable agency," suggesting that robots cannot be truly trustworthy teammates without this capability.