Demonstration Based Explainable AI for Learning from Demonstration Methods

Explainable feedback helps novices teach robots more effectively: The study of 26 participants found that when users received adaptive explanatory feedback showing both successful and unsuccessful robot trajectories, they were able to teach robots navigation tasks more efficiently and achieve better overall robot performance compared to systems without this feedback.

Showing representative examples of both successes and failures matters: Rather than only demonstrating what the robot learned correctly, the system strategically showed users a mix of successful and unsuccessful trajectories. This balanced approach to explanation appears to be key to helping users understand what the robot was learning and how to improve their teaching.

Better explanations lead to better user understanding: Beyond just improving robot performance, the study found that users who received the adaptive explanatory feedback demonstrated improved understanding of how the robot was learning—suggesting that XAI in learning systems isn't just about performance, but about building user comprehension and trust.

Adaptive explanations can reduce teaching burden: By selectively sampling which trajectories to show users rather than overwhelming them with all examples, the system improved teaching efficiency, indicating that thoughtfully designed explanations can make it easier for non-experts to work with AI systems.