Varying How We Teach: Adding Contrast Helps Humans Learn about Robot Motions
Tiffany Horter, Elena L Glassman, Julie Shah, and Serena Booth
ACM/IEEE International Conference on Human-Robot Interaction (HRI), Human-Interactive Robot Learning (HIRL) Workshop, 2023
Learning how robots move is difficult, but theories of human concept learning can be applied to support humans in this task. We draw insights from the Variation Theory of Learning, a theory that has been validated in the learning sciences through decades of classroombased studies. Variation Theory prescribes experiencing patterns of structured variation, where some aspects of concepts are held constant while other aspects vary. The result of experiencing these structured patterns is that human learners develop accurate and flexible conceptual models. Through a preliminary study, we show that using insights from Variation Theory improves humans’ ability to predict robot motions: accuracy in predicting motions increases from 52.4% using a familiarization-based strategy to 70.2% using a Variation-based strategy. Applying Variation Theory especially increases the human’s accuracy in predicting robot motions in novel settings (increasing from 50.0% to 72.4% accuracy).