Designing Interactions for Robot Active LearnersReport as inadecuate

Designing Interactions for Robot Active Learners

Designing Interactions for Robot Active Learners - Download this document for free, or read online. Document in PDF available to download.

This paper addresses some of the problems that arisewhen applying active learning to the context of human–robot interactionHRI. Active learning is an attractive strategy for robotlearners because it has the potential to improve the accuracy andthe speed of learning, but it can cause issues from an interactionperspective. Here we present three interaction modes that enable arobot to use active learning queries. The three modes differ in whenthey make queries: the first makes a query every turn, the secondmakes a query only under certain conditions, and the third makesa query only when explicitly requested by the teacher.We conductan experiment in which 24 human subjects teach concepts to ourupper-torso humanoid robot, Simon, in each interaction mode, andwe compare these modes against a baseline mode using only passivesupervised learning.We report results from both a learning and aninteraction perspective. The data show that the three modes usingactive learning are preferable to the mode using passive supervisedlearning both in terms of performance and human subject preference,but each mode has advantages and disadvantages. Based onour results, we lay out several guidelines that can inform the designof future robotic systems that use active learning in an HRI setting.

Socially Intelligent Machines Lab SIM - Socially Intelligent Machines Lab SIM Publications -

Author: Cakmak, Maya - Chao, Crystal - Thomaz, Andrea L. - -


Related documents