Transparent Active Learning for RobotsReportar como inadecuado

Transparent Active Learning for Robots

Transparent Active Learning for Robots - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

This research aims to enable robots to learn fromhuman teachers. Motivated by human social learning, we believethat a transparent learning process can help guide the humanteacher to provide the most informative instruction. We believeactive learning is an inherently transparent machine learning approachbecause the learner formulates queries to the oracle thatreveal information about areas of uncertainty in the underlyingmodel. In this work, we implement active learning on the Simonrobot in the form of nonverbal gestures that query a humanteacher about a demonstration within the context of a socialdialogue. Our preliminary pilot study data show potential fortransparency through active learning to improve the accuracyand efficiency of the teaching process. However, our data alsoseem to indicate possible undesirable effects from the humanteacher’s perspective regarding balance of the interaction. Thesepreliminary results argue for control strategies that balanceleading and following during a social learning interaction.

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

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


Documentos relacionados