Constructive Approach to Role-Reversal Imitation Through Unsegmented Interactions

摘要

This paper presents a novel method of a robot learning through imitation to acquire a user’s key motions automatically. The learning architecture mainly consists of three learning modules: a switching autoregressive model (SARM), a keyword extractor without a dictionary, and a keyword selection filter that references to the tutor’s reactions. Most previous research on imitation learning by autonomous robots targeted motions given to robots, were segmented into meaningful parts by the users or researchers in advance. To imitate certain behavior from continuous human motion, however, robots must find segments to be learned. To achieve this goal, the learning architecture converts a continuous time series into a discrete time series of letters using the SARM, finds meaningful segments using the keyword extractor without a dictionary, and removes less s meaningful segments from keywords using the user's reactions. In experiments, an operator showed unsegmented motions to a robot, and reacted to the motions the robot had acquired. Results showed that this framework enabled the robot to obtain several meaningful motions that the operator hoped it would acquire.