Research Interests
Thesis title: Comparing synchrony detection algorithms for robotic self-other discrimination.Thesis Advisor: Dr. Chris Prince.
Abstract
Robot learning techniques that are based on infant psychological development have lately been gaining popularity. Since infants learn, in part, by detecting synchrony between their sensory and motor systems, we hypothesize that synchrony detection can be a useful basis for improving the adaptive qualities of robots. The goal of this thesis was to compare two time-based synchrony detection algorithms with respect to their ability to perform visual self-other discrimination. Self-other discrimination is the ability to distinguish between any effect caused by one's self from any effect caused by some other entity in the world.The two algorithms chosen for comparison were Hershey and Movellan (2000) and Fitzpatrick (2003). The Hershey and Movellan (2000) algorithm had already been implemented in our laboratory (Mislivec, 2004 and Helder 2003) and was extended in this thesis to make it capable of self-other discrimination. The Fitzpatrick (2003) method was implemented by us, with some modi¯cations to suit our requirements. The Fitzpatrick method was able to discriminate between visual motion of self and visual motion of other with an accuracy of 74.34% as compared to 56.67% for Hershey and Movellan (2000). These results are for the most demanding case of self-other discrimination, i.e., when both the self and the other are simultaneously in motion. Thus we conclude that the Fitzpatrick (2003) method performs better under our test conditions than the Hershey and Movellan (2000) algorithm at our task of self-other discrimination. We also present some suggestions for future research, including possible improvements to the performance of the two methods.