Time normalization and analysis method in robot programming from human demonstration data

Abstract

Methods have been proposed to generate robot motions from human demonstrations. Humans demonstrate tasks without indicating what they want to teach. In future, the robot programming methods from human demonstrations will notice the intention of the human's actions and generate or modify the program from demonstrations. This paper describes two elements which are necessary in analyzing human demonstrations. The first is a time normalization method of demonstration data which humans show at different speeds. The second is a method to detect various changes in data without prior knowledge of how many or what kind of changes will occur. The scheme of robot programming from demonstrations is presented and the wavelet analysis result of time normalized demonstration data are discussed.