Learning Robot Skills Through Motion Segmentation and Constraints Extraction

Learning skills while interacting with human users allows robots to work in human environments and become efficient when performing everyday tasks. This paper presents a method of analyzing data obtained from human demonstrations of a task in order to: (a) extract specific constraints for each part of the task (important variables, the frame of reference to be used, and the most suitable controller for reproducing the task); (b) use these task features as continuously embeddable constraints in the learned robot motion (c) properly segment into subtask; The proposed method has been tested on a common kitchen task and its performance was compared against standard control modes.