A geometric approach to task understanding and playback: compactand robust task description for complex environments

摘要

The ability to execute similar tasks in various environments isimportant for robot systems, but there are few approaches for doingthis. We recently proposed a “teaching by showing” systemthat recognizes teaching tasks and executes corresponding tasks indifferent environments. The teaching task is recognized from geometricalconstraint and translated into a general task description (GTD).However, this method is difficult to apply in complex environmentsbecause of redundancy an description and lack of generalization. Thispaper presents an algorithm for both two- and three-dimensionalenvironments that solves these problems. The improved GTD describestasks in complex environments very compactly. It ignores objects that donot affect the task, which adds a useful robustness to the system