Generalizable task representation learning from human demonstration videos: a geometric approach | IEEE Conference Publication | IEEE Xplore