Uncovering manifold structures in Robonaut's sensory-data state space

This paper presents results from the application of dimensionality reduction algorithms to sensory-data time-series that were recorded from Robonaut - NASA's humanoid robot while it was being teleoperated through three tool manipulation tasks. The algorithms tested were principal component analysis, multidimensional scaling, and spatio-temporal Isomap. Structures were shown to exist in some cases, but their detection required careful analysis and a correct choice of parameters