Contributions to invariant representations of rigid-body trajectories with applications to the segmentation, recognition and generation of rigid body motions

Future robots are expected to operate in unpredictable and changing environments where there is a large variety in objects to handle and tasks to perform. For robots to successfully operate in these environments, they need to be capable of recognizing objects (and their affordance) and adapting plans online in a more general and reactive way. The research focus is on interpreting, analyzing and executing object-manipulating contact-tasks. The overall objective of this research is to design a general framework for robot learning of contact-tasks consisting of the following parts: 1. Fast and robust estimation of the contact-type based on measured motion and contact-force data. 2. Generalization of the contact-task model learned from human demonstrations by removing contextual information.  3. Reactive planning of shape-adjusted reference motions on the robot starting from the generalized contact-task models. To accomplish these objectives, coordinate invariant shape descriptors of contact-motions are used.