Gaussian-process-based robot learning from demonstration

Abstract Learning from demonstration allows to encode task constraints observing the motion executed by a human teacher. We present Gaussian-process-based learning (LfD) approach that robots learn manipulation skills demonstrations of By exploiting potential Gaussian process (GP) models offer, we unify in single, entirely GP-based framework, main features required for state-of-the-art LfD approach. address how GP can be used effectively policy trajectories space. To achieve an effective generalization across demonstrations, propose novel Task Completion Index (TCI) temporal alignment trajectories. Also, our probabilistic representation encoding variability throughout different phases task. Finally, method efficiently adapt fulfill new requirements and modulate robot behavior as function variability. This has been successfully tested real-world application, namely teaching TIAGo open types doors.