SKID RAW: Skill Discovery From Raw Trajectories

Integrating robots in complex everyday environments requires a multitude of problems to be solved. One crucial feature among those is equip with mechanism for teaching them new task an easy and natural way. When tasks that involve sequences different skills, varying order number these it desirable only demonstrate full executions instead all individual skills. For this purpose, we propose novel approach simultaneously learns segment trajectories into reoccurring patterns the skills reconstruct from unlabelled demonstrations without further supervision. Moreover, skill conditioning can used understand possible practical in, example, human-robot-interactions more intelligent adaptive robot behaviour. The Bayesian variational inference based evaluated on synthetic real human complexities dimensionality, showing successful learning segmentations libraries data.