go back

Learning Task-Parameterized Skills from Few Demonstrations

Jihong Zhu, Michael Gienger, Jens Kober, "Learning Task-Parameterized Skills from Few Demonstrations", ICRA / RA-L, 2022.


This is a follow-up of pub-4739 Moving away from repetitive tasks, robots nowadays demand versatile skills that adapt to different situations. Task- parameterized approaches improve the generalization of motion policy by encoding relevant contextual information in the task parameters, hence enabling flexible task executions. However, training such a policy often requires collecting multiple demon- strations in different situations. To create these situations, objects or sometimes even humans need to move around, which renders method the less applicable to real-world problems. Therefore, training with fewer demonstrations/situations is desirable. In the paper, we utilize task parameters to generate new demonstration data that augments the original training data set for policy improvements, thus allows learning task-parameterized skills with few demonstrations.

Download Bibtex file Download PDF