Tensor Based Knowledge Transfer Across Skill Categories for Robot Control

Chenyang Zhao, Timothy Hospedales, Freek Stulp, Olivier Sigaud

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Advances in hardware and learning for control are enabling robots to perform increasingly dextrous and dynamic control tasks. These skills typically require a prohibitive amount of exploration for reinforcement learning, and so are commonly achieved by imitation learning from manual demonstration. The costly non-scalable nature of manual demonstration has motivated work into skill generalisation, e.g., through contextual policies and options. Despite good results, existing work along these lines is limited to generalising across variants of one skill such as throwing an object to different locations. In this paper we go significantly further and investigate generalisation across qualitatively different classes of control skills. In particular, we introduce a class of neural network controllers that can realise four distinct skill classes: reaching, object throwing, casting, and ball-in-cup. By factorising the weights of the neural network, we are able to extract transferrable latent skills that enable dramatic acceleration of learning in cross-task transfer. With a suitable curriculum, this allows us to learn
challenging dextrous control tasks like ball-in-cup from scratch with pure reinforcement learning.
Original languageEnglish
Title of host publicationProceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI-17)
PublisherIJCAI Inc
Number of pages7
ISBN (Electronic)978-0-9992411-0-3
Publication statusPublished - 25 Aug 2017
Event26th International Joint Conference on Artificial Intelligence - Melbourne, Australia
Duration: 19 Aug 201725 Aug 2017


Conference26th International Joint Conference on Artificial Intelligence
Abbreviated titleIJCAI 2017
Internet address


Dive into the research topics of 'Tensor Based Knowledge Transfer Across Skill Categories for Robot Control'. Together they form a unique fingerprint.

Cite this