Touch-based Curiosity (ToC) learns what visibleobjects interactions are supposed to “feel” like . We encourage exploration by rewarding interactions where the expectation and the experience don’t match . We compare our cross-modal approach to single-modality (touch- or vision-only) approaches as well as othercuriosity-based methods and find that our method performs better and is moresample-efficient . We test our approach on a range of touch-intensive robotarm tasks (e.g. pushing objects, opening doors) across multiple experiments in a simulated setting, which we also release as part of this work . ToC is able to learn these difficult tasks throughsparse reward and curiosity alone . We found that our approach performs better, is more sample-efficient and is less sample-intensive

Author(s) : Sai Rajeswar, Cyril Ibrahim, Nitin Surya, Florian Golemo, David Vazquez, Aaron Courville, Pedro O. Pinheiro

Links : PDF - Abstract

Code :

Keywords : touch - tasks - based - curiosity - approach -

Leave a Reply

Your email address will not be published. Required fields are marked *