Learning to Represent Haptic Feedback for Partially-Observable Tasks. ICRA, 2017.

Learning to Represent Haptic Feedback for Partially-Observable Tasks. ICRA, 2017.

https://arxiv.org/pdf/1705.06243.pdf

Physical environments such as automated homes need to be able to take actions even in presence of limited, incomplete observations.

We formulate the task as a POMDP and learn an appropriate representation with deep recurrent neural networks that serves as the state for the POMDP model.