DG

Low dimensional state representation learning with reward-shaped priors

Nicolò Botteghi and Ruben Obbink and Daan Geijs and Mannes Poel and Beril Sirmacek and Christoph Brune and Abeje Mersha and Stefano Stramigioli (2021), 2020 25th International Conference on Pattern Recognition (ICPR), 3736--3743

Presents a reinforcement learning method that learns compact, low-dimensional state representations guided by reward-shaped priors, making robot control policies more efficient to train.