• Decision: submitted, no decision
• Abstract: We propose to deal with sequential processes where only partial observations are available by learning a latent representation space on which policies may be accurately learned.