Multi-agent active perception with prediction rewards (Supplementary material)