reinforcement_learning: true
perception_output_size: 256
decision_output_size: 128
action_output_size: 64