Modular Deep Reinforcement Learning for Emergent Locomotion on a Six-Legged Robot Conference Paper