Learning Motion Policy for Mobile Robots Using Deep Q-Learning
Nosan Kwak, Sujune Yoon, Kyungshik Roh · 2017
We present a deep Q-network (DQN) that learns motion policy for multiple mobile robots. Although DQNs have shown huge successes in several domains, there are few applications to learning for multiple robots. We propose the several advantages of deep Q-learning for multi-robots and present our implementation on a physics-based simulator. We show that the robots learned by the proposed DQN can successfully navigate an unseen environment and escape a dead-lock situation.