Generating collective foraging behavior for robotic swarm using deep reinforcement learning

摘要

This paper mainly discussed the generation of collective behaviors with raw camera images as the primary information input. The swarm robotic system exhibits considerable advantages when faced with individual-level failure or the lack of global information. Spatial information has always been a necessity in generating collective transport behavior. The rise of deep neural network technology makes it possible for a robot to perceive the environment from its visual input. In this paper, the use of deep reinforcement learning in training a robotic swarm to generate collective foraging behavior is shown. The collective foraging behavior is evaluated in a transportation task, where robots need to learn to process image information while cooperatively transport foods to the nest. We applied a deep Q-Learning algorithm and several improved versions to develop controllers for robotic swarms. The results of computer simulations show that using images as the main information input can successfully generate collective foraging behavior. Besides, we also combine the advantages of several algorithms to improve performance and perform experiments to examine the flexibility of the developed controllers.