Generating Collective Behavior of Robotic Swarm based on Deep Reinforcement Learning and Analysis of Controller

Abstract

Swarm robotics is the field of study that aims to generate a collective behavior of a large number of robots. In this paper, we applied deep reinforcement learning to design a controller for a robotic swarm. Deep reinforcement learning is a combination of reinforcement learning with deep neural networks that contain convolution layers. With the help of convolution layers, robots are able to learn from the camera image as the input of a deep neural network. When using deep reinforcement learning, however, it is difficult to understand the mechanism of improving performance. This paper analyses the convolution layers of the controller to understand the learning mechanism of a robotic swarm.