Forming Swarm Behavior by Vision-based Autonomous robots using Convolutional Neural Network

Abstract

Many studies have attempted to create life-like swarm behaviors on computer simulations. The most famous and simple model is Boids proposed by Reynolds where individuals just follow three local rules based on distances and positions to the others. The model can show the realistic and dynamic swarm behaviors but it is not plausible in a sense that all individuals can sense the distances to the others precisely in a real world. We have developed vision-based autonomous robots to form a swarm behavior on 2D environment using a convolutional neural network. This paper extends the model to 3D and investigates how the robots can form a swarm behavior. The results of 3D formation patterns by evolved robots are shown.