A GAN-based Active Terrain Mapping for Collaborative Air-Ground Robotic System

Abstract

Collaborative air-ground robotic system has recently emerged as an important research area and shown great potential in many practical applications of smart cities. This work aims to use such system to transform the aerial images from UAVs into terrain map exploited by UGVs to perform ground path planning or navigation tasks. We propose a novel GAN-based active terrain mapping (GAN-ATM) algorithm which integrates Active Learning (AL) strategy into Generative Adversarial Network (GAN) framework to build the terrain map efficiently with a very limited number of labeled data. The empirical results show that the proposed algorithm achieves the highest predictive accuracy of 90.35%. Due to a more accurate terrain map, the UAV using GAN-ATM can plan the shortest trajectory among all existing counterparts.