VGG-16-Based Map-Less Navigation Architecture with Temporal Vision Mosaic for Autonomous Ground Robots

摘要

This paper introduces a novel VGG-16-based visual navigation architecture for a differential drive robot, Social Plantroid, using a temporal vision mosaic, a novel approach which joins current and previous robot vision frames for heading estimation. As a minor contribution, it integrates a novel sunlight/shadow detection algorithm using Gabor filters. The neural network is trained with simulated data employing the artificial potential field method, which is another novelty for map-less robot navigation. Virtual and real-world experiments validate the effectiveness of this architecture in obstacle avoidance and navigation.