VGG-16 Neural Network-Based Visual Artificial Potential Field for Autonomous Navigation of Ground Robots
摘要
This paper describes a novel end-to-end robot visual navigation architecture using monocular gray scale images and a VGG-16-based neural network architecture to safely navigate a differential drive robot Plantroid, avoiding static and moving obstacles. Training data for the neural network is generated in simulated environments using the artificial potential field method. Experiments with the virtual and real robot were performed to validate the proposed architecture.