Learning based semantic segmentation for robot navigation in outdoor environment

This work presents a vision-based navigation control strategy for a wheeled robot traveling outside. The main achievements contain road detection with deep learning and navigation control scheme of a robot. A deep convolutional neural network is first employed to perform pixel-wise segmentation and thus to find road regions. Next, a fuzzy controller is designed for commanding a robot's activities including movement and speed. Experimental results verify the essential capability of navigation by using a deep architecture of convolutional network. In brief, the proposed approach certainly reaches autonomous traveling with collision avoiding and wayfinding functionalities.