Perception system for autonomous guided vehicles

Abstract

This report presents a perception system using deep convolutional neural networks that enables real time navigation of autonomous guided vehicles in urban environments. With the advancement of robotic technologies, there is a stark growth in the usage of robots in industry. To work in urban environments alongside humans, these robots adopt the see-think-act notion. The outputs of the perception system provide the robot, a more useful representation of its surrounding environment identifying known obstacles and predicting their behavior. The robot used in this project is an Unmanned Ground Vehicle (UGV) mounted with an array of sensors, including a 2D LIDAR and a RGB camera. The perception system performs two major tasks; (1) Mapping and (2) Object detection to give the robot a more useful representation of its surrounding environment. Mapping is performed using the Simultaneous Localization and Mapping algorithm. A deep convolutional neural network model is used to perform real time object detection to classify objects and find bounding boxes for those objects. These algorithms are realized in the robot using the Robot Operating System framework. The developed perception system shows promising results as it allows the robot to dynamically avoid obstacles while navigating from point to point.