Landmark-based localization for Unmanned Aerial Vehicles

Abstract

Localization is an aspect of robotics that is of fundamental importance in the deployment of autonomous vehicles. Robots need to know where they are relative to a global frame of reference, or other robots. Robot odometry is a trivial way of acquiring distance travelled by an autonomous vehicle. However, odometry has inherent flaws such as errors caused by wheel slippage on ground based vehicles. Other platforms like Unmanned Aerial Vehicles (UAVs) have built-in odometry capabilities that can be affected by drift. Further, the Global Positioning System (GPS) has a 10 foot error, which contributes a significant error to the robots location as discussed in [10]. Landmark-based localization is an ideal supplement to odometry and GPS. Recognition of landmarks such as tags or terrain using cameras can provide localization data. The Parrot AR drone was the platform for the landmark-based localization experiments. The real-time camera feeds from the drone along with ROS' (Robot Operating System) AR tag node provided the parameters: roll, pitch, yaw, x-metric, y-metric, z-metric. Utilizing a mathematical algorithm and the camera feed, the relative position of the drone to a point of origin was calculated. The error associated with the position was an acceptable 100mm-150mm, which was a significant improvement compared to other localization methods. Landmark-based localization is proving to be an effective way to attain position information when other sources such as GPS are unavailable as described in [11]. Despite its advantages, certain limitations and challenges need addressing. Dealing with limitations in camera image quality, lighting and locale restrictions would require further exploration.