Autonomous Flight and Navigation in Air-Ground Systems

Abstract

Robots are being used in situations that require more robust navigation and search. Often in these areas GPS can be unreliable or not present at all. In these situations, it becomes viable and even beneficial to consider collaboration between heterogeneous robots in order to utilize the different strengths of these robots, while avoiding situations that a specific robot’s weakness would decrease the probability of a successful mission. A specific example of this is a scenario involves unmanned ground and air vehicles (UGV, UAV), where the ground robot contributes its high payload capacity to provide computational resources, large battery life, and high accuracy sensors to provide high accuracy localization while the aerial robot can bring its high mobility to explore hard to get to regions. In these situations, the UAV can have different levels of sensing capability. UAVs with poor sensing capabilities need a way to navigate and explore different environments. For this problem a state lattice planner augmented with controller-based motion primitives (SLC) is used to provide UAVs with a robust way to navigate while utilizing a large variety of sensors and even the help of other robots in the environment. Another type of UAV has good sensing capabilities and is able to localize very well in x and y. However, the method for height estimation can often be naïve, especially when LIDAR data is readily available. This method is sensitive to surfaces with debris or objects on the ground plane, especially ramps, which can then lead to a poor height estimation. To solve this problem for the Search-Based Planning Lab (SBPL) my approach is to use these LIDAR scans to build a probabilistic elevation map (PEM) of the environment, which is then used to better estimate the height of the UAV as it traverses over these objects and environments. This method is also shown to work with dynamic obstacles and can be used to provide more features for the UAV.