Vision based localization and control of UGVs using a UAV

Official URL: https://risc01.sabanciuniv.edu/record=b2708497_(Table of contents)

Heterogeneous systems especially those that pair UAVs with UGVs are becoming increasingly popular. The advantages a multi-agent robot system brings are desirable for a multitude of tasks such as search and rescue, surveillance and reconnaissance. The mobility of UAVs by far makes them a better fit for perception and mapping tasks while the power and interaction capabilities of UGVs make them the desirable mode of operation for several other tasks. Thus, a system that looks to utilize them cooperatively has to solve the integral problem of localizing both robots in the same global map. Adding to this problem is the fact that a UAV-UGV pair is often desirable in scenarios where a common global sensor system such as GPS may not be available. In this thesis, a method to localize both vehicles in a single map is proposed by utilizing inertial sensors based estimates which are corrected through a single visual sensor onboard the UAV. We explore a method of generating a unified map which can be used to localize both the UGVs in the area as well as the UAV by levying the additional charge of conducting visual odometry and object detection on the UAV.