Autonomy is essential for flying on Mars because it’s impossible to remotely operate a flying vehicle due to the communications delay between Earth and the Red Planet. In order to navigate and explore new places, the vehicle must know where it has been and how far it has traveled. On Earth, this localization problem is solved with GPS, but on Mars we must use other means. Over the summer I integrated a state-of-the-art visual odometry technique into NASA’s vehicle. This technique, Semi-Direct Visual Odometry by Forster et al, allows for the vehicle to be localized with centimeter accuracy using only information from a camera facing the surface of the planet. Solving the localization problem paves the way for autonomous navigation and exploration.
Last summer, I worked at NASA’s Langley Research Center on a prototype for the first flying vehicle to be sent to Mars. Specifically, I developed the computer vision system aboard the vehicle using C++, OpenCV, the Robot Operating System, and Christian Forster’s Semi-Direct Visual Odometry Project. In the video above, I briefly overview the technologies I integrated to achieve camera-only localization in a simulated Mars environment. In the video below, I explain the project in detail during a technical talk at NASA.