Why follow the beaten path when untamed terrain holds so many exciting possibilities? That’s the trailblazing mindset behind the creation of Mississippi State University’s Autonomous Vehicle Simulator, or MAVS — a tool that augments MSU’s strengths in driverless research by providing off-road test capabilities.
MAVS is essentially a virtual proving ground where researchers can test autonomous software in a variety of digital environments and terrains. The simulator allows autonomy developers to test and debug algorithms in early stages and resolve flaws that could potentially lead to injury or vehicle and equipment damage during field testing.
MSU researchers also have access to an actual proving ground — a 50-acre property located near the CAVS facility where they test autonomous algorithms on real vehicles interacting with real environments.
MAVS is powered by Mississippi State’s expansive expertise in vehicle dynamics, engineering, robotics, computational physics, high-performance computing and related sciences, as well as a culture of interdisciplinary teamwork and collaboration.
Goodin points out that most military ground vehicles are designed to travel off-road, which makes MAVS an indispensable tool for accelerating the shift to autonomy. One area of interest is developing autonomous vehicles with leader-follower capabilities. Experience shows that military truck convoys are vulnerable to attacks, but if the first vehicle is manned and others are driverless vehicles “trained” to follow, fewer troops will be in harm’s way.
Driverless vehicles are equipped with a multitude of sensors that process information about their surroundings and help them “decide” where to go and how to get there. Vehicles won’t get very far without reliable, accurate environmental sensing algorithms that enhance their intelligence and agility.
When it comes to experiments involving human performance, there are many vehicle tests made possible by MAVS that are much more practical because of safety constraints. For example, researchers are using MAVS to conduct pedestrian detection and avoidance experiments by simulating human movements and behaviors in virtual environments.
MAVS also is creating a wealth of teaching, learning and research opportunities for faculty and students by providing hands-on experience in areas such as sensing perception, machine learning and autonomy.
Machine-learning algorithms developed by Mississippi State engineering students are not only improving vehicle capabilities but saving a tremendous amount of time and cost involved in image and data processing. Examples include algorithms that help vehicles recognize solid objects obscured by tall grass and estimate the density of overhead vegetation.
By opening new worlds of research data to graduate students, MAVS has been fertile ground for thesis and dissertation studies that have helped advance autonomous capabilities — for instance, by giving vehicles the capability to detect and quantify terrain roughness and recognize “negative obstacles” such as holes and ditches.
Mississippi State isn’t the only university benefiting from MAVS. The fact that MAVS is non-commercial, opensource software means any university in the state or country can use it. The MAVS technology also is available to private sector companies via licensing arrangements.