New Models for Optimizing Mission Control of Unmanned Aerial Vehicles
While unmanned aerial vehicles (UAVs) require no onboard pilots to accomplish their missions, controlling them from the ground is highly labor intensive. Though automated to follow a trajectory and stabilize when subject to wind gusts, UAVs cannot perform more sophisticated maneuvers, such as adjusting a flight plan based on unforeseen events and variable weather conditions, without human input. As a result, for each UAV performing a mission in the skies above Afghanistan and Pakistan, multiple Air Force pilots are needed to provide ground control.
But if algorithms that ECE Professors David Castañón and Christos Cassandras are developing gain traction, those pilots could end up controlling fleets of UAVs rather than teaming up to operate a single vehicle. Funded by the Air Force Office of Scientific Research, the Boston University duo is applying systems engineering techniques to achieve dramatic improvements in UAV onboard decision-making capability. Their research was featured in the April 1 edition of Science Daily.
“Currently, UAVs will go where you want them to go, but they don’t know why,” said Castañón. “We’re trying to develop approaches where teams of UAVs, given sufficient processed information, can determine what tasks each UAV should be doing next. Our goal is to be able to have one operator work with a team of eight to ten unmanned vehicles without having to micromanage them.”
Castañón and Cassandras’ algorithms seek to provide enough automation to enable UAV teams not only to determine and execute an optimal coordinated mission, but also to depart from that plan on the fly when unexpected conditions arise.
To evaluate their algorithms’ ability to perform real-time planning under uncertainty, Cassandras and Castañón developed a test scenario in which teams of small, sensor-toting robots represent the UAVs, and enlisted several undergraduate and graduate students to perform the testing. The robots are programmed to complete selected tasks such as finding objects where other robots aren’t searching—while subject to unexpected events such as the loss of a fellow robot.
“Our approach combines both strategic and tactical decision-making,” Castañón explained. “The algorithms use optimization techniques to enumerate the contingencies a robot would encounter over time, steer them into positions where they’re likely to succeed at their missions, and empower them to make their own decisions based on real-time information—rather than preplanning the entire mission ahead of time.”
Since starting the project three years ago, the BU team has developed algorithms that can make near-optimal decisions in the robotic test platform. They’re now customizing those algorithms to reflect the specific conditions encountered by military UAVs, and aim to begin boosting the autonomy of Air Force UAV fleets within three years.
Because they significantly reduce manpower requirements, these algorithms could lead to an increase in UAVs performing a wide range of military and civilian applications, including surveillance, air traffic control and urban traffic congestion monitoring, The technology could also be used to deploy swarms of robotic unmanned ground vehicles for disaster relief and other high-risk missions.
For example, hundreds of highly autonomous robots could be dispatched to explore terrain and find survivors in the aftermath of natural disasters such as the recent earthquake in Haiti.
“You don’t know the path beneath the rubble or what you’ll encounter,” said Castañón. “But with sufficient autonomy, you could coordinate a team of robots that figure out what to do next when they encounter different situations, and communicate with fellow robots in an efficient manner.”