Tiffany Lee, April 27, 2021
Bradley’s CAREER project to boost drones’ adaptability, performance
Drones are quickly becoming a ubiquitous tool in the United States, playing a role in the military, scientific research, journalism, weather forecasting and more — including, soon, Amazon package delivery. And the industry is expected to grow at a rapid clip: By 2023, the unmanned aircraft systems sector is projected to grow about 25%.
To many of us, today’s drones may still seem like something from a science fiction novel. But in terms of their capability, reliability and safety, the best is yet to come, said Justin Bradley, assistant professor of computer science and engineering at the University of Nebraska–Lincoln. Bradley is using a nearly $500,000, five-year grant from the National Science Foundation’s Faculty Early Career Development Program to give drones and other robots a highly advantageous quality that humans routinely capitalize on to get things done: the ability to sense and adapt to a changing environment, diverting energy to the most important tasks at hand while pausing lower-priority work.
“The best way to think about it is that when you are doing tasks, you can rapidly refocus your attention as you need to. If something happens in your environment, you can look, assess and shift your attention as needed to adapt to the situation,” Bradley said. “Robots don’t have that capability. This work would allow them to allocate that attention and other resources according to what they need to do.”
In the context of robotics, Bradley said, “attention” is the rough equivalent of computational power. To enable a drone or robot to sense its environment and direct computational power to a priority task, he will develop mathematical and computing foundations, analytical tools and algorithms that guide resource allocation processes and optimize overall performance. Bradley will then incorporate these frameworks into his final product: a resource-aware autopilot system that incorporates path-planning algorithms, along with new controllers for the drone.
The autopilot product will be compatible with the widely used, open-source Robot Operating System and will operate on common, small robotics hardware. Bradley is working to obtain a provisional patent for the technology.
“From an impact standpoint, this will be really useful for future missions that are complex and changing — which pretty much every mission is,” he said. “This will allow robotic aircraft to do longer and more efficient surveillance and reconnaissance missions where the environment is changing.”
Bradley’s autopilot system will also unlock the potential of “anytime algorithms,” which were originally devised in the 1980s to offer a tradeoff between computation time and the quality of the algorithm’s solution. These algorithms produce answers whose precision is commensurate to the resources devoted to solving them: The greater the time and computational resources available, the more accurate the answer.
Despite their utility in balancing computation time with precision, anytime algorithms haven’t been widely adopted because there hasn’t been a ubiquitous framework through which to deploy them, Bradley said. His work may change that.
“This is the first time they’ve been given a chance to take off, even though they’ve been on the theoretical shelf for 30 years,” he said.
Bradley will test his autopilot system at a rainforest in Costa Rica where drones are engaged in a sequence of tasks ranging from computationally simple to complex, making it the perfect testing grounds. For part of the mission, the drones fly waypoints, which means they are simply flying a route between different physical locations. But the other part of the mission, collecting samples from leaves, is more demanding.
“When the drone reaches a collection point, a lot of stuff has to happen. It builds a plan to navigate a very cluttered environment, drops its collector mechanism and grabs a leaf, all in the presence of wind and downwash from propellers. And as it does this, its characteristics change, including a change in its center of mass,” Bradley said. “It needs to adjust to these things and allow itself to refocus.”
Giving drones the ability to reallocate resources will be crucial to maximizing the full potential of artificial intelligence and machine learning, two rapidly growing branches of computer science that enable machines to perform tasks that typically require human intelligence. AI functions are intense and consume substantial resources, so it’s critical that a robot be able to pull computational power away from AI functions when trying to perform another task, and vice versa.
Right now, the type of technology Bradley is developing doesn’t exist, in part because of a dearth of expertise at the intersection of computer science and controlling. He’s uniquely prepared for this work because of his diverse educational background that includes degrees in computer engineering, electrical engineering and aerospace engineering. But there aren’t many other researchers with a similar interdisciplinary perspective that allows them to consider both computation and the design of a robot’s controller, he said.
That’s why, as part of the CAREER project, he will develop a new program of study that would enable Husker students to get a bachelor’s degree in robotics combined with a master’s degree in cyber-physical systems, known as CPS. When launched, the program would place Nebraska among just a handful of universities nationwide — including Vanderbilt, DePaul and Virginia – with CPS degree programs or certificates.
Bradley is also focused on engaging middle school girls in rural Nebraska in CPS and robotics concepts. Inspired by the capacity of his own middle school-aged daughter to do basic computer programming for school projects, Bradley will collaborate with Central Community College to hold a weeklong camp focused on robotics.
And because he believes that you can’t fully engage middle school girls without also talking to their parents and grandparents, who are key players in fostering STEM excitement at home, Bradley will team up with Nebraska’s Osher Lifelong Learning Institute to offer seminars for adults that focus on how CPS concepts can boost drones’ capabilities and allaying concerns about drones’ increasing presence in daily life.
The National Science Foundation’s CAREER award supports pre-tenure faculty who exemplify the role of teacher-scholars through outstanding research, excellent education and the integration of education and research.