Mimicking Mother Nature to maximize multi-agent systems
New faculty in autonomous systems combine expertise to build multi-agent UAV systems that are small but mighty
The cleverest engineer might take a lifetime designing a process that comes naturally to the simplest animal. How to locate food, for example, or how to move in a flock.
Luis Rodolfo Garcia Carrillo is one of a number of faculty hired recently with expertise in autonomous systems.
Luis Rodolfo Garcia Carrillo is an expert in control systems, multi-agent systems and computer vision. But he spends a good part of his day thinking about things like how ants swarm, how bacteria like E. coli locate food and how lions identify new migration destinations. Garcia Carrillo is part of a growing movement in engineering to draw on nature’s design principles to solve engineering problems.
“There is a new trend in robotics where we are trying to do complex things with simple ideas, so people are taking inspiration from nature in engineering,” Garcia Carrillo said.
The complex problem that Garcia Carrillo is tackling is how to program groups of small aerial vehicles to work together, autonomously, to accomplish a task.
“Most of my research is inspired by nature, by biology,” he said. “For example, ants are not powerful on their own but when you put all of them together they can do amazing things. This is what we are trying to mimic in our multi-agent systems.”
Optimizing small, multi-agent systems
For Garcia Carrillo, smaller UAVs working together are the future of autonomous systems. Smaller aerial vehicles are cheaper, more fuel efficient and safer. They also offer redundancy – each UAV can carry different sensors and the loss of one UAV to a crash or malfunction doesn’t lead to failure of the entire mission.
“When we have teams of agents that collaborate, they cooperate to accomplish a task faster and in a more efficient way,” Garcia Carrillo said. “Some people are interested in building one super UAV, but what happens if that super UAV crashes? You are going to lose everything. Here if one of these crashes, you are losing just one small part of your multi-agent system."
In theoretical simulations, researchers are able to move thousands of UAVs in concert, but in practice, Garcia Carrillo is working with groups of three or four UAVs. He’s focused on developing mathematical models and corresponding programs for three key behaviors: maintaining a formation, avoiding obstacles and target tracking.
Using these models, Garcia Carrillo’s UAVs could locate a lost hiker, for example, or track an animal through the wilderness. Garcia Carrillo’s UAVs use a rotating motion to hone in their target. The rotating motion is another trick he picked up by observing the natural world. E. coli bacteria, with their extremely limited perceptual systems, use a rotating motion to locate nutrients. Copying this rotation allows a UAV to scan a wide area and zero in on a location with the strongest signals.
UAV systems ready for field applications
To build a functional system, Garcia Carrillo is collaborating with Kostas Alexis, an assistant professor of computer science and engineering. In order for Garcia Carrillo’s robots to execute an obstacle avoidance program, for example, they first need to be able to sense obstacles in the environment. Likewise, tracking a target requires locating and locking onto the target. Those perceptual problems are the expertise of Alexis, and combining that with Garcia Carrillo’s background in control systems enables the duo to design a functional UAV system – one that Garcia Carrillo believes is ready to be used in the field, at least from a technical perspective. Regulations governing the use of UAVs are still in the works, and the state of Nevada as a Federal Aviation Administration test site is helping to test technology and develop policy that can guide the use of UAVs in commercial, research, and search and rescue scenarios.
“We have most of the hard work and the theoretical background that we need,” Garcia Carrillo said. “Government regulations and acceptance of society are the main challenges right now.”
Public perception of autonomous systems seems to swing between two polar extremes – on the one hand, fear about potential misuses and invasions of privacy, and on the other, the immense popularity of UAVs on the hobbyist market.
Garcia Carrillo strives to carefully situate his work within those perceptions. He appreciates the power that his flying robots have to attract students to study engineering, but he cautions prospective students to treat autonomous systems as complex technology, not a toy. Programming a UAV to find a missing hiker is significantly more complex than buying a drone on Amazon and flying it in your backyard, Garcia Carrillo said. Students drawn to the field by the fun of flying sometimes become frustrated by the time required to program and troubleshoot a functional UAV.
“Be patient and be prepared to spend a lot of hours dealing with things that are not working the way the math says they are going to work,” he said. “You don’t have to be a genius. You just have to be a hard worker.”
Keeping their eyes on the skies
Kostas Alexis, Assistant Professor, Computer Science and Engineering
In order to be capable of autonomous navigation and task completion, robots need advanced perception systems to operate in the world around them. Alexis’ research aims to develop smart robots capable of tasks like infrastructure inspection, surveillance and monitoring, and precision agriculture by using advanced perception techniques integrated with navigation and control.
Richard Kelley, Chief Engineer, NAASIC
As chief engineer for the Nevada Advanced Autonomous Systems Innovation Center, Kelley collaborates with researchers and industry to develop new technologies for aerial and ground-based autonomous systems. Currently, Kelley is developing software in collaboration with NASA that would coordinate air traffic for low-altitude, autonomous aircraft.
Logan Yliniemi, Assistant Professor, Mechanical Engineering
Yliniemi’s research exists at the interface of the fields of multi-objective optimization and multi-agent systems. He has developed computationally inexpensive algorithms that are able to simulate the coordinated movement of a large number of agents with one-tenth the computation cost of other comparable methods.