Landgren: Exploring the Science of Burritos and Belugas
Peter Landgren embraced inventing at an early age, and innovation continues to drive him at Princeton.
“I loved constructing things with my Dad and seeing how he solved different problems,” recalled the Oregon native, who expects to earn his Ph.D. in Mechanical and Aerospace Engineering in spring 2018. “I find a great many engineering avenues to be fascinating.”
Landgren’s current research avenue involves “cooperative multi-agent decision-making in noisy environments,” which explores the intersection of mathematics and group decision-making among both animals and humans and provides exciting new opportunities for design of networks.
“Humans and animals are remarkably adept at making complicated decisions, and we want to capture some of that magic and recreate it,” he said. “I like to think of our process as examining one small part of the human mind and decision-making capabilities, and then using mathematical tools to better understand what makes these strategies so effective.”
A lot of research has been done to develop decision-making algorithms that work in the “noisy” real world. For example, researchers have shown how a single agent can make an optimal sequence of choices among a set of uncertain alternatives, in what is known as the multi-armed bandit problem. Landgren’s work digs deeper.
“My specific innovation comes from the ‘cooperative multi-agent’ part,” he said. “Multi-agent simply means that there are multiple actors all making decisions at once, and cooperative means that they are working together to reach the best decision for all rather than competing as individuals.”
As an example, he describes a group of friends who meet at Qdoba on Nassau Street and want to buy the best burrito the Mexican restaurant has to offer. Collectively, they have to answer many questions. “How do you incorporate information from your different friends when making a decision about what to order?” he says. “Do you trust certain friends more? How do you weigh their different experience levels? This is a case where humans excel and complete the task with ease.
“My goal is to mathematically express these processes so we can do three things. First, doing so allows us to more accurately model and describe animal systems. For example, we could use it to model foraging deer and how they move. Second, a mathematical model allows us to enable robots to make principled decisions in the way humans do. For example, if you have a robot searching for an oil slick in the ocean, the robot needs to make decisions about where to go that trade off lots of conflicting goals. Third, by expressing behavior mathematically we can draw rigorous mathematical conclusions about the systems we’re studying.”
One practical application of Landgren’s research is the rapidly expanding world of connected devices ("smart" home devices, phones, etc.) and future robots (self-driving cars, for instance). “In order to provide a benefit, these devices will need to be able to learn and make hard decisions without human intervention, as well as communicate with other devices,” he said. “Our work could provide the mathematical framework for analysis and implementation of these complex networks.”
Landgren’s advisor, professor Naomi Ehrich Leonard ’85, said, “Peter’s highly creative work provides a new principled means to investigate, and leverage for design, the mechanisms that explain how networks of decision-making agents handle uncertainty. For instance, he can systematically predict better performing network topologies or which agent in the network will make the most rewarding decisions. Peter also has great talent and passion for design and robotics, and he is making exciting advances with a network of decision-making robotic vehicles.”
Leonard, the Edwin S. Wilsey Professor of Mechanical and Aerospace Engineering, and director of Princeton’s Council on Science and Technology, added that “Peter has excelled as a teaching assistant in the MAE junior design course in which student teams designed, built, and tested underwater robots and in the new course ‘Transformations in Engineering and the Arts.’ This summer Peter created and led a wonderful multi-day, hands-on robot-making workshop for students and postdocs.”
At Leonard’s Dynamical Control Systems Laboratory, he manages a 20,000-gallon cylindrical pool where researchers can test algorithms using underwater robots called Belugas. In 2014, he used the robots to demonstrate some of the new decision-making algorithms (click here to see a video presentation). He currently is working on making smaller, wireless versions of Belugas that will allow a wider variety of testing.
Landgren was awarded a National Defense Science and Engineering Graduate (NDSEG) Fellowship that started in September 2015. He received his bachelor’s degree in physics from Whitworth University in Spokane, Wash.