NSF GRFP recipient Kyle Horton, a second-year ESE PhD student, co-advised by Professor Nikolai Matni and Professor George Pappas, has research interests that range from data-driven controllers with online data collection to hierarchical control.
Horton’s undergraduate work from the University of Wisconsin-Madison informed the NSF GRFP Fellowship application, which focuses on the trade-offs between exploration and exploitation in data-driven controllers with online data collection.
“Oftentimes, the true dynamics of systems of interest, aircraft or robotic manipulators for example, are not known due to the complex physics at play,” Horton explains. “Thus, many methods have been developed for using collected system data to control the system without needing knowledge of the dynamics.”
Horton cites an airplane to illustrate the above argument. “When collecting data to use for a data-driven controller, it would be preferable if you did not crash the airplane in the process, but by not allowing certain risky actions to be taken, data is not able to be gathered about the effect of these particular actions,” Horton explains. “So ,the question to answer is then, how much efficacy of the controller is lost by not gathering this risky data, and how much risk is worth taking to gather better data?”
As is the case for many doctoral students, Horton’s research trajectory has shifted away from data-driven controllers to hierarchical control during his time at Penn. According to Horton, the core concept of hierarchical control divides complex tasks such as a robot picking up an object “into distinct control tiers to balance speed and accuracy.”
The two tiers, Horton explains, include both high-level and low-level controllers. The high level controller “plans abstract, long-horizon trajectories without focusing on granular environmental variables.” Whereas the low-level controller, executes fine grained adjustments to compensate for factors such a wind using PID controllers (Proportional-Integral-Derivative controller).”
The problem that Horton plans to address in his thesis research? While researchers have a thorough understanding of both controllers, “they do not fully understand how these levels interact or how to optimize tradeoffs so they play nicely together,” Horton explains.
When Horton isn’t working, he enjoys reading non-fiction, which includes history, math, and general science. In addition, he enjoys rock climbing, running, and gaming.