A Control-Systems Approach to Understanding Human Learning

点击翻译按钮获取中文摘要

The objective of this project is to understand how humans learn to control unknown dynamic systems. Humans possess exceptional learning capabilities that help them control complex systems with virtually no prior information. For example, humans ride bicycles, fly kites, and play with hula hoops. No existing control technique can match a human's ability to learn to interact with a wide variety of uncertain dynamic systems. This project addresses fundamental questions of human learning and control: What control strategies do humans learn? How do humans learn to control unknown dynamic systems? Understanding human learning has a high potential to transform a wide variety of technologies, such as human-interface devices and robotic-assist systems. For example, neurological injuries often lead to impaired motor control. Robotic-therapy devices have demonstrated some success in rehabilitation; however, improved understanding of human learning is necessary to unlock the potential of these technologies. As another example, learning to interact with complex systems, such as orthotic devices and haptic interfaces, can require significant training. Improved understanding of human learning will lead to interactive methods that accelerate learning. The control strategies that humans learn and the processes used to learn them are currently unknown. The predominant human-learning theory in neuroscience is the "internal model" hypothesis, which proposes that humans construct and use models for control. However, evidence in support of the internal model hypothesis is inconclusive. This project offers a new approach to human-learning research that applies principles from control systems to address fundamental questions of human learning and control. Specifically, a series of human-subject-based experiments will be performed to study human learning. First, this project seeks to identify the strategies that humans employ to control dynamic systems. These experiments focus on identifying the strategies that humans use for systems with challenging characteristics, such as nonlinearities, instabilities, nonminimum-phase-zero dynamics, and high relative degree. A novel subsystem identification method will be used to mathematically model the controllers employed by the human subjects. Next, this project aims to identify the learning mechanisms that allow humans to adapt to and control unknown dynamic systems. Human learning mechanisms will be studied by examining how humans learn at different frequencies, by comparing human learning to adaptive control, and by exploring how humans use persistently exciting signals to learn.