Learning-based Adaptive Optimal Control Principles for Human Movements

点击翻译按钮获取中文摘要

The study of human movement, driven by the desire to expand our understanding of the brain, holds great promise for devising new therapies for people who suffer from neurodegenerative disorders such as Parkinson's disease and Huntington's disease. This proposal aims to deepen our preliminary research in learning-based control theory as a new computational principle of sensorimotor control, while, at the same time, validating the proposed theory through numerical simulations and biological experiments. It is an interdisciplinary project that combines tools and methods from reinforcement learning, nonlinear control theory, and adaptive dynamic programming. Rigorous stability and robustness analysis as well as convergence proofs for the proposed learning algorithms will be carried out. A fundamentally novel aspect of the proposal is that attention will be paid to continuous-time dynamical systems described by differential equations as opposed to the conventional models used in the past literature such as discrete-time systems and Markov decision processes (MDPs). Intellectual merit:This interdisciplinary project is aimed at developing and validating robust adaptive dynamic programming as a theory of human sensorimotor learning and control. To this end, there is a great need to develop new results that go beyond the present literature by considering a wider class of continuous-time stochastic systems with both additive and multiplicative noise and strong nonlinearities. Human behavioural experiments are to be designed to test and develop computational models of sensorimotor control. Broader impacts:Even though human movement problems are targeted in this project, it is expected that the findings of this project toward learning-based control theory will be useful for other problems arising from engineering and computational neuroscience such as robotic rehabilitation. The proposal also includes a plan of educational activities centered on student supervision, curriculum development, knowledge dissemination, and collaboration with New York University Center for Neural Science. The PI hopes to organize workshops and invited sessions at major conferences, providing an opportunity to students and junior researchers to interact with leading authorities in automatic control and computational neuroscience.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.