12 units
Classical control techniques perform well when the system is linear. However, many interesting and practical systems are inherently nonlinear. In addition to nonlinearity, the plant parameters may not be completely known a priori. Furthermore, the parameters may change overtime. Neural networks are a tool that can be used to compensate for both nonlinearities and online changes in the plant. The objective of this course is to present an in-depth treatment of neural networks for estimation, controller design and stability analysis. The specific topics include: Lyapunov stability theory, uniform boundedness, multi-layer neural nets, backpropagation, radial basis functions, system identification, direct and indirect adaptive control strategies and robustness enhancing techniques. Specific applications will include control of robots and aircraft. Additionally, we will present a brief overview of some recent developments including stochastic neural network control and time-delay estimation techniques.
Prerequisites: Ordinary differential equations and 18-470 -- Fundamentals of Control. It is helpful, but not required, to have taken or to take concurrently: 18-771 -- Linear Systems and 16-264 -- Humanoids.
This course will be broadcast from the Silicon Valley Campus to Pittsburgh
Last updated on October 29, 2009
S10
Please note that the course history information is incomplete and/or may reflect different courses offered under the same course number.