Genetic programming for control

Staff Mentor: M. Mazo Jr. (Manuel)

Other Mentor(s):

C.F. Verdier, prof. R. Babuska


Machine learning; Intelligent control; Correct-by-design controller synthesis


Automated controller synthesis approaches for nonlinear systems with complex tasks, such as reinforcement learning and abstraction and (bi)simulation, result in controllers that can take the form of enormous tables or expressions. While useful, these controllers cannot always be implemented in embedded hardware with limited memory. Furthermore, these controllers are very though to be interpreted by humans. Finally, controllers from e.g. reinforcement learning approaches lack formal guarantees.

To overcome these limitations, we want to take one of these synthesized controllers and use genetic programming to find a compact analytic expression that can be implemented and/or further analyzed.

If you are interested, please contact

© Copyright Delft Center for Systems and Control, Delft University of Technology, 2017.