Skip to main content

Online learning of neural networks for real time control


Neural control strategies have been the subject of much attention of the engineering research community in recent years. However such control strategies are difficult to analyse theoretically. Subsequently few statements exist regarding the stability of closed loop systems containing neural controllers. The proposed project aims to build on the results of the HCM project of the same name and develop adaptive non-linear and/or time varying neural control strategies which are stable in a closed loop. Such algorithms would then be suitable for safety critical applications (aerospace, automobile). The developed adaptive algorithms will be evaluated on complex safety critical applications within the Daimler-Benz group (Deutsche Aerospace, Mercedes, AEG, Debis). Specifically the work package consists of developing frequency domain stability analysis for time varying linear systems, the development of a robust stable nonlinear neural controller which can tolerate plant uncertainty and the development and refinement of a neural adaptive controller.

Funding Scheme

RGI - Research grants (individual fellowships)


Daimler-Benz AG
91B,alt Moabit
10559 Berlin

Participants (1)

Not available