Signal processing, identification and control in switched non-linear systems

点击翻译按钮获取中文摘要

This research concerns the analysis and identification of control systems undergoing sudden parametric changes, to enable selection of alternate control schemes. Such systems exist in many physiological contexts,e.g., those used by the brain to control movement. Since usual assumptions of constant characteristics (stationarity) are violated, special algorithms are required to allow consistent estimation of system dynamics in the presence of sudden parametric changes and variable initial conditions.Hence, the first part of this research addresses the issue of developing software tools in MATLAB modules for easy distribution, which will have 2 purposes. First, fast classification of a system's response modes into potential configurations, using reduced models. Second, fitting system models to each of the potential modes using identification methods which take into account short response segments and variable initial conditions. Results here are finding direct application in CIHR funded research on clinical diagnosis and modelling of binocular control.The second major component examines the strategies deduced from the modelling of physiological motor control (CIHR supported work) and is porting so-deduced novel strategies to robot orientation systems and sensory fusion issues. For example, we have successfully implemented fast target pursuit in a binocular system using dynamic mode switching, and can coordinate motion of a supporting "neck" without additional controllers. The biologially-evoked approach offers the promise of faster and more accurate control, despite structurally simple controllers. Complex behaviour relies on internal switching which of course introduces non-linearities which must be addressed in the design and analysis aspects of the projects. This is being extended to allow innovative methods of sensory fusion and 'reflex' adaptation with context in both robotic heads and arms.