Adaptive gain scheduling with modular models

A modular representation system is combined with indirect adaptive control techniques to obtain an adaptive form of gain scheduling, used for control of unknown piecewise linear systems. Learning takes place online, i.e. at the same time as control is being effected. A probabilistic approach is taken for learning of the different linear dynamics and their scheduling, and also for generating the control signal. This is done by utilising the information from a bank of Kalman filters