Neurofuzzy approaches to intelligent collision avoidance problems in (Semi)Autonomous transportation

Model based methods for state estimation and control of linear systems is well established and applied. In practice the systems are non-linear, stochastic, temporal and only partially known. An alternative approach based upon empirical data based methods, which incorporate prior knowledge utilising linear additive 'nonlinear' models based upon neurofuzzy algorithms are introduced. For control and tracking, there is a surfeit of techniques, which could be applicable to non-linear problems if appropriate linearisation is achieved; here various forms of local neurofuzzy networks are discussed via a class of adaptive neurofuzzy networks. It is shown that they have good approximation, convergence and stability properties, as well as parametric parsimony making them ideal in control and tracking. It is then shown how these algorithms have been successfully applied to collision avoidance problems in cars, helicopters and ships.