Adaptive fuzzy systems for target tracking

Compares fuzzy and Kalman-filter control systems for real-time target tracking. Both systems performed well in the presence of additive measurement noise. In the presence of mild process (unmodelled-effects) noise, the fuzzy system exhibited finer control. The authors tested the robustness of the fuzzy controller by removing random subsets of fuzzy associations or 'rules', and by adding destructive or 'sabotage' fuzzy rules to the fuzzy system. They tested the robustness of the Kalman tracking system by increasing the variance of the unmodelled-effects noise process. The fuzzy controller performed well until over 50% of the fuzzy rules were removed. The Kalman controller's performance quickly depreciated as the unmodelled-effects variance increased. The authors used unsupervised neural-network learning to adaptively generate the fuzzy controller's fuzzy-associative-memory structure. The fuzzy systems did not require a mathematical model of how system outputs depended on inputs. >