Application of multi-valued AdaBoost to control of autonomous mobile robot

The authors proposed a tuning method, which made use of good generality of rule teaching and high accuracy of playback-type teaching. This method utilized a feature of boosting algorithm that constructed a strong classifier by combining multiple weak classifiers. However, the obtained controller had a limit in accuracy because this method was based on two-valued classifiers. This paper presents a modified boosting algorithm which is capable of aggregating multi-valued weak classifiers, and of identifying accurate and general control rule. Simulations of pass passage problem of a mobile robot are carried out to test the proposed method. It is shown that thus method acquires control rules more accurate than those obtained by the conventional two-valued boosting algorithm, and in the limited taught space, generalization of the final hypothesis by the proposed method is demonstrated.