Incremental learning based on extraction of action sequence for autonomous mobile robot

In this paper, an incremental learning algorithm based on extraction of action sequence is proposed. The proposed method extracts action sequences from experience obtained through learning in previously encountered environments and uses them in current environments. The proposed method is constructed by two modules. The first module is an action module that uses neural network and genetic algorithm to adapt to a given task autonomously. The second module is an extraction module that uses self-organized map to solve similar tasks quickly. Experimental simulation to evaluate the proposed method is conducted. The proposed method is applied to two tasks of the autonomous mobile robot and its effectiveness is shown. The first is a steering task including time lag and the second is a sweeping task. In addition, the experiment of a real robot is conducted and it is confirmed that the proposed method can be used for practical usage.