This paper provides an architecture of mobile robots for the cellular robotic system. The cellular robotic system (CEBOT) is an autonomous distributed robotic system composed of a number of robotic units called "cells." Since the CEBOT has variable structure, flexibility and extendibility are required in the control system for the CEBOT. To design the mobile robots for the CEBOT, we selected a parallel processing control system, with which it is easy to add new rules. In this paper we propose a method to integrate multiple processes for decision making on the behavior of the mobile robots. We define two relation matrices between the processes : a priority relation matrix and an interest relation matrix. Interprocess conflicts are solved by coordination between the priority relation matrix and the interest relation matrix. We also show a learning method to obtain the most suitable priority relation matrix.