The robot navigation problem is becoming one of the most important Problems that must be solved as automated and flexible manufacturing techniques become more advanced. This problem arises when a simulated robot searches for an optimal path from a given starting position to a destination position while avoiding the obstacles for an arbitrary navigation task. In particular, the difficulties of this problem are compounded in a multiagent environment in which there are many robots, each having individual navigation tasks in the same navigation area concurrently. In this paper, a new approach to the autonomous solution of the robot navigation problem is proposed. To realize an autonomous robot navigation mechanism, we applied a classifier system. This is a machine learning system in which simple string rules are learned to improve the performance in an arbitrary environment. Thus, each agent has its individual classifier system and searches for an individual solution with interaction among agents. Based on the proposed method, a robot navigation system was constructed and numerical experiments were carried out. The results of these experiments demonstrate the usefulness of the proposed method.