This paper presents a multi-agent approach to developing a flexible real-time control system for an autonomous mobile robot The main purpose of this study is to integrate heterogeneous algorithms and functions onboard the robot, while still to guarantee a reasonable responding time. The balance between generality and flexibility is also addressed. This strategy provides a framework for developing complex intelligent machines through teamwork of several people. The proposed control system has been implemented on an experimental robot based on a real-time Linux platform. Practical experiments show that the robot successfully navigates through complex environments by combining visual tracking, obstacle avoidance, and command receiving behaviors in a single system.