Aimed at the problem of traditional robots learning offline, taskspecific, inability to expand intelligence in the learning process, poor realtime performance and adaptability, based on the ideas of cognitive robot, neurobiology, and cognitive science, a robot visual selfdevelopment learning algorithm was proposed mimicing the dopamine regulation mechanism of the human brain. The hippocampus and prefrontal neural circuits of human brain were simulated, and the learning process was regulated by using dopamine to realize the visual learning ability mimicing human. The experimental results show that using the algorithm, selfdevelopment learning of visual images can be effectively realized and nonspecific tasks can be completed. It has the characteristics of high recognition rate, good realtime performance and strong adaptability.