This paper presents a new type of autonomous robotic fish with vision-based target tracking and obstacle avoidance. The robotic fish can sample images through its eye and recognize the meaningful color patch via a special vision algorithm. Specifically, an ARM9-centered microcontroller in conjunction with a CMOS camera for image acquisition is developed to effectively handle local visual information serving as the basis of underwater navigation. Meanwhile, a wireless interactive control console is devised, facilitating two-way information transfer. It can access the environmental information captured by the robotic fish in real time and display the obtained results in a visual format online. Experimental results demonstrate the effectiveness of the proposed scheme.