Intelligent robotic systems utilize sensory information to perceive the nature of their work environment. Of the many sensor modalities, vision is recognized a one of the most important and cost-effective sensors utilized in practical systems. In this paper, we address the problem of designing vision systems to perform a variety of robotic inspection and manipulation tasks. We describe the nature and characteristics of the robotic task domain and discuss the computational hierarchy governing the process of scene interpretation. We also present a case study illustrating the design of a specific vision system developed for performing inspection and manipulation tasks associated with a control panel.