Canonical targets for mobile robot control by deictic visual servoing

This paper extends recent work in deictic visual servoing by specifying the targets needed for general purpose navigation and by designing a local motion controller to drive a mobile robot with respect to those targets. Deictic visual servoing is a recent merger of visual servoing and shared control of mobile robots, providing a powerful new low-level control scheme which is also intuitive for human users. However, until now no method has been presented to identify the targets needed to control general purpose navigation. Since the targets presented here are based on the structure of the environment they are characteristic of the environment and are thus canonical. The targets are the 2D appearance of 3D object features represented in a visibility graph; they are used because a visibility graph indicates precisely the targets needed to control navigation. Previous work on tracking visual targets such as corners and edges suggests this approach is feasible. This paper presents a set of deictic commands capable of driving a robot along segments of a previously planned path, and reports experiments demonstrating the method´s success