Laser trail shape identification technique for robot navigation based on genetic programming

This paper proposes a meta-heuristic image processing application for mobile robot navigation. It classifies figures that are drawn on a wall by hand with a laser pointer. Image processing technique extracts optical flow of the laser beam trail, which represents vectors along edges of shapes. Genetic programming learns geometric characteristics of laser trail shapes and creates classification algorithm. Three typical figures, such as a circle, a triangle, and a square, are evaluated and identified in high accuracy. We have investigated the effects of genetic programming parameters on the performance of shape identification. As a result, proposal system makes it possible to command robots by easy and intuitive action of drawing a figure only with a laser pointer.