Real-time genetic obstacle avoidance controller for a differential wheeled exploratory robot

This paper presents the development of an autonomous differential wheeled robot able to avoid short-distance obstacles using signals from a set of infrared sensors. For the proposed implementation, the situations of imminent collision are solved on line using an adequately designed genetic algorithm (GA). In this way a knowledge database comprising the main set of rules that directly map the sensor information into the engine commands can be developed on the fly. Our experiments proved that a satisfactory behavior can be obtained even in cases when no initial knowledge database, usually obtained previously off-line through simulations, is provided. The implementation uses a MFC5213 Freescale microcontroller. The GA is developed in C language, using the CodeWarrior 7.2.2 IDE and was extensively tested to prove the viability of the proposed solution.