Learning Control Cycles for Area Coverage with Cyclic Genetic Algorithms

Area coverage is a type of path planning that is concerned with finding a pattern of movement that will result in coverage of all parts of an area. Most area coverage planning algorithms assume that the robot can maintain a track over the ground that will result in full coverage in obstacle free areas. This is easily done if the robot can be precisely controlled or has sufficient sensor capability to know the relationship of its current track to its previous one, but simple legged robots usually lack both of these attributes. A means of learning the optimal cycle of turns and straights to produce a full coverage track could greatly improve efficiency. In addition, a system of learning could compensate for the lack of calibration in robot turning systems. In this paper, we introduce a method for learning turn cycles that will produce the tracks required for area coverage. The learning is done using a cyclic genetic algorithm (a form of evolutionary computation designed to learn cycles of behavior). Tests of the simulated robot’s dead reckoning capabilities with the learned cycles show that using CGAs is an effective means of learning for