Generation of multiple cubic spiral paths for obstacle avoidance of a car-like mobile robot using evolutionary search

Cubic spiral is a curve whose direction is a cubic polynomial of arc length. It has been used as a path primitive for nonholonomic car-like mobile robot with restricted turning capability in free environment [1], [2]. In this paper, we investigate the use of cubic spiral segments for car-like mobile robots moving from a given start configuration to a given goal configuration in an environment cluttered with known static obstacles. We propose a parallel genetic algorithm (PGA) based on the island model to generate multiple collision-free paths in parallel via sets of a predefine number of intermediate configurations connected by collision-free cubic spirals. The architecture of PGA is simple: it consists of multiple islands, where subpopulation of each island runs a simple genetic algorithm independently with the same fixed-length real-coded chromosome representation of cubic spiral path segments, basic genetic-operators, one subgoals manipulation operator, and maintains diversity of subpopulations by synchronous migration every certain number of generations among the islands and a Big-bang activity. The generation of multiple collision-free paths by PGA is via the following strategy: (i) the generated collision-free paths are similar to an already found collision-free path, then they are eliminated and replaced by randomly generated paths for later evolution, (ii) otherwise, they are extracted as new alternative collision-free paths. Simulation results of mobile robot for a set of different start orientations show the effectiveness of this evolutionary approach to evolve a more rich set of collision-free paths.