Point-based Scan Matching by Differential Evolution

Nature-inspired metaheuristics have been applied in many different areas and have shown good results in comparison with traditional domain-specific optimization methods. In this work, we investigate the ability of a simple variant of differential evolution to solve 2D scan matching problem. It consists in finding an optimum affine transformation (rotation and translation) between two laser scans (2D pointclouds). Parameters of the affine transformation are in this approach determined by differential evolution. All steps of the proposed algorithm are data parallel and can be efficiently accelerated by massively parallel platforms including mobile graphical processing units. The proposed method was implemented and experimentally evaluated on a test data set. The obtained results show that it achieves a good accuracy and is a promising technique for real-world applications in mobile robotics.