Random Normal Matching: A robust probability-based 2D scan matching approach using truncated signed distance functions

Abstract

This paper proposes a novel, probability-based 2D scan matching approach called Random Normal Matching. It improves the robustness of ohm-tsd-slam - an existing 2D simultaneous localization and mapping algorithm. Using the SLAM's truncated signed distance map representation, a probability field is generated. The probability field is used to determine a pre-transformation for Iterative Closest Point scan matching. This combination results in an accurate and robust 2D SLAM, making it highly suitable for the application in rescue robotics with rough terrain. Test results show that even without processing odometry data, the proposed approach is well competitive with other state-of-the-art SLAM algorithms.