Mobile Platform Self-localization in Partially Unknown Dynamic Environments

Localization methods for mobile platforms are commonly based on an observation model that matches onboard sensors measures and environmental a priori knowledge. However, their effectiveness relies on the reliability of the observation model, which is usually very sensitive to the presence of unmodelled elements in the environment. Mismatches between the navigation map, itself an imperfect representation of the environment, and actual robot’s observations introduce errors that can seriously affect positioning. This article proposes a 2D point-based model for range measurements that works with a new method for 2D point matching and registration. The extended Kalman filter is used in the localization process since it is of the most efficient tool for tracking a robotic platform’s configuration in real time. The method minimizes the impact of measurement noise, mismodelling and skidding on the matching procedure and allows the extended Kalman filter observation model to be robust against skidding and unmodelled obstacles. Its O(n ·m) complexity enables real-time optimal points matching. Simulation and experiments demonstrate the effectiveness and robustness of the proposed algorithm in dynamic and partially unknown environments.