Altitude offset constraint for mobile robots’ localization

Localization is a fundamental element for mobile robotics. Using GNSS (Global Navigation Satellite Systems) for global position retrieving, odometry data (visual, laser or inertial) for pose estimation, feature matching for loop closure, and the use of HD (high-definition) maps are setups commonly used in self -driving vehicles' localization system. However, these setups constrain the vehicle to either rely on the GNSS, which can lack precision due to several factors including urban canyons, or dense vegetation on the surroundings; or on HD maps, usually dependent on constant map updates and, thus, requires driving the city all over again. Besides, localization based on odometry-only tends to drift over time, leading to imprecise localization in the long run. This paper proposes fusing odometry, compass, and altitude offset measurements for 2D pose estimation through a particle filter, given an elevation map of the environment. This approach does not need GNSS devices or internet access during navigation. We performed simulated experiments as this method's proof-of-concept as an alternative for global pose estimation. Despite simulated, the experiments demonstrate coherent convergence over relatively large tracks with realistic sensor noise. The source code of this project is available online and integrated with the ROS (Robotics Operating System) framework.