Vector Distance Transform Maps for Autonomous Mobile Robot Navigation

Publication Type:
Issue Date:
Full metadata record
Robot localization, where the position and the orientation of a mobile robot is estimated based on an 𝘢 𝘱𝘳𝘪𝘰𝘳𝘪 map, is a fundamental problem in autonomous mobile robot navigation. In this thesis, we present how environments can be represented using the vector distance transform for mobile robot localization, and show that it is a superior representation compared to alternatives such as occupancy grid maps and other distance transform variants such as the unsigned distance transform and the signed distance transform. We propose an approach based on non-linear least squares optimization for robot localization on environments represented by vector distance transform maps, that also captures the uncertainty of the position estimate. We also propose an approach based on extended Kalman filter for robot localization on environments represented by vector distance transform maps. Using simulations, public domain data and real world experiments, the proposed localization approaches are evaluated and compared with existing localization techniques in multiple robot platforms including both ground and aerial robots navigating in 2D and 3D environments using a series of sensors including LiDARs and cameras. We propose an information filtering based approach for mobile robot localization where a single endpoint range sensor can be used to accurately localize a mobile robot in motion by rotating it in a direction that will improve its position estimate and the corresponding uncertainty. The proposed approach is evaluated using simulations and real world experiments. Occupancy grid maps, the most popular type of map for robot localization using a range-bearing sensor, are assumed perfect when used for localization. It is a discrete representation of the environment and the probability of occupancy of each cell is independent from its neighbors. Given a set of robot poses and corresponding range-bearing measurements, both incorporated with uncertainty, a representation of environment based on the vector distance transform using non-linear least squares optimization that is both continuous and uncertain is proposed for robot localization. Simulations are used to demonstrate the accuracy of the map and its uncertainties.
Please use this identifier to cite or link to this item: