Efficient 2D and 3D map representations of both static and dynamic, indoor and outdoor environments are crucial for navigation of driving and flying robots. In this paper, we propose a fast and accurate approach for 2D and 3D Normal Distributions Transform (NDT) mapping based on indexed kd-trees. Similar to other approaches, we also model free space, which allows us to obtain occupancy probabilities. Additionally, we provide optional visibility based updates to enhance map consistency in case of noisy data, e.g. from stereo cameras. Unlike other available implementations, our approach is natively applicable to large-scale environments and in real-time, because our maps are able to grow dynamically. This also offers applicability to exploration tasks. To evaluate our approach, we present experimental results on publicly available datasets and discuss the mapping efficiency in terms of accuracy, runtime and memory management. As an exemplary use case, we apply our maps to Monte Carlo Localization on a well-known large-scale dataset.
[doi]
@inproceedings{SchulzIROS2018,
title = {{Efficient Map Representations for Multi-Dimensional Normal Distributions Transforms}},
author = {Cornelia Schulz and Richard Hanten and Andreas Zell},
booktitle = {2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
pages = {2679--2686},
address = {Madrid, Spain},
year = {2018},
month = {October},
days = {1--5},
abstract = {Efficient 2D and 3D map representations of both static and dynamic, indoor and outdoor environments are crucial for navigation of driving and flying robots. In this paper, we propose a fast and accurate approach for 2D and 3D Normal Distributions Transform (NDT) mapping based on indexed kd-trees. Similar to other approaches, we also model free space, which allows us to obtain occupancy probabilities. Additionally, we provide optional visibility based updates to enhance map consistency in case of noisy data, e.g. from stereo cameras. Unlike other available implementations, our approach is natively applicable to large-scale environments and in real-time, because our maps are able to grow dynamically. This also offers applicability to exploration tasks. To evaluate our approach, we present experimental results on publicly available datasets and discuss the mapping efficiency in terms of accuracy, runtime and memory management. As an exemplary use case, we apply our maps to Monte Carlo Localization on a well-known large-scale dataset.},
doi = {10.1109/IROS.2018.8593602},
}