Please use this identifier to cite or link to this item:
|Title:||A framework of probabilistic 3D map fusion for collaborative robots||Authors:||Yue, Yufeng||Keywords:||DRNTU::Engineering::Electrical and electronic engineering||Issue Date:||2019||Source:||Yue, Y. (2019). A framework of probabilistic 3D map fusion for collaborative robots. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||Utilizing multi-robot systems for environmental mapping, inspection or surveillance tasks have found significant attention over the past few years due to the inherent advantages of robustness and efficiency. Due to its limited sensing capabilities and computational power, each robot only has partial information of the surroundings. To maintain a comprehensive understanding of the environment under limited communication bandwidth, fusing local 3D maps generated by individual robots to a globally consistent map is a critical challenge in multi-robot mapping missions. Most of the existing approaches to collaborative mapping are subject to certain challenges such as uncertainty modeling, estimating relative transformation, information integrating, inconsistency detection, and communication management. This thesis proposes a systematic probabilistic multi-robot mapping framework that takes these challenges into account. First of all, a hybrid probabilistic and point set registration approach for fusion of 3D occupancy grid maps is proposed. Previous approaches ignored the foremost property of the 3D occupancy grid map, which is the occupancy probabilities of individual voxels. To incorporate the probability information, a novel error metric and optimization strategy need to be proposed. The Occupancy Iterative Close Point (OICP) algorithm combines positional coordinate with occupancy probability to register point cloud extracted from the 3D map. The weight between the positional error and the probability uncertainty is controlled by modeling errors using Shannon entropy and positional misalignment. In addition, an environment measurement model is applied to evaluate the relative transformation and a relative entropy filter is used to integrate the map dissimilarities. The accuracy of the proposed algorithm is evaluated using maps generated from both simulated and real-world environments, where the result is shown to generate more consistent global maps. Secondly, submap-based probabilistic inconsistency detection for multi-robot mapping is investigated. The primary goal of employing multiple robots in active mapping tasks is to generate a globally consistent map efficiently. However, all the current approaches focus on single robot inconsistency detection. The challenge comes from designing a framework that is able to detect inconsistency at single robot and multi-robot levels. In this thesis, a novel multi-level approach is introduced to measure the full 3D map inconsistency, where submap-based tests are performed on both single and multi-robot levels. The conformance test based on submaps is done by modeling the histogram of the misalignment error metric into a truncated Gaussian distribution. Besides, the detected inconsistency is further validated through the OICP map registration algorithm proposed in this thesis. The accuracy of the proposed method is evaluated using submaps generated from both indoor and outdoor environments, where the result illustrates its usefulness and robustness for multi-robot mapping tasks. Thirdly, hierarchical probabilistic fusion framework for matching and merging of 3D occupancy maps is studied by extending the positive results of integrating probabilistic map information in OICP. Existing approaches have relied on point set registration based algorithms for map matching. However, they have not provided a comprehensive analysis of uncertainty modeling, map matching, transformation evaluation, and map merging. In addition, high-level geometry features should be incorporated if available. The thesis addresses these issues by proposing a novel hierarchical probabilistic map fusion framework for 3D occupancy grid maps. Before the fusion of maps, map features and their uncertainties are explicitly modeled and integrated. For map matching, a two-level Probabilistic Map Matching (PMM) algorithm is developed to include high-level structural and low-level voxel features. The statistical testing model is applied to evaluate the matching result, where the acceptance threshold is auto-tuned based on the overlapping between the maps. The results verify the accuracy and efficiency of the proposed 3D occupancy map fusion approach and exhibit the improved convergence in these scenarios. Finally, a general probabilistic framework for collaborative 3D mapping using heterogeneous sensors is proposed. There have been few works on 3D map fusion of heterogeneous map data types in arbitrary (structured/unstructured) environments, especially merging of a sparse map with dense maps. In the thesis, a general probabilistic framework is proposed to address the integrated map fusion problem, which is independent of sensor types and SLAM algorithms. The framework provides the theoretical basis for computing the relative transformations (map matching) and merging probabilistic map information (map merging), which enables the team of robots to work with heterogeneous sensor types. Since maps produced by heterogeneous sensors possess diverse physical properties, an Expectation-maximization (EM) based map matching algorithm is proposed that automatically decides the numbers of voxels to be associated. To distribute the maps among robots, a time-sequential map merging framework is developed that makes fusing maps from distributed multi-robot system efficient. In addition, the uncertainty embedded in the map and caused by the relative transformation are also propagated and integrated. The proposed approach is evaluated in real-world experiments in various environments with heterogeneous sensors, which shows its accuracy and versatility in 3D map fusion for multi-robot mapping missions.||URI:||https://hdl.handle.net/10356/89844
|DOI:||10.32657/10220/47722||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.