Decentralized collaborative localization in urban environments with inter-agent ranging and 3D-mapping-aided (3DMA) GNSS

Collaborative;Localization;3D-mapping-aided GNSS;Shadow Matching;Intelligent Urban Positioning;Multi-Agent;GPS;UWB;Ranging;UAV;Decentralized Collaborative Localization;Robot Network;Urban Navigation

Tanwar, Siddharth ; Gao ; Grace X

关键词: Collaborative;    Localization;    3D-mapping-aided GNSS;    Shadow Matching;    Intelligent Urban Positioning;    Multi-Agent;    GPS;    UWB;    Ranging;    UAV;    Decentralized Collaborative Localization;    Robot Network;    Urban Navigation;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/104791/TANWAR-THESIS-2019.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

In recent times, there has been an increasing number of applications of autonomous systems such as unmanned aerial vehicles (UAVs) and self-driving cars in urban environments for tasks such as transport, delivery, photography, surveillance and search and rescue. Global Navigation Satellite System (GNSS) navigation in urban environments is prone to error sources such as multipath and signal blockage due to buildings in the environment. However, if we consider several agents, then the localization capabilities may differ among the agents due to reasons such as heterogeneity of sensors, different view of the sky, the structure of buildings around agents etc. This variety can be leveraged to localize all the agents better if they cooperate with each other. Collaborative localization (CL) is a way to aid navigation in a multi-agent system. However, CL algorithms face challenges such as scalability, robustness to noisy sensor data and single points of failure, and operability despite limited inter-agent communication. Additionally, a 3D Map of the environment can be used to predict and account for the effects of multipath and signal-blockage. Shadow Matching has shown high potential in 3D-mapping-aided (3DMA) GNSS navigation for a single agent in urban environments, but it suffers from errors such as ambiguity which cannot be reliably mitigated without external sensing and integrated systems. Shadow Matching naturally compliments range-based GNSS algorithms and is ideally used alongside it. This integration, however, is not guaranteed to reduce ambiguity and may in fact increase it. In this thesis, we present three novel decentralized collaborative localization methods for navigation of multiple agents in an urban environment. The proposed frameworks are applicable to sparsely connected networks and information exchange is limited to only those agents which obtain relative inter-agent measurements. Furthermore, we allow the agents to carry out the updates asynchronously.First, we present an algorithm which allows fordeep coupling of agents' GPS measurements with inter-agent ranging measurements. We build this work upon a decentralized extended Kalman filter based collaborative localization framework and take advantage of the variable visibility of the sky for different agents. We propose a methodology for relaying satellite information between agents to augment the set of visible satellites on each agent with virtual satellites, thereby providing more constraint equations to each agent. The proposed method is validated on real world dataset involving an aerial vehicle, ground agents, and several range-only sensors.Next, we incorporate the 3D city map and present a novel snapshot algorithm which couples 3DMA GNSS measurements with inter-agent ranging modality. We build this upon the Intelligent Urban Positioning (IUP) 3DMA algorithm which uses Shadow Matching (SM) and Likelihood-based 3DMA Ranging (LB-3DMAr) algorithms. The introduction of multiple agents equipped with ranging sensors in the framework enables ambiguity error mitigation (which is a source of error in both SM and LB-3DMAr) and further improves accuracy (by introducing an additional sensor measurement). This method works by constraining an agent's probability distribution using its neighbors in a discretized grid of position hypothesis. Finally, we extend the above method to a multi-epoch variant. Unlike the snapshot algorithm, this allows for a mechanism to account for temporal correlations between poses for each agent. This enables the proposed methodology to further improve ambiguity error mitigation and localization accuracy by preventing jump discontinuities at subsequent time steps and introducing an additional sensor measurement (in the form of pseudorange rate measurements). The method adapts the discretized Bayesian filter to associate temporal correlations between agent states. These snapshot and multi-epoch methods are validated on simulated datasets in an urban area of Champaign, Illinois, with multiple agents in a variety of scenarios. We demonstrate the improved performance in terms of positioning accuracy and ambiguity mitigation. We also analyze the impact of network connectivity, and size of network on positioning accuracy.