【摘要】 为了提升复杂环境下的协同导航精度与鲁棒性，提出了融合深度学习与多模型滤波的无人车协同导航方法。将深度学习网络与交互式多模型预测算法（IMM）深度融合，并融入协同导航系统的设计中，实现了数据层面的高效融合与互补，显著增强了导航系统在复杂、高动态环境中的适应性与精确性。复杂环境下实车实验结果表明，在200 m的测试路径上，所提方法协同导航系统最大误差为0.3 m，较最初的激光/惯性协同导航方法提升了27.9%，验证了所提方法在卫星拒止环境下协同导航系统的显著优势与工程实用价值，为未来智能无人系统在拒止条件下的自主导航提供了有力的技术支撑。更多还原

【Abstract】 In order to improve the accuracy and robustness of cooperative navigation in complex environments, an unmanned vehicle cooperative navigation method integrating deep learning and multi-model filtering is proposed. The deep learning network is deeply integrated with the interactive multiple model(IMM) prediction algorithm and incorporated into the design of the cooperative navigation system.Efficient data-level integration and complementarity have been achieved, significantly enhancing the adaptability and accuracy of the navigation system in complex and highly dynamic environments. To validate the effectiveness of the proposed algorithm, real-vehicle tests are conducted in complex environments, where the maximum error of the cooperative navigation system is merely 0.3 m over a 200 m test path, which is increased by 27.9% compared with the original laser/inertial cooperative navigation method. This result confirms the significant advantages and engineering practical value of the proposed method in the cooperative navigation system under satellite rejection environments, providing robust technical support for future autonomous navigation of intelligent unmanned systems under extreme conditions.更多还原