In many transport applications, one of the safety critical function is the localization. This is all the more true for land transport applications such as autonomous vehicles. While the democratization of satellite positioning systems, such as GPS, Galileo, Beidou or Glonass, has made it possible to consider a global solution applicable anywhere in the world, the principle of positioning by receiving signals from satellites more than twenty thousand kilometers away shows limits when they are confronted with disturbances related to the environment close to the receiver. However, for these safety-critical applications, the requirements are strong and sometimes even conflicting. The developed function must meet a defined level of precision, availability, continuity of service, integrity, operational safety and finally robustness to environment changes. Taken separately, these requirements can be achieved by actions recommended by the literature. For more precision and availability, coupling between absolute GNSS data and relative INS and odometer data, is recommended. To increase safety and integrity, a fault detection layer is essential, but this will negatively impact availability. One therefore needs a fault management layer. A harmonious policy, thought at the function design, makes it possible to achieve all the objectives. In this study, we propose a framework based on a tripartite approach: the tight fusion of GNSS and IMU data, the development of a diagnostic layer based on information theory and using the very promising alpha Rényi divergence, as well as a fault isolation layer. The diagnostic layer is designed to be robust and adaptive to changing environment through a deep neural network. The proposed framework is tested on data acquired in the field. Encouraging results allow to consider the generalization of the concept.

在许多运输应用中，安全关键功能之一是定位。对于自动驾驶汽车等陆地交通应用来说更是如此。虽然 GPS、伽利略、北斗或格洛纳斯等卫星定位系统的民主化使得考虑适用于世界任何地方的全球解决方案成为可能，但通过接收两万公里之外的卫星信号进行定位的原理显示出局限性当它们遇到与接收器附近的环境有关的干扰时。然而，对于这些安全关键型应用来说，要求非常严格，有时甚至是相互冲突的。开发的功能必须满足规定的精度、可用性、服务连续性、完整性、操作安全性以及最终对环境变化的鲁棒性水平。单独来看，这些要求可以通过文献推荐的行动来实现。为了获得更高的精度和可用性，建议将绝对 GNSS 数据与相对 INS 和里程计数据耦合。为了提高安全性和完整性，故障检测层至关重要，但这会对可用性产生负面影响。因此需要一个故障管理层。和谐的政策，在功能设计上的思考，使得所有目标的实现成为可能。在这项研究中，我们提出了一个基于三方方法的框架：GNSS 和 IMU 数据的紧密融合，基于信息论并使用非常有前途的 alpha Rényi 散度开发诊断层，以及故障隔离层。诊断层被设计为通过深度神经网络保持鲁棒性并适应不断变化的环境。所提出的框架在现场获取的数据上进行了测试。令人鼓舞的结果允许考虑概念的概括。