Abstract A significant growth in Unmanned Aircraft System (UAS) operations has been observed over the past decade, largely driven by the emergence of new commercial opportunities and use-cases. This has posed new technological and regulatory challenges in order to address the complex safety, efficiency and sustainability requirements associated with UAS operations in an increasingly congested airspace. The growing need for trusted autonomy in UAS operations imposes demanding performance requirements on Navigation and Guidance Systems (NGS), both in terms of accuracy, integrity, continuity and availability. In most current NGS implementations, system autonomy is tightly constrained within a specified set of operational and environmental conditions through a large number of explicit rules. Recent breakthroughs in Artificial Intelligence (AI)-based methods and the emergence of highly-parallelized processor boards with low form-factor has led to the opportunity to employ Machine Learning (ML) techniques to enhance navigation system performance, particularly for small UAS (sUAS), which account for the majority of current and future unmanned aircraft use-cases. sUAS navigation systems typically employ diverse low Size, Weight, Power and Cost (SWaP-C) sensors such as Global Navigation Satellite System (GNSS) receivers, MEMS-IMUs, magnetometers, cameras and Lidars for localization, obstacle detection and avoidance. This paper presents a comprehensive review of conventional sUAS navigation systems, including aspects such as system architecture, sensing modalities and data-fusion algorithms. Additionally, performance monitoring and augmentation strategies are critically reviewed and assessed against current and future UAS Traffic Management (UTM) requirements. The primary focus is on the identification of key gaps in the literature where the use of AI-based methods can potentially enhance navigation performance. A critical review of AI-based methods and their application to sUAS navigation is conducted, along with an assessment of the performance benefits they provide over conventional navigation systems. Reviewed methods include but are not restricted to Artificial Neural Networks (ANN) such as Convolutional and Recurrent Neural Networks (CNN and RNN), Support Vector Machines (SVM) and ensemble techniques. The key challenges associated with adapting these methods to address sUAS operational objectives are clearly identified. The review also covers the assurance of predictable, deterministic system behaviour which is a key requirement to support system certification. The review and analysis will inform the reader of the applicability of various AI/ML methods to sUAS navigation and autonomous system integrity monitoring, and its likely role in the ongoing UTM evolution.

中文翻译：

---

小型无人机智能自主导航系统研究进展

摘要 在过去十年中，无人机系统 (UAS) 的运营出现了显着增长，这主要是由于新商业机会和用例的出现。这带来了新的技术和监管挑战，以解决与无人机系统在日益拥挤的空域中运行相关的复杂安全、效率和可持续性要求。在 UAS 操作中对可信自治的需求不断增长，对导航和制导系统 (NGS) 提出了苛刻的性能要求，包括准确性、完整性、连续性和可用性。在大多数当前的 NGS 实施中，系统自治通过大量明确的规则被严格限制在一组指定的操作和环境条件内。最近在基于人工智能 (AI) 的方法方面取得的突破以及具有低外形尺寸的高度并行处理器板的出现，使得有机会采用机器学习 (ML) 技术来增强导航系统性能，特别是对于小型无人机 (sUAS) )，占当前和未来无人驾驶飞机用例的大部分。sUAS 导航系统通常采用各种小尺寸、重量、功率和成本 (SWaP-C) 传感器，例如全球导航卫星系统 (GNSS) 接收器、MEMS-IMU、磁力计、摄像头和激光雷达，用于定位、障碍物检测和避让。本文全面回顾了传统 sUAS 导航系统，包括系统架构、传感模式和数据融合算法等方面。此外，根据当前和未来的 UAS 交通管理 (UTM) 要求严格审查和评估性能监控和增强策略。主要重点是确定文献中的关键差距，其中使用基于人工智能的方法可以潜在地提高导航性能。对基于人工智能的方法及其在 sUAS 导航中的应用进行了严格审查，同时评估了它们相对于传统导航系统提供的性能优势。审查的方法包括但不限于人工神经网络 (ANN)，例如卷积和循环神经网络（CNN 和 RNN）、支持向量机 (SVM) 和集成技术。明确指出了与调整这些方法以解决 sUAS 操作目标相关的主要挑战。审查还涵盖了可预测、确定性系统行为的保证，这是支持系统认证的关键要求。审查和分析将告知读者各种 AI/ML 方法对 sUAS 导航和自主系统完整性监控的适用性，以及它在正在进行的 UTM 演变中的可能作用。