如何编写地信测绘信息相关的综述论文-总结版本
A. Hamissi, A. Dhraief, and L. Sliman, “A Comprehensive Survey on Conflict Detection and Resolution in Unmanned Aircraft System Traffic Management,” IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, 2024 DEC 10, 2024.
摘要
痛点:Several surveys have reviewed conflict resolution methods for UAVs. However, to the best of our knowledge, there is no survey specifically addressing conflict detection and resolution methods in UTM, particularly those using AI-based methods.
方法分类:The methods into two categories: non-learning methods and learning-based methods.
Non-learning methods typically rely on pre-defined algorithms.
Learning-based methods, including Machine Learning (ML) and Reinforcement Learning (RL), enable UAVs to adapt to their environment, autonomously
研究贡献:This article can serve as a foundational resource for researchers in guiding their selection of methods for conflict resolution, particularly those relevant to UTM systems.
I. Introduction
无人机广泛应用在多个领域;
欧洲无人机报告指明了未来的巨大市场空间;
多个国家发展UTM系统来管理UAV活动;
各个国家具体的系统项目建设情况;
存在动静态冲突,冲突事故与日俱增,带来了财产损失;
各种Collision Avoidance Systems (CAS)发展起来了;
prior knowledge & pre-mission path planning have notable limitations particularly in dynamic environments. To overcome these limitations, many researchers have started exploring new conflict resolution methods using Machine Learning (ML) and more particularly Reinforcement Learning (RL). This is because RL methods excel in dynamic environments due to their ability to learn from interactions with the environment adapting and optimize their strategies overtime to handle complex and evolving scenarios. RL方法的优势
检索关键词及检索到的综述类文章(配套表)
this survey aims to outline the fol-lowing anticipated contributions:
i) 经典方法涵盖冲突解决服务中的三层:战略去冲突、战术去冲突和碰撞避免; Providing a comprehensive review of the primary classical methods employed in collision avoidance systems, encompassing sensing, detection, and res-olution. We compare these methods, presenting the advantages and disadvantages of each. Our study includes a detailed examination of classical methods that can be applied within the context of an UTM, spanning all the three layers of the conflict resolution service in an UTM: strategic deconfliction, tactical deconfliction and collision avoidance;
ii) 经典方法在 UTM 中应用的挑战和考虑因素 Delving deeply into the challenges and considerations associated with these meth-ods for an UTM.
iii) 三大机器学习方法:监督学习、无监督学习和强化学习 Offering an extensive review of conflict resolution methods based on the three major machine learning approaches: supervised learning, unsupervised learning, and RL, for both UAVs in general and within the UTM framework;
iv) 无人机冲突解决方法时的主要挑战和关注点 Identifying and analyzing the main challenges and concerns related to conflict resolution when employing UAVs conflict resolution methods.
本文的结构
II. FUNDAMENTALS: UAV, UTM
A. Unmanned Aerial Vehicle (UAV)
B. Unmanned Aircraft System Traffic Management (UTM)
UTM(无人机交通管理系统)代表了一个由相关方组成的联盟,合作提供一组服务,这些服务主要旨在确保无人机操作的安全和成功执行。相关方可以是个人、组织或团队,他们对空域管理或无人机操作有利益、参与或影响。相关方的集合包括以下实体 [19]:i) 操作员:合法实体或个人通过远程飞行员或自动机载系统控制无人机,并承担进行安全操作的责任;ii) 监管机构:制定政策和规章以确保空域安全;iii) 服务提供商:为操作员提供的主要服务提供商称为 USS(无人机系统服务供应商),提供包括注册和识别在内的服务。而补充数据服务提供商(SDSPs)提供与天气、地面风险观察等有关的信息,包括建筑物等特征的数据和人口密度等信息;iv) 其他相关方:包括公共安全组织和普通公众。
为了高效和安全地运行 UTM 概念,无人机系统(UAS)操作员必须获得与交通管理、导航和协调相关的特定服务 [31]。在交通管理方面,这些服务包括注册和识别、跟踪、空域管理(包括地理围栏)和冲突管理。对于导航,UTM 提供任务管理服务,而在数据和协调方面,它涉及应急管理工作、监控和与 ATC(空中交通管制)的接口 [4]。ATC 为有人驾驶飞机提供指导和指令,以确保安全和高效的空中交通流量。
III. THE UTM CONFLICT MANAGEMENT SERVICE
the safe operation of the UAVs.
hazard: environmental hazards arising from factors such as severe weather conditions, technical hazards stemming from tech-nical deficiencies, or organizational hazards associated with communication problems
risk: associated risk measure that represents the probability of a hazard materializing
conflict: denotes the case wherein two or more UAVs operate in close proximity or follow intersecting flight trajectories, thereby creating the potential for collision or safety hazards
accident: The causes of accidents and incidents include a pilot’s loss of awareness of the vehicle’s position, collisions with dynamic and static objects, issues in components, or navigation system malfunctions.
incident and damage: repercussions [ 后果 ] affecting the environment, humans, or other vehicles.
在空管(空中交通管制)领域,strategic separation(战略分离)是指在航班计划和调度阶段采取的长距离或长时间的分离措施。这种分离措施通常是在飞行前的规划阶段进行,主要用于确保航班在整个飞行过程中保持安全距离,避免空中相撞,并优化空中交通流量。
塔台会采取 tactical separation 措施,确保飞机之间的安全距离。
各种UTM项目
FAA 的 UTM 项目,于 2016 年由 FAA 和 NASA 合作启动。FAA 的 UTM 通过多个层次的分离措施确保无人机系统的安全操作,包括战略规划和战术措施。超出视线范围(BVLOS)的运营商必须通过 USS 网络共享飞行计划,以避免冲突。UTM 根据操作风险水平调整分离协议,在高流量区域要求使用先进的 DAA(探测和避让)技术,并在必要时采取战术冲突解决措施。
欧洲的 U-Space 项目由 SESAR 联合事业(SJU)于 2017 年启动,将在其第二开发阶段(2022 年)开始管理冲突解决。据预测,对于一个拥有 100 万人口的城市,每日无人机操作量可能在 1,000 到 100 万之间,每分钟约有 700 次飞行,任何时间点上空可能有 10,000 架无人机在飞行。对于一个拥有 2000 万人口的国家,每日操作量可能在 2 万到 2000 万之间,每分钟约 14,000 次飞行,同时可能有 21 万架无人机在空中飞行。无人机密度可能在每平方千米 0.0012 到 1.2 架之间。
类似于 FAA 的 UTM,U-Space 的冲突管理包括三个层次:战略去冲突、战术分离提供和碰撞避免。在这一背景下,BUBBLES 项目 [37] 定义了 U-Space 中一种新的分离管理方法,强调基于风险、基于性能和情景依赖的方法。该项目旨在通过定义 UAS 操作的目标安全水平(TLS)来限制碰撞风险到可接受的水平。项目提出了根据交通类别、操作环境和 CNS(通信、导航和监视)系统性能的动态分离最小值。
USEPE 项目引入了 D2-C2 方法,用于 U-Space 中的无人机分离,该方法结合了动态空域划分和城市区域上方的高速走廊。它利用机器学习(ML)技术支持 U-Space 服务提供商(USSPs)完成诸如路线规划、冲突解决和动态划分更新等任务。该项目旨在通过在 BlueSky ATM 中进行机器学习模拟来验证这种方法,模拟的情景包括城市监控、最后一公里配送和紧急情况。提出的冲突管理方法考虑了现实世界的因素,如天气条件、无人机性能和技术限制。
几何方法包括从初始轨迹偏离的机动 [60] 以及速度的改变,无论是方向还是大小 [57]、[58]。Ho 等人 [57] 引入了一种新的 ORCA 算法版本,称为适应性 ORCA,专门设计用于碰撞避免。
IV. NON-LEARNING METHODS FOR CONFLICT MANAGEMENT
detect and avoid (DAA) systems to maintain safe airspace separation
DAA system for unmanned aircraft uses algorithms, displays, and procedures to maintain safe separation and prevent mid-air collisions by providing pilots with timely alerts to avoid other aircraft.
The dynamic rerouting service assists the operators by modifying the UAV trajectories and providing guidance to alter flight paths.
The goal of a CAS, that is ensuring the safety of the UAV flights by preventing collisions, aligns with the primary objective of an UTM.
势场法:不足在于容易陷入局部最小值、需要大量计算资源和时间、要求预先了解障碍物的信息难以动态
The force field methods [48], [49], [50] also known as potential field methods involve applying the principles of attraction and repulsion to various elements characterizing the flight path of an UAV, treating it as a charged particle.
some methods, like potential field, optimized trajectory and geometric methods, have high computational complexity to be applied in real-world problems [9].
The inherent principle of these methods may lead to either local minima or the risk of failing to reach the destination if an obstacle is in close proximity [8], [9]. In the optimized trajectory methods [51], [52], [53], collision avoidance is modeled as an optimization problem aimed at finding optimal solutions for a collision-free trajectory based on avoiding obstacles, considering the constraints of the problem. This method, in certain instances, results in high computational complexity [9]. It requires a prior knowledge of obstacles, including their dimensions and positions, rendering this approach challenging to apply in dynamic environments and less effective in avoiding dynamic obstacles
具体而言,当存在障碍物时,会产生排斥力,而吸引力则指向目标或目的地 [7]、[8]、[9]、[10]。文献指出,势场方法存在诸多局限性。特别是,它们在实际无人机场景中的应用具有挑战性,需要大量的计算能力和时间 [7]、[8]。这些方法的内在原理可能导致陷入局部最小值,或者在障碍物靠近时无法到达目的地 [8]、[9]。在优化轨迹方法 [51]、[52]、[53] 中,避障被建模为一个优化问题,旨在根据避障需求和问题的约束条件找到无碰撞的最优轨迹。在某些情况下,这种方法会导致较高的计算复杂度 [9]。它需要预先了解障碍物的信息,包括其尺寸和位置,这使得该方法在动态环境中的应用具有挑战性,并且在避动态障碍物时效果较差。
强化学习(RL)和势场方法被结合用于路径规划。在文献 [111] 中,提出了一种增强的路径规划方法,将黑洞势场(BHPF)与强化学习集成,形成了黑洞势场深度 Q 学习(BHDQN)方法。该方法解决了传统人工势场(APF)方法的局限性,特别是在复杂环境中容易陷入局部最小值的问题。黑洞势场(BHPF)作为学习环境,使智能体能够自主适应并导航朝向目标,同时避开障碍物。该方法通过静态和动态实验进行了验证,表明智能体能够有效地逃离局部最小值,并且快速适应新出现的障碍物。
而在文献 [112] 中,方法则相反。实际上,通过集成人工势场(APF)来增强强化学习(RL)方法。文献 [112] 引入了 B-APFDQN,这是一种为无人机(UAV)设计的高级路径规划方法,它结合了深度 Q 学习(DQN)和 APF,以提高收敛速度和路径效率。仿真结果表明,B-APFDQN 显著减少了试错的频率,并且相比于传统的 DQN 方法,实现了更快的收敛。
Path planning method with improved artificial potential field—A reinforcement learning perspective
在复杂环境中,代理(agent)容易陷入局部稳定点,因为当合力为零时,代理就会停止移动。这个问题的主要原因是环境中障碍物的形状多样和位置关系复杂。为了克服这个问题,研究人员做出了很多努力。
贾等人通过离散化障碍物的轮廓来改变障碍物的排斥势,从而解决局部稳定点问题 [14]。李等人提出了一种改进的人工势场回归搜索方法,用于完全已知环境中的自主移动代理路径规划 [15]。奥罗斯科-罗萨斯等人提出了一种膜进化人工势场方法来解决移动代理路径规划问题。该方法利用遗传算法寻找生成可行且安全路径的参数 [16]。里兹基等人设计了一种势函数,引导四旋翼飞行器前往目标并避开障碍物。该算法通过利用沿墙行为来解决局部稳定点问题 [17]。
目前,解决局部稳定点问题的主要思路是改变势场,以减少局部稳定点的出现。
the environment is a Markov decision process (MDP)
V. LEARNING-BASED METHODS FOR CONFLICT MANAGEMENT
VI. CHALLENGES AND FUTURE RESEARCH
可以将高级优化技术集成到经典算法中,XAI可以提高基于学习的决策的透明度。