Mission-based Architecture for Swarm Composability (MASC) 基于任务的集群组合性框架

Mission-based Architecture for Swarm Composability (MASC)

基于任务的集群组合性框架

闲着没事借助google翻译了一篇集群方面的文章
主要思想是用自上而下和自下而上形式结合做集群。把要完成的任务分成不同的阶段（飞行前准备、进入任务、任务、退出任务、任务后收尾），在不同阶段又将任务分为不同的层次。在核心的任务段（例如侦察、情报、搜救），提取一些典型的集群策略（例如搜索、跟踪、攻击），在策略的下层使用集群行为（例如发射、回收、到航点、群体分开、群体连接、群体避障等），再集群行为下设置集群算法层、最下层是个体无人机层。详见第三章。
个人觉得每一层都是都做成库的形式，层与层之间可以由上到下的调用。做好了这种框架，可以上下并行开发系统了。

摘要

Unmanned aerial vehicle (UAV) swarm design and mission application is a burgeoning area of research. The applications of swarm technology to unmanned systems are in the infancy of realization, although clear benefits from the enhanced capabilities can be easily envisioned for commercial and government missions including persistent search, long-term monitoring, sensor data collection, object retrieval, and offensive attack missions. Designing a UAV swarm architecture without considering the mission doctrine is imprudent, as is developing a mission doctrine without understanding the unique capabilities of swarm technology. This paper provides a literature review of current UAV swarm design architectures, most of which employ bottom-up methods with little consideration for an intended operational mission. The proposed top-down framework and modeling approach is focused on developing a modular UAV swarm playbook-type architecture by decomposing a subset of US Navy missions, and developing tactics that can be applied to those missions and support future missions. The framework proposed herein is derived from a consolidation of heuristics from the model-based systems engineering, robotics, human systems integration, biology, and computer science disciplines. The goal of this framework is to provide a standard methodology for designing and operating swarm unmanned systems that will meet the performance requirements for military missions. In this paper, an integrated framework for designing a UAV swarm system with swarm mission doctrine as a primary design factor is presented.

无人机（UAV）群体设计和任务应用是一个新兴的研究领域。群体技术在无人系统中的应用正处于实现的初期阶段，尽管可以轻松地为商业和政府任务设想增强功能的明显好处，包括持续搜索，长期监测，传感器数据收集，对象检索和攻击性攻击任务。在不考虑任务原则的情况下设计无人机群体架构是不谨慎的，正如在不了解群体技术的独特能力的情况下开发任务原则一样。本文提供了当前无人机群体设计架构的文献综述，其中大多数采用自下而上的方法，很少考虑预期的操作任务。建议的自上而下的框架和建模方法侧重于通过分解美国海军任务的一部分来开发模块化无人机群体游戏书型架构，并制定可应用于这些任务并支持未来任务的策略。本文提出的框架源自基于模型的系统工程，机器人，人体系统集成，生物学和计算机科学学科的启发式整合。该框架的目标是为设计和运行满足军事任务性能要求的群体无人系统提供标准方法。本文提出了一种设计无人机群系统的集成框架，该系统以群体任务原则为主要设计因素。

1，背景和介绍

Interest in swarm technology has increased over the last two decades. Much of this interest can be attributed to the dynamic field of unmanned systems technology, which has been rapidly developing both in government and in the private sector. Unmanned system technology has expanded from physically hazardous, extended-endurance, highaltitude, military missions to agriculture, search and rescue, environmental research, and commercial missions1. Unmanned systems provide many advantages over manned systems. In the case of UAVs, they are less constrained by human factors such as crew rest, G-tolerance, environmental conditions, and comfort. Unmanned systems can be expendable and could have lower life-cycle costs than manned systems; however, low system reliability2 low technology readiness levels, large logistical footprints, and an ironically increased manpower requirement have marginalized cost advantages.

在过去的二十年中，对群体技术的兴趣有所增加。 这种兴趣很大程度上归功于无人系统技术的动态领域，无人系统技术在政府和私营部门都在迅速发展。 无人系统技术已从物理危险，延长耐力，高度，军事任务扩展到农业，搜索和救援，环境研究和商业任务1。 与载人系统相比，无人系统具有许多优势。 在无人机的情况下，它们受到人员因素的限制，例如船员休息，G耐受性，环境条件和舒适性。 无人系统可以是消耗性的，并且可以比有人系统具有更低的生命周期成本; 然而，低系统可靠性2，低技术准备水平，大量的物流足迹以及具有讽刺意味的增加的人力需求使得成本优势边缘化。

Swarm application to unmanned systems derives impetus from biology. Large numbers of individuals such as birds, fish, or insects may work together to accomplish useful tasks that cannot be completed by an individual or any group of non-cooperative individuals. Members of the swarm may be unintelligent and inefficient on an individual scale, yet inter-agent interactions produce emergent behavior that enables advantages such as robustness, flexibility, and scalability3. It is the local interactions among the agents and between the agents and their environment that may elicit beneficial collective behavior3,4. For this paper, a swarm is defined as a group of 10 or more individual, self-organized, homogeneous UAVs that perform a mission through local interactions under a decentralized control architecture5,6. The number 10 was chosen as a lower limit because it likely exceeds the number of UAVs that one human can directly control under operationally relevant workloads7,8. Decentralized control is stipulated because it permits scalability, robustness, and self-organization6.

Swarm应用于无人系统源于生物学的推动力。诸如鸟类，鱼类或昆虫之类的大量个体可以一起工作以完成无法由个人或任何非合作个体组完成的有用任务。群体成员在个体规模上可能是非智能和低效的，但代理间交互产生的紧急行为能够实现诸如健壮性，灵活性和可扩展性等优势3。代理人之间以及代理人与他们的环境之间的本地互动可能引发有益的集体行为3,4。对于本文，群体被定义为一组10个或更多个体，自组织，同质的无人机，通过分散控制架构下的本地交互执行任务5,6。数字10被选为下限，因为它可能超过一个人在操作相关工作负荷下可直接控制的无人机数量7,8。规定了分散控制，因为它允许可扩展性，稳健性和自组织6。

Swarm system behaviors are difficult to predict and control due to their complexity9. The number of possible behavioral interactions increases exponentially with the number of agents in the swarm9. Yet a key advantage of swarm technology is that the swarm is designed to be composed of simple, modular, identical components6. These homogeneous agents are not programmed for a specific role and do not operate under a centralized coordinating agent. Accordingly, the loss of an individual does not cause a significant decrease in system performance because another agent or agents can assume its duties. Thus, a homogeneous swarm can be adaptable, robust, and scalable3,6. Bachrach et al. eloquently underscored the need for a high-level language to orchestrate multi-agent operations by composing new behavior plans from existing lower-level programs9.

由于其复杂性，群体系统行为难以预测和控制9。 随着swarm9中代理的数量，可能的行为交互的数量呈指数增长。 然而，群体技术的一个关键优势是群体设计由简单，模块化，相同的组件组成6。 这些同质代理不是针对特定角色编程的，不在集中协调代理下运行。 因此，个人的损失不会导致系统性能的显着降低，因为另一个或多个代理可以承担其职责。 因此，同质群可以是可适应的，稳健的和可扩展的3,6。 Bachrach等。 雄辩地强调需要一种高级语言来编排多代理操作，方法是从现有的低级程序中编写新的行为计划9。

2，当前的方法

Bottom-up modeling approaches focus on assembling sub-components of systems to build more complex systems. Agent-based modeling (ABM), finite state machines and Petri Nets are bottom-up modeling methods frequently used in swarm system design. Bottom-up models are advantageous from a modularity and composability perspective, but they often risk failing to meet higher-level system requirements if design begins before a higher-level system architecture is established. Consequently, combining bottom-up and top-down models is a software development heuristic10.
自下而上的建模方法侧重于组装系统的子组件以构建更复杂的系统。 基于代理的建模（ABM），有限状态机和Petri网是在群系统设计中经常使用的自下而上建模方法。 从模块化和可组合性的角度来看，自下而上的模型是有利的，但如果在建立更高级别的系统架构之前开始设计，它们通常有可能无法满足更高级别的系统要求。 因此，自下而上和自上而下模型相结合是一种软件开发启发式10。

Modeling agent-level interactions can provide valuable information regarding the emergent behavior inherent to the complex adaptive system (CAS). ABM is a commonly used approach for modeling a CAS as a group of autonomous agents who make decisions individually based on their assessment of the environment and in accordance with a rule set11,12. McCune et al.13 used ABM to investigate command and control of a UAV swarm, Bonabeau simulated human systems using ABM methods11, and Munoz14 studied defensive UAV swarm employment using ABM. The emergent behavior of a swarm system is a critical attribute to consider when designing a swarm system, developing swarm tactics, or devising an assessment methodology for a swarm system.

728/5000
建模代理级别的交互可以提供有关复杂自适应系统（CAS）固有的紧急行为的有价值信息。 ABM是一种常用的方法，用于将CAS建模为一组自主代理，这些代理根据其对环境的评估并根据规则集11,12单独做出决策。 McCune等人13使用ABM来研究无人机群的命令和控制，Bonabeau使用ABM方法模拟人类系统11，Munoz14使用ABM研究防御性无人机群体就业。 群体系统的紧急行为是设计群体系统，开发群体策略或设计群体系统评估方法时需要考虑的关键属性。

Finite state machines (FSM) have been used to model multi-vehicle autonomous, unmanned system architectures15,16,17,18. Within an FSM architecture, each agent operates within one of several defined states at a given time with trigger events causing the agent to transition between states precipitated by environmental conditions it senses or events it encounters. This type of structure is applicable in developing military swarm systems as the states and triggers can be defined deterministically, which is necessary for high-risk mission events. Weiskopf et al. demonstrated a control architecture based on a library of basic tasks to support a search and track reference mission18. Conversely, there may be other mission events, such as covert searching, in which some bounded degree of unpredictability is desired. In those cases, probabilistic finite state machines (PFSM), can be used to allow for different behaviors within a state or by allowing multiple transitions between states17. Using this architecture, the transition
probabilities can remain fixed or change over time19. PFSM have been used to produce collective behaviors such as task allocation and aggregation15.
有限状态机（FSM）已被用于模拟多车辆自主，无人系统架构15,16,17,18。在FSM架构内，每个代理在给定时间内在若干定义状态之一内操作，触发事件使代理在其感知的环境条件或其遇到的事件引发的状态之间转换。这种类型的结构适用于开发军事群体系统，因为可以确定性地定义状态和触发器，这对于高风险任务事件是必要的。 Weiskopf等人。展示了基于基本任务库的控制架构，以支持搜索和跟踪参考任务18。相反，可能存在其他任务事件，例如隐蔽搜索，其中需要一些有限程度的不可预测性。在这些情况下，概率有限状态机（PFSM）可用于允许状态内的不同行为或允许状态之间的多个转换17。使用这种架构，转换概率可以保持固定或随时间变化19。 PFSM已被用于产生集体行为，如任务分配和聚合15。
Behavior-based design, in which the individual behavior of each agent is developed iteratively until the desired swarm behavior is acquired, has been a typical design method in swarm robotics, in part due to the influence of biological swarms. The term behavior is commonly used in robotics literature to describe the actions being performed by robots, or agents. Behaviors apply to individuals and environments, as well as to groups, which are often called collective behaviors. Behaviors may also be categorized as higher-level abstract behaviors and lower-level primitive behaviors, or simply primitives19. The term primitives is borrowed from the computer science discipline and functions similarly in robotics literature; they act as building blocks for programming higher-level functions. A seminal behavior-based design is Brooks’ subsumption architecture which uses a layering approach for controlling systems and incorporates augmented FSM processors for managing inputs and outputs20. A key contribution of Brooks’ work was that he partitioned the robot control problem into behaviors rather than into functional modules. As a result, his hierarchy of layered behaviors allows higher levels of behaviors to subsume lower layers of less complex behavior. Nicolescu and Matarić presented a behavior-based hierarchical architecture for robots in which reusable primitive behaviors support a library of abstract behaviors21. The proposed MASC framework presents an approach similar to Nicolescu and Matarić’s, but inspired from a navy doctrinal perspective, that may provide applications across the gamut of prospective military swarm operations to enhance mission suitability of future swarm systems
1706/5000
基于行为的设计，其中每个代理的个体行为被迭代地开发直到获得期望的群体行为，已经是群体机器人中的典型设计方法，部分地由于生物群的影响。术语行为通常用于机器人文献中，以描述由机器人或代理执行的动作。行为适用于个人和环境，以及通常称为集体行为的群体。行为也可以归类为更高级别的抽象行为和更低级别的原始行为，或者仅仅是原始行为19。原语这个术语来自计算机科学学科，在机器人文学中也有类似的功能;它们充当编程更高级函数的构建块。基于行为的开创性设计是Brooks的包容体系结构，它使用分层方法来控制系统，并采用增强型FSM处理器来管理输入和输出20。布鲁克斯的工作的一个关键贡献是他将机器人控制问题划分为行为而不是功能模块。因此，他的分层行为层次结构允许更高级别的行为包含更低层次的复杂行为。 Nicolescu和Matarić为机器人提出了一种基于行为的分层架构，其中可重用的原始行为支持抽象行为库21。拟议的MASC框架提出了类似于Nicolescu和Matarić的方法，但受到海军理论观点的启发，可能提供跨越未来军事群体操作的应用，以增强未来群体系统的任务适用性

Top-down methods are surfacing to augment the bottom-up models and expedite their progression from labs and simulations to operationally useful systems. Model-based composable architectures, such as playbooks, merge the benefits of top-down and bottom-up approaches and promote reusable designs. DeLoach et al. developed the MultiAgent Systems Engineering methodology for analyzing, designing, and producing heterogeneous multi-agent systems22. Their seven-step method uses graphically based models to define the agents and interfaces. Brambilla et al. proposed a four-phase, property-driven, top-down design method that formally describes the features of the system the designer wants to realize23.
自上而下的方法正在浮出水面，以增强自下而上的模型，加快从实验室和模拟到操作上有用的系统的发展。 基于模型的可组合体系结构（如playbooks）融合了自上而下和自下而上方法的优点，并促进了可重用的设计。 DeLoach等 开发了MultiAgent系统工程方法，用于分析，设计和生成异构多代理系统。 他们的七步法使用基于图形的模型来定义代理和接口。 Brambilla等 提出了一种四阶段，属性驱动，自上而下的设计方法，正式描述了设计者想要实现的系统的特征23。

2.2.1. Playbooks
The idea of using a playbook, or collection of pre-defined tactics or action plans for multi-robot unmanned systems is not new and has been used in both UAVs24 and UGVs25. A playbook can be described as a library of action plans that enable the operator to manage the agents’ behavior at multiple levels and tailor the level of autonomy to the mission26. Rather than controlling each sequence of actions, the catalog of pre-defined behaviors simplifies user control, reduces necessary communications, and synchronizes agent tasking26. A good playbook should be constructed at a high enough level of abstraction such that action plans (tactics in the proposed MASC) are common to several missions, and can be tailored for a variety of uses by adjusting tune-able parameters. A benefit to using a playbook architecture for UAVs is the possibility to extend the pool of potential operators beyond specifically trained UAV operators.
对多机器人无人系统使用剧本或预定义战术或行动计划的集合的想法并不新鲜，并且已经在UAV24和UGVs25中使用。 剧本可以被描述为一个行动计划库，使操作者能够在多个层面管理代理人的行为，并为任务26定制自治水平。 预定义行为的目录不是控制每个动作序列，而是简化用户控制，减少必要的通信，并同步代理任务26。 应该在足够高的抽象层次上构建一个好的剧本，以便行动计划（拟议的MASC中的策略）对于几个任务是通用的，并且可以通过调整可调参数来针对各种用途进行定制。 使用无人机的剧本架构的好处是可以将潜在操作者的范围扩展到经过专门训练的无人机操作者之外。

Playbooks may provide a means to improve human-machine teaming, an important factor for operating swarm systems. Squire et al. studied the effects of various multi-vehicle control architectures on human performance27. This work was built upon Parasuraman et al.'s experiments that used a simplified version of the “Playbook” interface in the RoboFlag multi-vehicle simulation environment28. The simplified RoboFlag Playbook used a hierarchical task model to task robots to perform various functions under temporal and conditional constraints, using three architectures of different automation levels. The authors evaluated the effects of these three architectures on human performance under defensive, offensive, and mixed conditions. The results of this study suggest that a playbook type architecture provides operators with the requisite flexibility of balancing manual control and cognitive workload when managing unforeseeable conditions28. Furthermore, the playbook system architecture looks to be promising for overcoming the “single robot parenting” paradigm.
Playbooks可以提供一种改进人机组合的方法，这是操作群系统的一个重要因素。 Squire等人。 研究了各种多车辆控制架构对人体表现的影响27。 这项工作建立在Parasuraman等人的实验基础之上，该实验在RoboFlag多车辆仿真环境中使用了“Playbook”界面的简化版本28。 简化的RoboFlag Playbook使用分层任务模型，使任务机器人在时间和条件约束下使用三种不同自动化级别的体系结构执行各种功能。 作者评估了这三种架构在防御，进攻和混合条件下对人类表现的影响。这项研究的结果表明，在管理不可预见的条件时，Playbook类型的体系结构为操作员提供了平衡手动控制和认知工作负载的必要灵活性28。 此外，剧本系统架构似乎有望克服“单一机器人育儿”范式。

Browning et al.'s “STP” (skills, tactics, and plays) multi-robot architecture is particularly applicable to this research because it was designed to control a team of autonomous robots in an adversarial environment: RoboCup robot soccer29. In the STP architecture, skills are the low-level control algorithms that support the tactics, which in turn compose the higher-level plays. In this architecture, tactics and skills can be detached from plays to support a hierarchical control structure for operating individual heterogeneous robots that perform different roles.
Goldman et al. addressed the problem of over-constrained missions in which the system cannot meet the requests of the operator, due to temporal, geographic or other constraints24. They improved the Smart Information Flow Technology (SIFT) Playbook-enhanced Variable Autonomy Control System (PVACS) project by adding a “best effort” planning mode, using cost-based optimization, to relax constraints and provide viable alternative play options24. The capability to automate modifications to plays extends the potential for multi-vehicle unmanned systems to be operated by users who may not be specifically trained for the UAV’s capabilities.
In 2017, DARPA announced a program called OFFSET in order to advance swarm technology by focusing on human-swarm teaming and swarm autonomy within a realistic gaming environment30. Their method takes a hierarchical approach to the swarm framework which is composed of a mission, tactics, primitives, and supporting algorithms30. This program focuses exclusively on the urban operational environment, with a goal of building a playbook of tactics to support the framework.
Browning等人的“STP”（技能，战术和游戏）多机器人体系结构特别适用于这项研究，因为它旨在控制对抗环境中的自主机器人团队：RoboCup机器人足球29。在STP架构中，技能是支持战术的低级控制算法，而后者又构成了更高级别的游戏。在这种体系结构中，策略和技能可以与游戏分离，以支持用于操作执行不同角色的各个异构机器人的分层控制结构。
Goldman等人。解决了过度约束的任务问题，由于时间，地理或其他限制，系统无法满足运营商的要求24。他们通过添加“尽力而为”计划模式，使用基于成本的优化，放宽约束并提供可行的替代游戏选项24，改进了智能信息流技术（SIFT）Playbook增强型可变自治控制系统（PVACS）项目。自动修改游戏的能力扩展了多车辆无人系统由未经过UAV专门训练的用户操作的可能性。
2017年，DARPA宣布了一项名为OFFSET的计划，旨在通过在真实的游戏环境中专注于人群聚集和群体自治来推进群体技术30。他们的方法采用分层方法来处理群体框架，该框架由任务，策略，原语和支持算法组成30。该计划专注于城市运营环境，目标是建立一个支持框架的战术手册。

3 Mission-based Architecture for Swarm Composability (MASC)基于任务的框架

3. Mission-based Architecture for Swarm Composability (MASC)
   Terminology for swarm system functions varies depending on the academic field. A standard taxonomy for swarm system functions that is relatable across academic disciplines would facilitate architecture design. This research proposes a top-down methodology to integrate mission doctrine and system architecture design for a UAV swarm, and aims to identify a taxonomy for swarm unmanned system architecture design. MASC is focused on developing a swarm playbook-type architecture by decomposing a subset of US Navy missions, and developing tactics that can be applied to those missions and support future missions. Fig. 1 shows the decomposition of the proposed swarm operational framework, MASC. The focus of this research is on the tactic and play levels of the architecture.
   群体系统功能的术语因学术领域而异。 针对群体系统功能的标准分类法可以跨学科进行关联，这将有助于架构设计。 本研究提出了一种自上而下的方法来整合无人机群的任务原则和系统架构设计，旨在确定群体无人系统架构设计的分类。 MASC致力于通过分解美国海军任务的子集，开发可用于这些任务并支持未来任务的战术来开发群体游戏式架构。 图1显示了所提出的群体操作框架MASC的分解。 这项研究的重点是建筑的策略和游戏水平。

The swarm operational team is based on the UAV swarm field experimentation conducted by the Advanced Robotic Systems Engineering Laboratory (ARSENL) at the Naval Postgraduate School. The team is comprised of a Swarm Commander, Swarm Monitor, and Ground Crew. The Swarm Commander selects the swarm tactics and is responsible for the overall execution of the mission. The Swarm Monitor oversees the health and function of the swarm, and separates errant individual UAVs from the swarm33. The Ground Crew is responsible for UAV preflight, launch, and postflight dutie
群体运行团队基于由海军研究生院的高级机器人系统工程实验室（ARSENL）进行的无人机群体实地试验。 该团队由Swarm Commander，Swarm Monitor和Ground Crew组成。 群体指挥官选择群体战术并负责任务的整体执行。 群体监视器监视群体的健康和功能，并将错误的单个无人机与群体33分开。 Ground Crew负责无人机预检，发射和后飞行

The following UAV swarm mission taxonomy is proposed to describe the overall mission architecture, and includes the following terms:
A swarm mission describes the overall task and purpose delineating the action assigned to the swarm. Example UAV swarm missions include: intelligence, surveillance, and reconnaissance (ISR), maritime interdiction operations (MIO), humanitarian assistance and disaster relief (HADR), and search and rescue (SAR). Each swarm mission is the parent of several swarm mission phases.
A swarm mission phase describes a distinct time period within the mission. There are five operational phases in a swarm mission: Preflight, Ingress, OnStation, Egress, and Postflight. The three phases that cover the in-flight portion of the mission— Ingress, OnStation, and Egress—are the focus of this research. A swarm mission phase is composed of one or more swarm tactics.
提出以下无人机群体任务分类来描述整体任务架构，并包括以下术语：
群体任务描述了描述分配给群体的动作的总体任务和目的。 无人机群体任务包括：情报，监视和侦察（ISR），海上拦截行动（MIO），人道主义援助和救灾（HADR）以及搜救（SAR）。 每个群体任务都是几个群体任务阶段的父母。
群体任务阶段描述了任务中的不同时间段。 群体任务有五个操作阶段：Preflight，Ingress，OnStation，Egress和Postflight。 这项研究的重点是覆盖飞行任务部分Ingress，OnStation和Egress的三个阶段。 群体任务阶段由一个或多个群体战术组成。

A swarm tactic is the employment and ordered arrangement of agents in relation to one another for the purpose of performing a specific task31. Swarm tactics include searching, tracking, monitoring, evading, and attacking. A swarm tactic is composed of one or more swarm plays. For example, the Efficient Search tactic is composed of four different search pattern options (plays). Swarm tactics are designed to be used in multiple missions.
A swarm play describes the maneuvers and behaviors of the swarm as a collective of agents31. The artificial intelligence and robotics communities use the term “behavior” to describe the interaction of the agent with the environment32. Swarm plays can be described as behaviors with specific triggers and temporal constraints, and are the building blocks for swarm tactics. Example swarm plays include launch, transit to waypoint, split, join, and orbit. Swarm play parameters are tunable characteristics of a play—such as altitude blocks or geographic boundaries—that can be changed based on the mission or rules of engagement (ROE). A swarm play is composed of one or more swarm algorithms. Swarm plays are re-used within multiple different swarm tactics in the MASC framework.
Swarm algorithms are the step-by-step procedures used by the controlling software to solve a recurrent task such as sorting, path planning or foraging. Swarm algorithms are the building blocks of swarm plays. A swarm play is composed of one or more swarm algorithms31. Swarm algorithms use data from the individual UAVs such as position, heading, velocity, altitude, attitude, health status, and state31.
群体策略是代理人为了执行特定任务而相互关联的就业和有序安排31。群体策略包括搜索，跟踪，监视，回避和攻击。群体战术由一个或多个群体游戏脚本组成。例如，高效搜索策略由四种不同的搜索模式选项（脚本）组成。群体战术旨在用于多个任务。
群体脚本描述了群体作为代理人集体的动作和行为31。人工智能和机器人社区使用术语“行为”来描述代理与环境的交互32。群体脚本可以被描述为具有特定触发和时间约束的行为，并且是群体战术的构建块。示例群体脚本包括发射，转移到航点，分裂，连接和轨道。群体脚本参数是脚本的可调特性 - 例如高度块或地理边界 - 可以根据任务或参与规则（ROE）进行更改。群脚本由一个或多个群算法组成。 Swarm脚本在MASC框架中的多个不同群体策略中重复使用。
Swarm算法是控制软件用于解决诸如排序，路径规划或觅食等周期性任务的逐步过程。 Swarm算法是群体脚本的构建块。群脚本由一个或多个群算法31组成。群体算法使用来自各个无人机的数据，例如位置，航向，速度，高度，姿态，健康状态和状态31。

3.1. MASC Applied to a Notional Mission Scenario
The MASC framework was applied to a fictional maritime interdiction operation (MIO) in which the US Navy is supporting the Indonesian Navy in combatting illegal shipping in the Sunda Strait. MIO is a US Navy mission, typically executed with maritime air support that involves surveillance and interception of private or commercial vessels, boarding and searching of suspect vessels, and detaining, diverting or seizing vessels found in violation of United Nations sanctions or other international laws34.
MASC框架适用于虚构的海上拦截行动（MIO），其中美国海军正在支持印度尼西亚海军打击Su他海峡的非法航运。 MIO是美国海军的任务，通常由海上空中支援执行，涉及监视和拦截私人或商业船只，登船和搜查可疑船只，以及扣留，转移或扣押违反联合国制裁或其他国际法律的船只34。

The mission is composed of the five phases introduced earlier: Preflight, Ingress, OnStation, Egress, and Postflight. The Preflight phase begins when the swarm is powered on and concludes when the swarm is in a flight ready status. The Ingress phase commences when the first UAV of the swarm has been loaded onto the launcher and concludes when the swarm has arrived at the on-station waypoint. The OnStation phase begins when the entire swarm reaches the assigned on-station area and ends when the first UAV in the swarm reaches low fuel or the on-station tasking ends. The Egress phase starts when the swarm is on a flight path to return to the ship and concludes when the entire swarm has landed. The Postflight phase begins when the swarm has landed and ends when the mission has been debriefed. The inflight phases of the mission (Ingress, OnStation, and Egress) are the focus of the MASC framework. The Preflight, Ingress, Egress, and Postlight phases are similar amongst different mission scenarios, with the greatest variety between missions occurring in the OnStation phase. The flow of mission phases for the MIO mission is shown in the activity diagram in Fig. 2.

任务由前面介绍的五个阶段组成：Preflight，Ingress，OnStation，Egress和Postflight。预检阶段在群体开启时开始，并在群体处于准备就绪状态时结束。当群体的第一个无人机被装载到发射器上时，入口阶段开始，并且当群体到达站上航路点时结束。 OnStation阶段在整个群体到达指定的站内区域时开始，并在群中的第一个UAV达到低燃料或站上任务结束时结束。当群体在飞行路径上返回船舶时，出口阶段开始，并在整个群落落地时结束。 Postflight阶段在群体着陆时开始，在任务被汇报后结束。任务的飞行阶段（Ingress，OnStation和Egress）是MASC框架的重点。 Preflight，Ingress，Egress和Postlight阶段在不同的任务场景中是相似的，在OnStation阶段发生的任务之间的差异最大。 MIO任务的任务阶段流程如图2中的活动图所示。

Each of the phases contains sub-activity diagrams depicting lower-level activities of the swarm, swarm operations team, and external actors, as signified by the “decomposed” notation in the lower part of the block. Fig. 3 shows the breakdown of on-station MIO activities by participating unit. The focus of this research is on the swarm’s activities in the top branch of Fig. 3, which are depicted at the tactics level in Fig. 4
每个阶段都包含子活动图，描述了群体，群体运营团队和外部参与者的低级活动，如块下部的“分解”符号所示。 图3显示了参与单元对站内MIO活动的细分。 本研究的重点是图3顶部分支中的群体活动，如图4中的战术层面所示。

The swarm’s activities during the MIO mission OnStation phase are decomposed and shown in Fig. 4. The light green blocks with labels beginning with “t” (i.e. t3, t4, t5, t8, t9, t13) are tactics the swarm executes. In this mission, the swarm is divided into two sub-swarms that concurrently execute different tactics. Based on the type of cueing information it receives, Sub-swarm 1 monitors or searches the assigned area, tracks the target, and evades the target if it is threatened. Meanwhile, Sub-swarm 2 operates at a higher altitude and relays the communication and sensor data from Sub-swarm 1 to the command and control and other participating units. Concurrently, the swarm is consuming power that is being monitored by the Swarm Monitor in a separate swarm team activity diagram. Finally, the swarm receives the off-station decision from the Swarm Commander, coordinated by the trigger (green parallelogram), and proceeds to the Egress phase.
MIO任务OnStation阶段期间群体的活动被分解并显示在图4中。标有“t”开头的浅绿色块（即t3，t4，t5，t8，t9，t13）是群体执行的策略。 在这个任务中，群体被分成两个子群，同时执行不同的策略。 根据它接收的提示信息的类型，子群1监视或搜索指定区域，跟踪目标，并在目标受到威胁时回避目标。 同时，子群2在更高的高度操作并且将来自子群1的通信和传感器数据中继到命令和控制以及其他参与单元。 同时，群体正在消耗由Swarm Monitor在单独的群组活动图中监视的功率。 最后，群体接收来自Swarm Commander的离站决定，由触发器协调（绿色平行四边形），并进入出口阶段。
Underlying each tactic is a sequence of plays. Each tactic is composed of one or more sensor plays and one or more maneuver plays. Modifiable play parameters such as altitude, velocity, execution times, offset distances, geographic constraints, and frequency settings enable the plays to be tailored to mission specific conditions. Fig 5. is an example of the sequence of plays that compose the Evade tactic. The Evade tactic is triggered by a perceived threat that precipitates a concurrent sensor and maneuver loop. The sensors can be set to a passive emissions control (EMCON) mode or an active jamming mode based on ROE or the commander’s intent. The maneuver part of the Evade tactic disperses the individual UAVs away from the threat and joins them back together to confuse the threat. The Evade tactic can be executed until the threat is gone, or for a specified time period as directed by the Swarm Commander through repetition of the Disperse play
每种策略的基础都是一系列戏剧。 每个战术由一个或多个传感器脚本和一个或多个机动脚本组成。 可修改的脚本参数（例如高度，速度，执行时间，偏移距离，地理限制和频率设置）使得脚本能够根据任务特定条件进行定制。 图5是组成Evade战术的脚本序列的一个例子。 Evade战术是由感知到的威胁引发的，这种威胁促成了同时发生的传感器和机动回路。 可以根据ROE或指挥官的意图将传感器设置为无源发射控制（EMCON）模式或主动干扰模式。 Evade战术的机动部分使各个无人机远离威胁并将它们连接在一起以混淆威胁。 Evade战术可以在威胁消失之前执行，或者在Swarm Commander指示的指定时间段内通过重复分散脚本来执行

Each play is composed of one or more swarm algorithms, which are the self-contained sequence of procedures to be performed by the swarm control software in order to execute a specific task. Within the context of the MASC framework, the swarm algorithms show the boundary wherein the operational architecture ends and the solution architecture begins. There is a considerable body of research in robotics algorithms, especially with regard to search and tracking functions. Typically, these optimization problems are treated as local, small-scale sub-spaces of the larger, global (NP-complete) problem in order to satisfy practical applications35. Many algorithms applicable for swarm systems, such as Boid’s flocking, ant colony optimization, and particle swarm optimization can be characterized as biologically-inspired or evolutionary36,37,38.
每个脚本由一个或多个群算法组成，这些算法是由群控制软件执行以执行特定任务的独立过程序列。 在MASC框架的上下文中，群体算法显示了操作体系结构终止且解决方案体系结构开始的边界。 机器人算法中有大量研究，特别是在搜索和跟踪功能方面。 通常，这些优化问题被视为较大的全局（NP完全）问题的局部小尺度子空间，以满足实际应用35。 适用于群体系统的许多算法，例如Boid的植绒，蚁群优化和粒子群优化，可以被描述为生物启发或进化36,37,38。

4 Conclusions and Future Work

MASC provides a mission-based, top-down
framework and modeling approach focused on developing a modular UAV swarm playbook-type architecture by decomposing a subset of US Navy missions, and developing tactics that can be applied to those missions. The framework is derived from a consolidation of heuristics from multiple academic disciplines and extends the swarm UAV field experimentation research conducted by ARSENL. This framework provides a standard methodology for designing swarm unmanned systems that will operate within military missions. To progress from remotely piloted single UAVs to remotely supervised UAV swarms, automation must be incorporated into the system architecture. By operating at the tactics level rather than the play level, Swarm Commanders will be aided with mission-based automation tools designed to facilitate scaling up the UAV-to-operator ratio to achieve large swarm (greater than 50) capability. Due to the common task patterns across missions (i.e. searching, tracking, communication relay), this architecture has the potential to be re-used across multiple mission types. Future work in this research area includes employing this framework into a prototype virtual environment, collecting feedback from prospective users, expanding the mission set, and defining specific algorithms for each play
4.结论和未来工作
MASC提供基于任务的自上而下的功能
框架和建模方法侧重于通过分解美国海军任务的一部分来开发模块化无人机群体剧本类型的架构，并开发可应用于这些任务的策略。该框架源自多个学科的启发式整合，并扩展了ARSENL进行的群体无人机现场实验研究。该框架提供了一种标准方法，用于设计将在军事任务中运行的群体无人系统。为了从遥控的单个无人机发展到远程监督的无人机群，必须将自动化整合到系统架构中。通过在战术层面而不是游戏层面进行操作，Swarm Commanders将配备基于任务的自动化工具，旨在帮助扩大无人机与操作员的比例，以实现大群（超过50）的能力。由于各任务的共同任务模式（即搜索，跟踪，通信中继），该架构有可能在多种任务类型中重复使用。该研究领域的未来工作包括将该框架应用于原型虚拟环境，收集潜在用户的反馈，扩展任务集，并为每个游戏定义特定算法