CN114428695A - Abnormal hierarchical processing method and system for behavior tree of group unmanned system - Google Patents

The invention discloses an exception grading processing method and system for a behavior tree of a group unmanned system. The method comprises S1, configuring a behavior tree engine on a console and a group unmanned aerial vehicle platform; s2, constructing a task tree and a platform behavior tree according to the business process; s3, when a platform behavior tree is constructed, a recovery node is constructed on a node of which the first abnormality occurrence probability reaches a first threshold value to form a local abnormality processing link, and an abnormality monitoring function is constructed on leaf nodes of the platform behavior tree and is used for abnormality monitoring and abnormality throwing; s4, placing a centralized exception handling link in an exception detection sub-thread started when a behavior tree thread is constructed; and S5, when the group unmanned system cooperatively executes the business process, the running state of the behavior tree on the unmanned system is fed back to the task tree in real time and can be used for task tree-level exception handling. The three-level layer-by-layer exception handling mode ensures that exceptions of each level of behavior tree can be handled timely, efficiently and intelligently when the three-level layer-by-layer exception handling mode is oriented to a complex business process of a group unmanned system.

一种针对群体无人系统行为树的异常分级处理方法及系统A kind of abnormal classification processing method and system for group unmanned system behavior tree

技术领域technical field

本发明涉及无人系统技术领域，更具体地说，特别涉及一种针对群体无人系统行为树的异常分级处理方法及系统。The present invention relates to the technical field of unmanned systems, and more particularly, to a method and system for abnormal classification processing for group unmanned system behavior trees.

背景技术Background technique

近年来，无人系统(以无人机、无人车等为载体的智能操作系统)发展迅速，并随着无人系统业务实施的范围增大，业务逻辑演变急剧复杂，单体无人系统已无法满足未来业务场景的需求，进而促使群体无人系统得到了大规模的爆发式发展。因为群体无人系统是以业务执行过程中扮演的角色为单位，一个业务角色通常会对应多个单体无人系统，而不同业务角色可能会有相同的具体应用动作，这对群体无人系统业务流程的可复用性提出了非常迫切的需求，因此一种新型管理业务执行流程的方法，行为树应运而生。In recent years, unmanned systems (intelligent operating systems based on unmanned aerial vehicles, unmanned vehicles, etc.) have developed rapidly, and with the expansion of the scope of unmanned system business implementation, the business logic has evolved rapidly and complicatedly. It has been unable to meet the needs of future business scenarios, which has led to the large-scale explosive development of group unmanned systems. Because the group unmanned system is based on the role played in the business execution process, a business role usually corresponds to multiple single unmanned systems, and different business roles may have the same specific application actions. The reusability of business processes presents a very urgent need, so a new method for managing business execution processes, behavior trees, emerges as the times require.

在行为树流程管理机制未出现之前，无人系统普遍采用有限状态机来对一套业务流程进行编排。有限状态机维护了一张图，图的节点是每个状态抽象的类，节点和节点的连线是状态间根据一定的规则做的状态转换，整个管理这些状态切换的载体就是有限状态机。这种实现机制就注定了有限状态机存在各个状态类之间相互依赖严重，耦合度很高，且结构不灵活，可扩展性不高等一系列缺陷。而行为树是将整套业务流程的每个具体动作搭建为一棵树，父节点是行为分支，叶节点是行为的具体表现。行为树将群体无人系统的角色行为包装为一个对象，符合面向对象的设计理念。行为树将行为逻辑和状态数据剥离，降低耦合，方便策划配置，而无需对每个无人系统的行为进行代码控制，提高了可视性，简化问题排查，提高效率。同时，借助行为树自身良好的可扩展性，可将群体无人系统的业务流程划分为多级行为树，层层嵌套。Before the emergence of the behavior tree process management mechanism, unmanned systems generally use finite state machines to orchestrate a set of business processes. The finite state machine maintains a graph. The nodes of the graph are the abstract classes of each state. The connection between the nodes and the nodes is the state transition between states according to certain rules. The entire carrier for managing these state transitions is the finite state machine. This implementation mechanism is destined to have a series of defects such as severe interdependence among various state classes, high coupling degree, inflexible structure, and low scalability of finite state machines. The behavior tree is to build each specific action of the entire business process into a tree, the parent node is the behavior branch, and the leaf node is the specific performance of the behavior. The behavior tree wraps the role behavior of the group unmanned system into an object, which conforms to the object-oriented design concept. The behavior tree strips behavior logic and state data, reduces coupling, facilitates planning and configuration, and does not require code control of the behavior of each unmanned system, which improves visibility, simplifies troubleshooting, and improves efficiency. At the same time, with the good scalability of the behavior tree itself, the business process of the group unmanned system can be divided into multi-level behavior trees, which are nested layer by layer.

通常，群体无人系统执行大业务时所涉及的机器平台数量较多，场景范围较广，无线通信链路极容易受到干扰甚至中断，导致无人平台与控制台失去联系，从而影响业务流程的正常运转；加之机器平台上装载的传感器数量和种类也比较繁多，而大多数传感器的应用环境都有比较严苛的条件限制，面对变幻莫测的场景环境，一个完备的行为树系统必须具备对各种可能发生的异常情况进行自我修复的能力，从而更高效、更智能地完成预期的整套业务流程。因此，为了使群体无人系统的控制核心行为树运转得更加稳定可靠，提出一种针对群体无人系统行为树的异常分级处理方法及系统成为当前亟待解决的技术难题。Usually, a large number of machine platforms are involved in the execution of large-scale business by a group unmanned system, and the scope of the scene is wide. The wireless communication link is easily disturbed or even interrupted, resulting in the loss of contact between the unmanned platform and the console, thus affecting the operation of the business process. In addition, the number and types of sensors loaded on the machine platform are also relatively large, and the application environment of most sensors has relatively strict conditions. Facing the unpredictable scene environment, a complete behavior tree system must have The ability to self-heal for various abnormal situations that may occur, so as to complete the expected set of business processes more efficiently and intelligently. Therefore, in order to make the control core behavior tree of the swarm unmanned system run more stably and reliably, it has become a technical problem to be solved urgently to propose an abnormal classification processing method and system for the behavior tree of the swarm unmanned system.

发明内容SUMMARY OF THE INVENTION

本发明的目的在于提供一种针对群体无人系统行为树的异常分级处理方法及系统，以克服现有技术所存在的缺陷。The purpose of the present invention is to provide an abnormal classification processing method and system for group unmanned system behavior tree, so as to overcome the defects existing in the prior art.

为了达到上述目的，本发明采用的技术方案如下：In order to achieve the above object, the technical scheme adopted in the present invention is as follows:

一种针对群体无人系统行为树的异常分级处理方法，包括以下步骤：An abnormal classification processing method for group unmanned system behavior tree, comprising the following steps:

S1、在控制台和群体无人机平台上配置行为树引擎；S1. Configure the behavior tree engine on the console and swarm drone platform;

S2、根据业务流程构造任务树和平台行为树，将所述任务树置于中心控制台主线程，构造异常检测子线程一，将所述平台行为树置于无人平台主线程，构造异常检测子线程二；S2. Construct a task tree and a platform behavior tree according to the business process, place the task tree in the main thread of the central console, construct an abnormality detection sub-thread 1, place the platform behavior tree in the unmanned platform main thread, and construct anomaly detection child thread two;

S3、构造平台行为树时，在第一异常发生概率达到第一阈值的节点上构建恢复节点，以形成局部异常处理环节，并在所述平台行为树叶子节点构建异常监视函数，用于异常的监视和异常抛出；S3. When constructing the platform behavior tree, construct a recovery node on the node where the probability of occurrence of the first abnormality reaches the first threshold to form a local abnormality processing link, and construct an abnormality monitoring function on the child node of the platform behavior tree, which is used for abnormal monitoring and exception throwing;

S4、构造平台行为树主线程时开启的异常检测子线程中放置集中式异常处理环节，所述平台行为树运行时发送运行状态至集中式异常处理环节的事件队列，待后续进入异常分析，对已发生事件的逻辑进行分析评判；且所述平台行为树叶子节点通过监视函数监视异常，若发现异常则将该异常发送至集中式异常处理环节的异常队列，并对所述异常队列进行异常过滤，对过滤结果进行评判，若评判的难易程度小于预设值则调用异常服务对异常进行异常处理，若难易程度大于预设值则调用综合信息进行异常分析，再调用异常服务进行异常处理；S4. A centralized exception processing link is placed in the abnormality detection sub-thread opened when the main thread of the platform behavior tree is constructed. When the platform behavior tree is running, the running status is sent to the event queue of the centralized exception processing link, and the exception analysis is performed later. The logic of the event that has occurred is analyzed and judged; and the platform behavior tree child node monitors the abnormality through the monitoring function, and if an abnormality is found, the abnormality is sent to the abnormality queue of the centralized abnormality processing link, and the abnormality queue is abnormally filtered. , to judge the filtering results. If the difficulty of the judgment is less than the preset value, the exception service is called to handle the exception. If the difficulty is greater than the preset value, the comprehensive information is called to analyze the exception, and then the exception service is called to handle the exception. ;

S5、群体无人系统以集中式协同执行业务流程时，无人系统上的平台行为树运行状态实时反馈给任务树，若无人系统端因通信中断而与控制台或其他无人系统失联时，触发行为树的异常处理，将当前执行节点切换为备用节点，所述备用节点在评估自身状态时与控制台尝试重连；若任务树检测到状态同步超时，触发任务树的异常处理，将当前执行的行为树子树切换为备用节点并提示人为干预；当无人系统与控制台恢复通信后，所述任务树重新同步集群的行为树运行状态，并对业务流程的下一步进行重新规划。S5. When the group unmanned system executes the business process in a centralized manner, the running status of the platform behavior tree on the unmanned system is fed back to the task tree in real time. If the unmanned system end is disconnected from the console or other unmanned systems due to communication interruption When the exception handling of the behavior tree is triggered, the current execution node is switched to a standby node, and the standby node attempts to reconnect with the console when evaluating its own state; Switch the currently executed behavior tree subtree to the standby node and prompt human intervention; when the unmanned system and the console resume communication, the task tree resynchronizes the behavior tree running state of the cluster, and re-runs the next step of the business process. planning.

进一步地，所述步骤S3中的恢复节点为具有两个子节点的控制流类型节点，所述恢复节点的控制逻辑为：仅第一子节点返回成功时该恢复节点返回成功；仅在第一子节点返回失败时执行第二子节点；第二子节点用于恢复操作，若恢复操作成功，则再次执行第一子节点。Further, the recovery node in the step S3 is a control flow type node with two child nodes, and the control logic of the recovery node is: only when the first child node returns successfully; the recovery node returns successfully; only when the first child node returns successfully; When the node fails to return, execute the second child node; the second child node is used for the recovery operation, and if the recovery operation succeeds, execute the first child node again.

进一步地，所述集中式异常处理环节在初始化时，对集中式异常的异常描述符进行配置。Further, during initialization, the centralized exception processing link configures the exception descriptor of the centralized exception.

进一步地，所述集中式异常处理环节中，在无人平台行为树的叶子节点中放置异常检测的监视函数，若行为发生异常时，执行监视函数通知集中式异常处理环节，将异常请求发送到集中式异常处理环节的异常队列，行为树引擎将对当前执行的叶子节点唯一标识ID予以保护。Further, in the centralized exception handling link, a monitoring function for anomaly detection is placed in the leaf node of the unmanned platform behavior tree. If the behavior is abnormal, the monitoring function is executed to notify the centralized exception handling link, and the exception request is sent to. In the exception queue of the centralized exception processing link, the behavior tree engine will protect the unique ID of the currently executed leaf node.

进一步地，所述步骤S4中的异常描述符为：一个异常所需要的资源都集中在这个结构体描述中，该结构体会在行为树初始化的时候被分配到一个数组中存放，其中数组下标代表的是异常代码。Further, the exception descriptor in the step S4 is: the resources required by an exception are concentrated in the description of the structure, and the structure is allocated to an array for storage when the behavior tree is initialized, where the subscript of the array is Represents the exception code.

进一步地，所述步骤S4的异常队列的底层数据结构为环形双端队列，该环形双端队列的头部用于读取和存储最早发出的请求，环形双端队列的尾部用于下个写入请求。Further, the underlying data structure of the abnormal queue in the step S4 is a circular deque, the head of the circular deque is used to read and store the earliest sent request, and the tail of the circular deque is used for the next write. input request.

进一步地，所述步骤S4中的事件队列用于存储平台行为树的运行状态，所述事件队列用于集中式异常处理环节对异常作出综合信息判断。Further, the event queue in the step S4 is used to store the running state of the platform behavior tree, and the event queue is used for the centralized exception processing link to make comprehensive information judgment on exceptions.

进一步地，所述步骤S4中的异常服务中存储异常列表，每个异常对应有异常处理程序，所述异常列表包括通用异常和用户注册异常，所述异常处理为根据异常代码调用对应的异常处理程序。Further, the exception service in the step S4 stores an exception list, each exception corresponds to an exception handling program, the exception list includes general exceptions and user registration exceptions, and the exception processing is to call the corresponding exception processing according to the exception code. program.

进一步地，在所述异常处理程序完成处理后，根据引起异常的事件类型将控制权返回给当前叶子节点、将控制权返回给下一个叶子节点或终止发生异常的节点。Further, after the exception handler completes processing, the control right is returned to the current leaf node, the control right is returned to the next leaf node, or the node where the exception occurred is terminated according to the event type that caused the exception.

本发明还提供一种根据上述的针对群体无人系统行为树的异常分级处理方法，包括：The present invention also provides a kind of abnormal classification processing method for group unmanned system behavior tree according to the above, including:

配置模块，用于在控制台和群体无人机平台上配置行为树引擎；Configuration module for configuring the behavior tree engine on console and swarm drone platforms;

构造模块，用于根据业务流程构造任务树和平台行为树，为平台行为树叶子节点构造异常监视函数，将所述任务树置于中心控制台主线程，构造异常检测子线程一，将所述平台行为树置于平台主线程，并构造异常检测子线程二；The construction module is used to construct a task tree and a platform behavior tree according to the business process, construct an anomaly monitoring function for the child nodes of the platform behavior tree, place the task tree on the main thread of the central console, construct an anomaly detection sub-thread one, and place the The platform behavior tree is placed in the main thread of the platform, and an anomaly detection sub-thread 2 is constructed;

局部异常处理模块，用于构造平台行为树时在第一异常发生概率达到第一阈值的节点上构建恢复节点，以形成局部异常处理环节；The local exception handling module is used to construct a recovery node on the node where the probability of occurrence of the first exception reaches the first threshold when constructing the platform behavior tree, so as to form a local exception handling link;

集中式异常处理模块，用于在构造平台行为树主线程时开启的异常检测子线程中放置集中式异常处理环节，所述平台行为树运行时发送运行状态至集中式异常处理环节的事件队列，待后续进入异常分析，对已发生事件的逻辑进行分析评判；且所述平台行为树叶子节点通过监视函数监视异常，若发现异常则将该异常发送至集中式异常处理环节的异常队列，并对所述异常队列进行异常过滤，对过滤结果进行评判，若评判的难易程度小于预设值则调用异常服务对异常进行异常处理，若难易程度大于预设值则调用综合信息进行异常分析，再调用异常服务进行异常处理；The centralized exception processing module is used to place a centralized exception processing link in the exception detection sub-thread opened when the main thread of the platform behavior tree is constructed, and the platform behavior tree sends the running status to the event queue of the centralized exception processing link when the platform behavior tree is running. After entering into the exception analysis later, analyze and judge the logic of the event that has occurred; and the platform behavior tree child node monitors the exception through the monitoring function, and if an exception is found, the exception is sent to the exception queue of the centralized exception processing link, and the The abnormality queue performs abnormal filtering, and judges the filtering results. If the difficulty level of the judgment is less than a preset value, the abnormality service is called to process the abnormality. If the difficulty level is greater than the preset value, the comprehensive information is called to analyze the abnormality. Then call the exception service for exception handling;

任务树级异常处理模块，用于群体无人系统以集中式协同执行业务流程时，平台行为树运行状态实时反馈给任务树，若无人系统端因通信中断而与控制台或其他无人系统失联时，触发行为树的异常处理，将当前执行节点切换为备用节点，所述备用节点在评估自身状态时与控制台尝试重连；若任务树检测到状态同步超时，触发任务树的异常处理，将当前执行的行为树子树切换为备用节点并提示人为干预；当无人系统与控制台恢复通信后，所述任务树重新同步集群的行为树运行状态，并对业务流程的下一步进行重新规划。The task tree-level exception handling module is used when the group unmanned system executes business processes in a centralized manner, and the running status of the platform behavior tree is fed back to the task tree in real time. When the connection is lost, the abnormal processing of the behavior tree is triggered, and the current execution node is switched to the standby node, and the standby node attempts to reconnect with the console when evaluating its own state; if the task tree detects that the state synchronization has timed out, the abnormality of the task tree is triggered Processing, switching the currently executed behavior tree subtree to the standby node and prompting human intervention; when the unmanned system and the console resume communication, the task tree re-synchronizes the behavior tree running state of the cluster, and the next step of the business process is Replanning.

与现有技术相比，本发明的优点在于：本发明通过三个层级异常处理的逐层处理模式，群体无人系统执行复杂业务流程时，各级行为树发生的异常都得到了及时高效、智能地处理。Compared with the prior art, the present invention has the advantages that: the present invention adopts the layer-by-layer processing mode of three-level exception processing, when the group unmanned system executes the complex business process, the abnormality that occurs in the behavior tree at all levels is timely, efficient, and effective. Handle intelligently.

附图说明Description of drawings

为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

图1是本发明针对群体无人系统行为树的异常分级处理的三个层级结构图。FIG. 1 is a three-level structure diagram of anomaly classification processing for group unmanned system behavior tree according to the present invention.

图2是本发明局部异常处理环节的恢复节点示意图。FIG. 2 is a schematic diagram of a recovery node of a local exception processing link of the present invention.

图3是本发明集中式异常处理环节的流程示意图。FIG. 3 is a schematic flow chart of the centralized exception processing link of the present invention.

图4是本发明任务树级异常处理环节的流程示意图。FIG. 4 is a schematic flowchart of a task tree-level exception processing link of the present invention.

图5是本发明针对群体无人系统行为树的异常分级处理系统的原理图。FIG. 5 is a schematic diagram of the abnormal classification processing system for group unmanned system behavior tree according to the present invention.

具体实施方式Detailed ways

下面结合附图对本发明的优选实施例进行详细阐述，以使本发明的优点和特征能更易于被本领域技术人员理解，从而对本发明的保护范围做出更为清楚明确的界定。The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, and the protection scope of the present invention can be more clearly defined.

参阅图1-图4所示，本实施例公开了一种针对群体无人系统行为树的异常分级处理方法，包括以下步骤：Referring to FIG. 1-FIG. 4, the present embodiment discloses an abnormal classification processing method for a group unmanned system behavior tree, including the following steps:

步骤S1、在控制台和群体无人机平台上配置行为树引擎，包括行为树内和树间的通信机制。Step S1, configure the behavior tree engine on the console and the swarm UAV platform, including the communication mechanism within the behavior tree and between trees.

步骤S2、根据业务流程构造任务树和平台行为树，将所述任务树置于中心控制台主线程，构造异常检测子线程一，将所述平台行为树置于无人平台主线程，构造异常检测子线程二。Step S2, construct a task tree and a platform behavior tree according to the business process, place the task tree in the main thread of the central console, construct an abnormality detection sub-thread one, place the platform behavior tree in the unmanned platform main thread, and construct an abnormality Detect child thread two.

步骤S3、构造平台行为树时，在第一异常发生概率达到第一阈值的节点上构建恢复节点，以形成局部异常处理环节，并在所述平台行为树叶子节点构建异常监视函数，用于异常的监视和异常抛出；Step S3, when constructing the platform behavior tree, construct a recovery node on the node where the probability of occurrence of the first abnormality reaches the first threshold to form a local abnormality processing link, and construct an abnormality monitoring function on the child node of the platform behavior tree for abnormality. monitoring and exception throwing;

其中，恢复节点为具有两个子节点的控制流类型节点，恢复节点的控制逻辑为：当且仅当第一子节点返回成功时该恢复节点返回成功；仅在第一子节点返回失败时执行第二子节点；第二子节点用于恢复操作，例如重新初始化系统或其他恢复行为，若恢复操作成功，则再次执行第一子节点。Among them, the recovery node is a control flow type node with two child nodes, and the control logic of the recovery node is: if and only when the first child node returns successfully, the recovery node returns successfully; only when the first child node fails to return, execute the first Two child nodes; the second child node is used for recovery operations, such as reinitializing the system or other recovery behaviors, and if the recovery operation is successful, the first child node is executed again.

具体的，用户可以指定在返回失败之前应该执行多少次恢复操作，以此来控制局部异常的处理时间。Specifically, the user can specify how many recovery operations should be performed before returning a failure, so as to control the processing time of local exceptions.

步骤S4、构造平台行为树主线程时开启的异常检测子线程中放置集中式异常处理环节，平台行为树运行时，为了观测整个行为树的运行状态，每个节点发送事件至集中式异常处理环节的事件队列，使得异常检测环节对每个节点都有所观测，若某个或多个节点通过监视函数发现异常则将该异常发送至集中式异常处理环节的异常队列，并对异常队列进行异常过滤，对过滤结果进行评判(多节点相同请求，不同节点请求冲突等情况)，若评判的难易程度小于预设值(可以为简单易处理)则调用异常服务对异常进行异常处理，若难易程度大于预设值(可以为复杂)则调用综合信息进行异常分析，再调用异常服务进行异常处理。In step S4, a centralized exception processing link is placed in the exception detection sub-thread opened when the main thread of the platform behavior tree is constructed. When the platform behavior tree is running, in order to observe the running state of the entire behavior tree, each node sends an event to the centralized exception processing link. If one or more nodes find an exception through the monitoring function, the exception will be sent to the exception queue of the centralized exception processing link, and the exception queue will be abnormal. Filter, judge the filtering results (multiple nodes have the same request, different nodes have conflicting requests, etc.), if the difficulty of the judgment is less than the preset value (it can be simple and easy to handle), call the exception service to handle the exception. If the degree of ease is greater than the preset value (it can be complex), the comprehensive information is called for exception analysis, and then the exception service is called for exception handling.

本实施例中，集中式异常处理环节的主要工作流程如下：In this embodiment, the main workflow of the centralized exception handling link is as follows:

注册异常，在系统初始化时，用户需自行完成对此类异常的注册，主要包含异常描述符的配置。异常描述符：一个异常所需要的资源都集中在这个结构体描述中，该结构体会在平台行为树初始化的时候被分配到一个数组中存放，其中数组下标代表的是异常代码。Register exceptions. When the system is initialized, users need to complete the registration of such exceptions by themselves, mainly including the configuration of exception descriptors. Exception descriptor: The resources required by an exception are concentrated in the description of this structure, which is allocated to an array when the platform behavior tree is initialized, and the array subscript represents the exception code.

异常检测，注册异常和异常检测需配合使用，异常检测的监视函数位于平台行为树的叶子节点，当行为发生异常时，执行监视函数通知集中式异常处理环节，将异常请求发送到处理环节的异常队列，并让异常处理环节作出相应处理。异常抛出的同时，行为树引擎将对当前执行的叶子节点唯一标识ID做现场保护，以便异常处理结束时恢复运行。Anomaly detection, registration anomaly and anomaly detection need to be used together. The monitoring function of anomaly detection is located in the leaf node of the platform behavior tree. When an abnormal behavior occurs, the monitoring function is executed to notify the centralized exception processing link, and the exception request is sent to the exception processing link. Queue, and let the exception handling process deal with it accordingly. When the exception is thrown, the behavior tree engine will protect the unique ID of the currently executed leaf node on the spot, so that the operation can be resumed when the exception processing ends.

异常队列，异常队列的底层数据结构为环形双端队列，队列的头部是读取请求的地方，存储最早发出的请求；尾部是下个写入请求的地方。异常队列主要解决以下问题：异常消息处理不得阻塞调用者；请求可以合并处理；接受请求和处理请求解耦，使得两者可异步处理；优先级请求；单播多写入队列，只有集中式异常处理环节可以读取，其他节点可以写入。Exception queue, the underlying data structure of the exception queue is a circular double-ended queue, the head of the queue is where the read request is stored, and the earliest issued request is stored; the tail is where the next write request is. The exception queue mainly solves the following problems: exception message processing must not block the caller; requests can be combined and processed; accepting requests and processing requests are decoupled, so that the two can be processed asynchronously; priority requests; unicast multi-write queues, only centralized exceptions The processing link can read, and other nodes can write.

事件队列，事件队列存储平台行为树的运行状态，其他对象可以针对这个事件做出回应，供集中式异常处理环节对异常作出综合信息判断。The event queue stores the running status of the platform behavior tree, and other objects can respond to this event for the centralized exception processing link to make comprehensive information judgments on exceptions.

异常过滤器，对异常队列的所有异常请求进行过滤，进行异常的分析，比如重复异常的请求合并，简单异常以及复杂异常的判断。The exception filter filters all exception requests in the exception queue and analyzes exceptions, such as merging repeated exception requests, and judging simple exceptions and complex exceptions.

异常服务，存储着异常列表，针对每个异常代码都有相应的异常处理程序，供异常处理环节读取请求的异常之后调用服务接口作出处理(服务被限制在集中式异常处理线程中，只允许该环节访问)。异常列表分为系统通用异常和用户注册异常两部分，其中通用系统异常为一些常见的通用性异常，系统运行前需事先定义；而用户注册的面向具体业务流程的异常，可与监视函数配合使用。The exception service stores the exception list, and has a corresponding exception handler for each exception code, for the exception processing link to read the requested exception and then call the service interface for processing (the service is limited to the centralized exception processing thread, and only allows visit this link). The exception list is divided into two parts: system general exceptions and user registration exceptions. The general system exceptions are some common general exceptions, which need to be defined before the system runs. The exceptions registered by users for specific business processes can be used in conjunction with monitoring functions. .

异常处理，主要是根据不同的异常代码，调用相应异常处理程序解决异常的过程。异常处理程序执行时，可根据异常等级确认是否同时暂停所有正在执行的行为树分支，启动安全模式(一棵提前设置好的行为树子树)，让平台进入安全模式，防止陷入混乱。Exception handling is mainly the process of calling corresponding exception handlers to resolve exceptions according to different exception codes. When the exception handler is executed, it can confirm whether to suspend all the executing behavior tree branches at the same time according to the exception level, start the safe mode (a subtree of the behavior tree set in advance), and let the platform enter the safe mode to prevent confusion.

控制权返回，当异常处理程序完成处理后，根据引起异常的事件类型，会引发下列三种情况中的一种：处理程序将控制权给当前叶子节点(当事件发生时正在执行的叶子节点)；处理程序将控制权返回给下一个叶子节点(当事件发生时正在执行的下一个叶子节点)；处理程序终止发生异常的节点。Control returns. When the exception handler finishes processing, one of the following three situations will be triggered depending on the type of event that caused the exception: The handler gives control to the current leaf node (the leaf node that was executing when the event occurred) ; The handler returns control to the next leaf node (the one that was executing when the event occurred); the handler terminates the node where the exception occurred.

步骤S5、群体无人系统以集中式协同执行业务流程时，平台行为树运行状态实时反馈给任务树，若无人系统端因通信中断而与控制台或其他无人系统失联时，自身无法完成状态同步，因此会触发行为树的异常处理，将当前执行节点切换为备用节点，备用节点在评估自身状态时与控制台尝试重连；若控制台任务树检测到状态同步超时，触发任务树的异常处理，将当前执行的行为树子树切换为备用节点并提示人为干预；当无人系统与控制台恢复通信后，控制台任务树重新同步集群的行为树运行状态，并对业务流程的下一步进行重新规划。Step S5, when the group unmanned system executes the business process in a centralized manner, the running status of the platform behavior tree is fed back to the task tree in real time. The status synchronization is completed, so the exception processing of the behavior tree will be triggered, and the current execution node will be switched to the standby node. The standby node will try to reconnect with the console when evaluating its own status; if the console task tree detects that the status synchronization has timed out, the task tree will be triggered When the unmanned system resumes communication with the console, the console task tree re-synchronizes the running status of the cluster's behavior tree, and the operation status of the business process is updated. The next step is to re-plan.

本发明还提供一种针对上述群体无人系统行为树的异常分级处理方法的系统，包括：配置模块1，用于在每个单体无人系统和控制台上配置行为树引擎；构造模块2，用于根据业务流程构造任务树和平台行为树，为平台行为树叶子节点构造异常监视函数，将所述任务树置于中心控制台主线程，构造异常检测子线程一，将所述平台行为树置于平台主线程，并构造异常检测子线程二；局部异常处理模块3，用于构造平台行为树时在第一异常发生概率达到第一阈值的节点上构建恢复节点，以形成局部异常处理环节；集中式异常处理模块4，用于在构造平台行为树主线程时开启的异常检测子线程中放置集中式异常处理环节，所述平台行为树运行时发送运行状态至集中式异常处理环节的事件队列，待后续进入异常分析，对已发生事件的逻辑进行分析评判；且所述平台行为树叶子节点通过监视函数监视异常，若发现异常则将该异常发送至集中式异常处理环节的异常队列，并对所述异常队列进行异常过滤，对过滤结果进行评判，若评判的难易程度小于预设值则调用异常服务对异常进行异常处理，若难易程度大于预设值则调用综合信息进行异常分析，再调用异常服务进行异常处理；任务树级异常处理模块5，用于群体无人系统以集中式协同执行业务流程时，平台行为树运行状态实时反馈给任务树，若无人系统端因通信中断而与控制台或其他无人系统失联时，触发行为树的异常处理，将当前执行节点切换为备用节点，所述备用节点在评估自身状态时与控制台尝试重连；若任务树检测到状态同步超时，触发任务树的异常处理，将当前执行的行为树子树切换为备用节点并提示人为干预；当无人系统与控制台恢复通信后，所述任务树重新同步集群的行为树运行状态，并对业务流程的下一步进行重新规划。The present invention also provides a system for the abnormal classification processing method for the above-mentioned group unmanned system behavior tree, comprising: a configuration module 1 for configuring a behavior tree engine on each single unmanned system and a console; a construction module 2 , for constructing a task tree and a platform behavior tree according to the business process, constructing an anomaly monitoring function for the child nodes of the platform behavior tree, placing the task tree in the main thread of the central console, constructing an anomaly detection sub-thread one, and converting the platform behavior The tree is placed in the main thread of the platform, and an abnormality detection sub-thread 2 is constructed; the local exception processing module 3 is used to construct a recovery node on the node where the probability of occurrence of the first exception reaches the first threshold when constructing the platform behavior tree, so as to form a local exception handling The centralized exception processing module 4 is used to place a centralized exception processing link in the exception detection sub-thread opened when the main thread of the platform behavior tree is constructed. The event queue, after entering into the exception analysis, analyzes and judges the logic of the event that has occurred; and the platform behavior tree child node monitors the exception through the monitoring function, and if an exception is found, the exception is sent to the exception queue of the centralized exception processing link , and perform abnormal filtering on the abnormal queue, and evaluate the filtering results. If the degree of difficulty of the evaluation is less than the preset value, the abnormal service is invoked to handle the abnormality. If the degree of difficulty is greater than the preset value, the comprehensive information is invoked. Exception analysis, and then call the exception service for exception processing; task tree level exception processing module 5, when the group unmanned system executes the business process in a centralized manner, the running status of the platform behavior tree is fed back to the task tree in real time. When the connection with the console or other unmanned systems is lost due to communication interruption, the exception handling of the behavior tree is triggered, and the current execution node is switched to the standby node. The standby node attempts to reconnect with the console when evaluating its own status; if the task The tree detects that the state synchronization has timed out, triggers the exception handling of the task tree, switches the currently executed behavior tree subtree to the standby node and prompts human intervention; when the unmanned system and the console resume communication, the task tree resynchronizes the cluster's The behavior tree runs the state and re-plans the next step of the business process.

本发明将群体无人系统行为树的异常处理机制分为三个层级，由低到高分别为：局部异常自处理环节，集中式异常处理环节，任务树级异常处理环节。局部异常自处理环节用于解决一些平台行为树级的简单异常，凭借基本的传感器重启，等待，参数重置等预设操作就能得到解决。此类异常可通过构建平台行为树时添加恢复节点来处理，将异常处理的权限下放至局部节点，可提高简单异常事件的处理效率，若恢复节点无法对该异常进行修复，则将该异常提交至集中式异常处理环节进行处理。集中式异常处理环节用于收集局部异常自处理环节无法解决的异常问题，并进行统一处理。其中，较为简单的异常事件，可通过异常过滤器和异常请求服务，获得异常处理入口，来具体解决异常；而针对比较复杂的异常事件(例如，多项异常事件存在关联关系，因果关系等)，则需要先经过异常分析器的预处理，再进一步请求异常服务。通常，单体无人系统上行为树运转时发生的异常情况可通过上述两层异常处理机制进行主观能动的自我消化和恢复，而面对群体协同业务流程时，则需要任务树级异常处理环节的参与。任务树级异常处理环节用于解决无人平台与中心控制台间通信中断导致任务树运转发生异常，或是单体无人系统上集中式异常处理环节产生难以处理的异常的情况，通过启动备用节点或人工干预的方式，强制执行最高优先级的异常处理方案。本申请通过三个层级异常处理模块的逐层处理，群体无人系统执行复杂业务流程时，各级行为树发生的异常都得到了及时高效、智能地处理。The invention divides the exception handling mechanism of the group unmanned system behavior tree into three levels, from low to high: local exception self-processing link, centralized exception processing link, and task tree-level exception processing link. The local exception self-handling link is used to solve some simple exceptions at the platform behavior tree level. It can be solved by basic sensor restart, wait, parameter reset and other preset operations. Such exceptions can be handled by adding a recovery node when building the platform behavior tree. Delegating the exception handling authority to a local node can improve the processing efficiency of simple exception events. If the exception cannot be repaired by the recovery node, the exception will be submitted. To the centralized exception handling link for processing. The centralized exception processing link is used to collect exception problems that cannot be solved by the local exception self-handling link, and handle them uniformly. Among them, for relatively simple abnormal events, the exception processing entry can be obtained through the exception filter and exception request service to solve the exception; for more complex abnormal events (for example, multiple abnormal events are correlated, causal, etc.) , it needs to be preprocessed by the exception analyzer before further requesting the exception service. Usually, the abnormal situation that occurs during the operation of the behavior tree on a single unmanned system can be subjectively digested and recovered through the above two-layer exception handling mechanism, while in the face of group collaborative business processes, a task tree-level exception handling link is required. Participation. The task tree-level exception handling link is used to solve the problem that the operation of the task tree is abnormal due to the communication interruption between the unmanned platform and the central console, or the centralized exception handling link on a single unmanned system produces an intractable exception. Node or human intervention to enforce the highest priority exception handling scheme. Through the layer-by-layer processing of the three-level exception processing modules in the present application, when the group unmanned system executes complex business processes, the exceptions that occur in the behavior trees at all levels are handled in a timely, efficient and intelligent manner.

虽然结合附图描述了本发明的实施方式，但是专利所有者可以在所附权利要求的范围之内做出各种变形或修改，只要不超过本发明的权利要求所描述的保护范围，都应当在本发明的保护范围之内。Although the embodiments of the present invention are described in conjunction with the accompanying drawings, the patent owner can make various changes or modifications within the scope of the appended claims, as long as the protection scope described in the claims of the present invention is not exceeded, all should be within the protection scope of the present invention.