WO2024098438A1 - 多智能体超视距网联协同感知动态决策方法及相关装置 - Google Patents

多智能体超视距网联协同感知动态决策方法及相关装置 Download PDF

Info

Publication number
WO2024098438A1
WO2024098438A1 PCT/CN2022/131637 CN2022131637W WO2024098438A1 WO 2024098438 A1 WO2024098438 A1 WO 2024098438A1 CN 2022131637 W CN2022131637 W CN 2022131637W WO 2024098438 A1 WO2024098438 A1 WO 2024098438A1
Authority
WO
WIPO (PCT)
Prior art keywords
path
agent
obstacle
time
information
Prior art date
Application number
PCT/CN2022/131637
Other languages
English (en)
French (fr)
Inventor
魏平
杨腾
程翔
李鹏
刘克勤
辛景民
郑南宁
Original Assignee
西安交通大学
北京大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 西安交通大学, 北京大学 filed Critical 西安交通大学
Publication of WO2024098438A1 publication Critical patent/WO2024098438A1/zh

Links

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0231—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means
    • G05D1/0238—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means using obstacle or wall sensors
    • G05D1/024—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means using obstacle or wall sensors in combination with a laser
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0212—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory
    • G05D1/0219—Control of position or course in two dimensions specially adapted to land vehicles with means for defining a desired trajectory ensuring the processing of the whole working surface

Definitions

  • the present invention relates to the field of robot perception and multi-agent collaboration, and in particular to a multi-agent beyond-visual-range networked collaborative perception dynamic decision-making method and related devices.
  • the purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a multi-agent beyond-visual-range networked collaborative perception dynamic decision-making method and related devices.
  • a point cloud-based multi-agent beyond-horizon networked collaborative perception dynamic decision-making method includes the following steps:
  • Step 1 Build a three-dimensional scene map to locate the agent in real time
  • Step 2 After obtaining the original point cloud, the obstacle point cloud is obtained through downsampling, ground segmentation, coordinate system transformation and scene background separation preprocessing operations, and then the obstacles outside the original scene are clustered by the Euclidean clustering method;
  • Step 3 Build a collaborative perception dynamic decision-making scheduling system based on a centralized communication framework.
  • the scheduling system introduces an agent obstacle list and a global abnormal path list based on real-time agent operation status information, positioning information, and obstacle information.
  • a map of executable paths is maintained by traversing the global topological path and the global abnormal path.
  • global path planning for multi-agent collaboration is achieved, and the path is broadcast to achieve the combination of collaborative perception and path planning.
  • Step 4 After the intelligent agent receives the broadcasted path information, it executes path following with dynamic speed adjustment based on time window optimization, switches to obstacle avoidance state according to real-time obstacle detection information, or uploads abnormal information to perform dynamic replanning, and sends real-time intelligent agent operation status information, positioning information, and obstacle information.
  • the scene background in step 2 is divided into three-dimensional grid maps by obtaining a global scene map, removing the ground and top, and filtering the real-time point cloud according to the three-dimensional grid map to separate obstacles from the scene; according to the geometric coordinate information of each cluster, the bounding box of each obstacle perpendicular to the coordinate system is obtained.
  • each agent senses obstacles in the surrounding environment and uploads obstacle information to the scheduling system; the scheduling system maintains an obstacle list for each agent, the obstacle list stores the position and bounding box size information of each obstacle in the global map uploaded by the current agent in real time, and judges the occupation of the global topological path according to the position and bounding box size information of the obstacles in each obstacle list, adds the occupied path segment to the abnormal path list, and maintains a real-time global abnormal path list based on each obstacle list, and the real-time global abnormal path list stores all the path segments occupied by obstacles in the global topological path;
  • the scheduling system maintains a real-time topological map of executable paths by traversing the global topological path and the global exception path;
  • the scheduling system When the scheduling system receives input of the starting and ending points of a task, it first assigns the task to the nearest idle agent, uses the topological map of the real-time executable path as the map input, executes the shortest path planning algorithm and performs time window allocation, completes the path planning, and sends the path to the agent.
  • the scheduling system when the scheduling system traverses the obstacle information uploaded by the current agent, it also traverses the positioning information of other online agents to determine whether the position of other agents is within the bounding box of the currently traversed obstacle; if so, the currently traversed obstacle is identified as an agent and is not added to the current agent's obstacle list.
  • the dispatching system receives the location and status information of the agent, registers information for each agent, and performs task allocation and path planning based on the location and status information of the agent;
  • a path planning algorithm that reuses space and time is used in path planning.
  • a time window mechanism based on path segments is introduced on the basis of the shortest path planning. Multiple paths are retained during the shortest path planning. According to the time window allocation, the shortest path is used as the planned path.
  • step 4 when the agent in step 4 performs path following or obstacle avoidance, if the agent encounters an obstacle blocking the path ahead and cannot avoid it, it will stop driving and set the state to abnormal, and upload the state information at the same time;
  • the scheduling system When the scheduling system receives abnormal information, it traverses the current tasks of all agents and compares them with the abnormal path list. Based on the comparison results, it re-plans feasible paths for all agents whose tasks contain abnormal paths, and at the same time sends the task paths to each agent to realize the function of multi-vehicle weight planning.
  • the scheduling system replans the path of the agent.
  • the intelligent body detects in real time whether there are obstacles blocking the current path, and if not, it drives normally;
  • the obstacle avoidance algorithm is called to calculate the local running trajectory, plan a local obstacle avoidance path, calculate the time cost of running the obstacle avoidance path at the maximum speed, and at the same time calculate the time cost of running the original path at the maximum speed; compare the time cost with the time to reach the target point planned by the scheduling system. If it can be arrived within the planned time, the obstacle avoidance algorithm is executed to avoid the obstacle; if it cannot be arrived within the planned time or the obstacle cannot be avoided, the driving is suspended, the state of the intelligent body is adjusted, and re-planning is triggered.
  • the agent calculates the real-time required speed based on the time point of the next target point and the real-time distance to the next target point, and dynamically adjusts the speed of the agent during driving, so that the agent can complete the operation task on the current path within the time window planned by the scheduling system, ensuring that the agent travels according to the planned time window and path.
  • a point cloud-based multi-agent beyond-horizon networked collaborative perception dynamic decision-making device comprising a three-dimensional scene map construction module, an original point cloud processing module, a scheduling system construction module and an agent execution module;
  • the three-dimensional scene map construction module is used to construct a three-dimensional scene map to locate the intelligent agent in real time;
  • the original point cloud processing module is used to obtain the original point cloud, obtain the obstacle point cloud through downsampling, ground segmentation, coordinate system transformation and scene background separation preprocessing operations, and then cluster the obstacles outside the original scene through the Euclidean clustering method;
  • the scheduling system building module is used to build a scheduling system based on the centralized communication framework collaborative perception dynamic decision-making.
  • the scheduling system introduces the agent obstacle list and the global abnormal path list based on the real-time agent operation status information, positioning information and obstacle information.
  • a map of the execution path is maintained by traversing the global topological path and the global abnormal path.
  • the global path planning of multi-agent collaboration is realized, and the path is broadcast to realize the combination of collaborative perception and path planning;
  • the intelligent agent execution module is used to receive the broadcast path information, perform path following with dynamic speed adjustment based on time window optimization, switch to obstacle avoidance state or upload abnormal information to perform dynamic replanning according to real-time obstacle detection information, and send real-time intelligent agent operation status information, positioning information and obstacle information.
  • a computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making method are implemented.
  • a computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the above-mentioned point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making method.
  • the point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making method of the present invention combines collaborative perception with path planning by maintaining the obstacle list of the agent and the global abnormal path list, and can realize multi-agent collaborative perception and multi-agent collaborative planning and scheduling tasks.
  • the introduction of a re-planning mechanism enables the agent system to make dynamic decisions based on the perceived information, thereby improving its adaptability to dynamic and complex environments.
  • the present invention can ensure that multiple agents can dynamically, collaboratively, and efficiently complete driving tasks.
  • a path following algorithm for time window path planning is introduced, which can dynamically adjust the running speed according to time conditions and provide control support for the collaborative path planning algorithm.
  • an improved obstacle avoidance algorithm when encountering an obstacle, it will first determine whether the intelligent agent can avoid the obstacle autonomously. If the obstacle cannot be avoided, re-planning will be performed, which reduces the frequency of re-planning and improves the operating efficiency of the intelligent agent.
  • the point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making device of the present invention includes a specific module for completing the above-mentioned dynamic decision-making method.
  • the computer device and storage medium for point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making are used to implement the specific steps of the above-mentioned decision-making method.
  • FIG1 is a framework flow chart of the decision-making method of the present invention.
  • FIG2 is a diagram of a perception and positioning framework used in the present invention.
  • FIG3 is a three-dimensional mapping visualization result of the present invention.
  • FIG4 is a diagram of the motion control framework of an intelligent body according to the present invention.
  • FIG5 is a flow chart showing the interaction between each agent and the scheduling system during runtime
  • FIG6 is a path planning framework diagram of the dispatching system of the present invention.
  • FIG7 is a schematic diagram of a path replanning framework of a dispatching system according to the present invention.
  • FIG8 is a schematic diagram of an executable path maintenance framework of the scheduling system of the present invention.
  • FIG9 is a block diagram of a point cloud-based multi-agent beyond-horizon networked collaborative perception dynamic decision-making device according to the present invention.
  • FIG. 10 is a schematic diagram of a computer device in an embodiment of the present invention.
  • a point cloud-based multi-agent over-the-horizon networked collaborative perception dynamic decision-making method is provided.
  • the framework flow chart of the specific implementation of the present invention is shown in Figure 1.
  • the intelligent agent performs three-dimensional obstacle detection, mapping, and positioning to provide real-time environmental perception, static maps, and real-time positioning information support for the dispatching system;
  • the dispatching system implements the functions of multi-vehicle global path planning and dynamic re-planning, and sends the planning results to the intelligent agent; after obtaining the planning results, the intelligent agent performs path following and obstacle avoidance based on real-time positioning and environmental perception information.
  • the main implementation process includes the following steps:
  • Step 1 Deploy a map construction and real-time positioning framework based on 3D lidar point cloud.
  • the existing relatively mature 3D SLAM method is deployed to realize the construction of 3D scene map and real-time positioning based on the 3D scene map.
  • the result of the implemented scene mapping is shown in Figure 3.
  • Step 2 Scene filtering and 3D obstacle detection based on point cloud map.
  • the obstacle point cloud is obtained mainly by obtaining the original point cloud, downsampling, ground segmentation, coordinate system transformation, and scene background separation preprocessing operations, and then the obstacles outside the original scene are clustered by the Euclidean clustering method.
  • the scene background separation is performed by removing the ground and top of the global scene map obtained in step 1 to generate a three-dimensional grid map, and then the real-time point cloud is filtered according to the three-dimensional grid map to separate the obstacles from the scene.
  • the enclosing box of each obstacle perpendicular to the coordinate system is obtained.
  • step 1 and step 2 refers to Figure 2.
  • the open source sc-lego-loam SLAM framework is used in combination with a 16-line laser radar and a 9-axis IMU to realize three-dimensional mapping of the experimental site, and the mapping result shown in Figure 3 is obtained;
  • the open source hdl-localization positioning framework is used in combination with a 16-line laser radar and a 9-axis IMU to realize the real-time repositioning of the intelligent agent in the experimental site;
  • the real-time conversion matrix between the intelligent agent's own coordinate system and the global map coordinate system can be obtained according to the result of real-time repositioning, and the point cloud information perceived by the intelligent agent in real time can be converted to the global coordinate system according to the conversion matrix, and the scene background separation and obstacle detection described above are completed according to the mapping result of Figure 3, and finally the obstacle bounding boxes
  • Step 3 Build a collaborative perception and dynamic decision-making scheduling system in a centralized communication framework.
  • this invention builds a one-to-many centralized network communication based on the 5G communication module. Based on this communication framework, a multi-agent networked collaborative shared perception and dynamic decision-making scheduling system is built.
  • a connectionless UDP communication is established, in which the central control computer receives the uploaded information of each intelligent agent and broadcasts the path information, and the intelligent agent receives the path information and sends the real-time intelligent agent operation status information, positioning information and obstacle information.
  • the specific communication interaction process between the intelligent agent and the scheduling system refers to the interactive flow chart of each intelligent agent and the scheduling system during operation shown in Figure 5. After the intelligent agent is powered on, it runs the positioning and obstacle detection program and uploads relevant information.
  • the scheduling system After receiving the status information, positioning information and obstacle information, the scheduling system completes the registration for the intelligent agent; when there is a task added to the scheduling system, task allocation and path planning will be carried out and sent to the intelligent agent; after receiving the path, the intelligent agent performs path following or obstacle avoidance according to the environmental information.
  • the abnormal state information is uploaded; after receiving the abnormal state information, the scheduling system replans the path according to the current executable path list and sends it to the intelligent agent again. If there is no path to be planned, the user is prompted to clear the obstacle.
  • an agent obstacle list and a global abnormal path list are introduced.
  • the abnormal path list is combined with the global map and a dynamic replanning mechanism is introduced.
  • each intelligent agent In order to realize the combination of 3D perception of intelligent agents and scheduling planning, after each intelligent agent perceives obstacles in the surrounding environment, it uploads the obstacle information to the scheduling system through communication.
  • the scheduling system maintains an obstacle list for each intelligent agent.
  • an intelligent agent identifies other intelligent agents as obstacles, it eliminates them through position judgment, following the principle that the environment near a single intelligent agent is mainly based on the perception of the intelligent agent.
  • the scheduling system refers to each obstacle list to jointly maintain a global abnormal path list. Through the mechanism of the abnormal path list, the sharing of perception information of each intelligent agent can be realized.
  • the combination of collaborative perception and path planning can be realized by traversing the global topological path and the global abnormal path to jointly maintain a map of executable paths.
  • the dispatching system After receiving the location and status information of the agent, the dispatching system registers information for each agent, as shown in Figure 7, and performs task allocation and path planning based on the location and status information of the agent.
  • a path planning algorithm that reuses space and time is used.
  • a time window mechanism based on path segments is introduced to ensure that when allocating the planned path, only one agent will appear on the same path in the same time period, so as to realize the collaborative operation of multiple agents.
  • the agent can make dynamic autonomous decisions when the path ahead is blocked by obstacles and cannot be avoided.
  • a replanning mechanism is introduced on the basis of the spatiotemporal multiplexing multi-agent collaborative scheduling system.
  • the agent encounters an obstacle blocking the path ahead and cannot avoid it, it stops driving and sets the state to abnormal, and uploads the state information at the same time; as shown in Figure 8, the central control scheduling system receives the abnormal information, triggers replanning, traverses the current tasks of all agents, and compares them with the abnormal path list.
  • the agent whose task contains the abnormal path is replanned and sent to each agent to realize the function of multi-vehicle replanning.
  • this replanning function determines whether to replan the path for the agent by checking the jointly maintained abnormal path list. If other agents detect that obstacles block some paths of this agent outside the line of sight, it will also trigger the replanning of this agent, realizing dynamic perception decision beyond the line of sight.
  • Step 4 Implement a path following algorithm with dynamic speed optimization based on target time.
  • the real-time required speed is calculated, and the speed is dynamically adjusted during the driving process to ensure that the agent drives according to the planned time window and path.
  • Step 5 Implement an obstacle avoidance algorithm that combines local path planning with global path planning.
  • a dynamic collaborative path planning design based on 3D obstacle detection and path planning is used.
  • the scheduling system will re-plan and allocate paths for all tasks containing abnormal paths.
  • This design will cause re-planning to occur all the time in scenes with many obstacles, making it difficult to ensure operational efficiency.
  • an obstacle avoidance function is added.
  • an obstacle avoidance framework combining local path planning with global path planning is designed. The details are as follows:
  • the intelligent body During the operation of the intelligent body, it detects in real time whether there are obstacles blocking the current path. If not, it drives normally.
  • the obstacle avoidance algorithm logic is called to plan a local obstacle avoidance path, and the time cost of the obstacle avoidance path when running at the maximum speed is calculated, and the time cost of the original path when running at the maximum speed is added;
  • the time cost is compared with the time to reach the target point planned by the scheduling system. If the target point is reached within the planned time, the obstacle avoidance algorithm is used. If the target point cannot be reached within the planned time or the obstacle cannot be avoided, the driving is suspended, the state of the intelligent body is adjusted, and re-planning is triggered.
  • the speed selector is used to determine whether to execute path following or obstacle avoidance.
  • path following is performed when there are no obstacles on the path ahead.
  • the intelligent agent executes path following, which can reduce the frequency of re-planning and improve the operation efficiency of the intelligent agent.
  • a point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making device is provided, see Figure 9, which is a principle block diagram of the point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making device of the present invention
  • the point cloud-based multi-agent beyond-visual-range networked collaborative perception dynamic decision-making device includes a three-dimensional scene map construction module, an original point cloud processing module, a scheduling system construction module and an agent execution module;
  • the three-dimensional scene map construction module is used to construct a three-dimensional scene map and perform real-time positioning of the agent;
  • the original point cloud processing module is used to obtain the original point cloud, and then obtain the obstacle point cloud through down sampling, ground segmentation, coordinate system transformation and scene background separation preprocessing operations, and then cluster the obstacles outside the original scene through the Euclidean clustering method;
  • the scheduling system construction module is used to build a cluster-based A scheduling system for collaborative perception and dynamic decision-making in a Chinese communication framework.
  • the scheduling system introduces an agent obstacle list and a global abnormal path list based on real-time agent operation status information, positioning information and obstacle information.
  • a map of execution paths is maintained by traversing global topological paths and global abnormal paths.
  • an agent execution module is used to receive broadcast path information, execute path following with dynamic speed adjustment based on time window optimization, switch to obstacle avoidance state according to real-time obstacle detection information, or upload abnormal information to perform dynamic replanning, and send real-time agent operation status information, positioning information and obstacle information at the same time.
  • a computer device which may be a server, and its internal structure diagram may be as shown in FIG10.
  • the computer device includes a processor, a memory, a network interface, and a database connected via a system bus.
  • the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, a computer program, and a database.
  • the internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium.
  • a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program: Step 1, constructing a three-dimensional scene map to locate an intelligent agent in real time;
  • Step 2 After obtaining the original point cloud, the obstacle point cloud is obtained through downsampling, ground segmentation, coordinate system transformation and scene background separation preprocessing operations, and then the obstacles outside the original scene are clustered by the Euclidean clustering method;
  • Step 3 Build a collaborative perception dynamic decision-making scheduling system based on a centralized communication framework.
  • the scheduling system introduces an agent obstacle list and a global abnormal path list based on real-time agent operation status information, positioning information, and obstacle information.
  • a map of executable paths is maintained by traversing the global topological path and the global abnormal path.
  • global path planning for multi-agent collaboration is achieved, and the path is broadcast to achieve the combination of collaborative perception and path planning.
  • Step 4 After the intelligent agent receives the broadcasted path information, it executes path following with dynamic speed adjustment based on time window optimization, switches to obstacle avoidance state according to real-time obstacle detection information, or uploads abnormal information to perform dynamic replanning, and sends real-time intelligent agent operation status information, positioning information, and obstacle information.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: Step 1, constructing a three-dimensional scene map to locate the intelligent agent in real time;
  • Step 2 After obtaining the original point cloud, the obstacle point cloud is obtained through downsampling, ground segmentation, coordinate system transformation and scene background separation preprocessing operations, and then the obstacles outside the original scene are clustered by the Euclidean clustering method;
  • Step 3 Build a collaborative perception dynamic decision-making scheduling system based on a centralized communication framework.
  • the scheduling system introduces an agent obstacle list and a global abnormal path list based on real-time agent operation status information, positioning information, and obstacle information.
  • a map of executable paths is maintained by traversing the global topological path and the global abnormal path.
  • global path planning for multi-agent collaboration is achieved, and the path is broadcast to achieve the combination of collaborative perception and path planning.
  • Step 4 After the intelligent agent receives the broadcasted path information, it executes path following with dynamic speed adjustment based on time window optimization, switches to obstacle avoidance state according to real-time obstacle detection information, or uploads abnormal information to perform dynamic replanning, and sends real-time intelligent agent operation status information, positioning information, and obstacle information.
  • Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Optics & Photonics (AREA)
  • Electromagnetism (AREA)
  • Processing Or Creating Images (AREA)
  • Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于点云的多智能体超视距网联协同感知动态决策方法及装置，涉及机器人感知以及多智能体协作领域。方法通过维护智能体的障碍物列表及全局异常路径列表，将协同感知与路径规划相结合，能够实现多智能体协同感知及多智能体协同规划调度任务。引入了重规划机制，可使智能体系统根据感知信息进行动态决策，提升了对动态复杂环境的适应能力。而引入时间窗路径规划的路径跟随算法，可根据时间条件动态调整运行速度，为协同路径规划算法提供控制支撑。通过添加改进的避障算法，遇到障碍物时会先判断智能体能否自主避障，若无法避障则执行重规划，降低了重规划的频率，提高智能体运行效率。

Description

多智能体超视距网联协同感知动态决策方法及相关装置 技术领域
本发明涉及机器人感知以及多智能体协作领域，具体涉及多智能体超视距网联协同感知动态决策方法及相关装置。
 背景技术
随着时代的发展进步，自动化工厂、智能仓储物流等应用场景对多智能体感知与协作能力的需求越来越高。传统的多智能体任务基于结构化场景地图，智能体按照固定路线行驶，各智能体完成各自固定任务，无法实时更改运行策略，抗干扰能力弱、效率低、无法应对突发状况，无法适应结构不规则、没有预定义的数据模型且动态变化的复杂非结构化场景。而5G通信技术的发展、智能体自主感知能力的提升，为多智能体在非结构化场景下通过通信网联协同感知，协同动态调度规划提供了可能。
 技术问题
本发明的目的在于克服上述现有技术的缺点，提供多智能体超视距网联协同感知动态决策方法及相关装置。
 技术解决方案
一种基于点云的多智能体超视距网联协同感知动态决策方法，包括以下步骤：
步骤1、构建三维场景地图，对智能体进行实时定位；
步骤2、获取原始点云后，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；
步骤3、搭建基于集中式通信框架的协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个可执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；
步骤4、当智能体端接收到广播的路径信息后，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
进一步的，步骤2中的场景背景分通过获取全局场景地图，进行去地面和去顶的处理，生成三维栅格地图，根据所述三维栅格地图对实时的点云进行过滤，实现障碍物与场景的分离；根据每个聚类的几何坐标信息，获取到每个障碍物垂直于坐标系的包围框。
进一步的，在步骤3中的调度系统进行路径规划前，各智能体感知周围环境障碍物后，将障碍物信息上传给调度系统；所述调度系统为每个智能体分别维护障碍物列表，所述障碍物列表存储有当前智能体实时上传的各障碍物在全局地图下的位置及包围框尺寸信息，根据各障碍物列表中的障碍物的位置及包围框尺寸信息，判断对全局拓扑路径的占用情况，将被占用的路径段添加到异常路径列表中，基于各障碍物列表维护一个实时的全局异常路径列表，所述实时的全局异常路径列表存储有全局拓扑路径中所有被障碍物占用的路径段；
在进行任务路径规划时，调度系统通过遍历全局拓扑路径和全局异常路径来共同维护一个实时的可执行路径的拓扑地图；
当调度系统有任务起点终点输入时，首先将任务分配给距离最近的空闲的智能体，将所述实时的可执行路径的拓扑地图作为地图输入，执行最短路径规划算法并进行时间窗分配，完成路径规划，并将路径下发给智能体。
进一步的，在为每个智能体添加障碍物列表的过程中，当调度系统遍历当前智能体上传的障碍物信息时，同时遍历其它在线的各智能体的定位信息，判断是否有其它智能体的位置在当前遍历的障碍物的包围框内；若有，则将当前遍历的障碍物识别为智能体，不添加到当前智能体障碍物列表中。
进一步的，当步骤3中的调度系统进行路径规划时，所述调度系统接收智能体定位及状态信息后，为每个智能体注册信息，同时根据智能体的位置及状态信息，进行任务分配和路径规划；
在路径规划使用空间与时间复用的路径规划算法，在最短路径规划的基础上引入以路径段为单位的时间窗机制，在最短路径规划时保留多条路径，根据时间窗分配情况，将其中路径最短的一条作为规划路径。
进一步的，步骤4中的智能体进行路径跟随或避障时，当智能体端遇到前方路径有障碍物遮挡且无法避让时，暂停行驶并置状态异常，同时上传状态信息；
当调度系统接收到异常信息时，遍历所有智能体的当前任务，并与异常路径列表作对比，基于对比结果为所有任务包含异常路径的智能体重新规划出可行使路径，同时下发任务路径给各智能体，以实现多车重规划的功能。
进一步的，当障碍物在智能体的视距外时，若有其它智能体检测到障碍物遮挡了所述智能体的路径，则调度系统对所述智能体的路径进行重规划。
进一步的：在智能体运行过程中，实时检测当前路径是否有障碍物遮挡，若无则正常行驶；
若当前路径有障碍物遮挡，则调用避障算法计算局部运行轨迹，规划出局部的避障路径，计算以最大速度运行所述避障路径的时间代价，同时以最大速度运行下原本路径下的时间代价；将时间代价与调度系统规划的到达目标点的时间作对比，若能够在规划的时间内到达，则选择执行避障算法进行避障；若无法在规划的时间内到达或无法避障，则暂停行驶，调整智能体状态，触发重规划。
进一步的，当步骤4中智能体进行路径跟随时，智能体结合下一个目标点的时间点和实时到达下一目标点的距离，计算实时所需速度，在行驶的过程中动态调节所述智能体的速度，使得智能体可以在调度系统规划的时间窗内完成在当前路径上的运行任务，保证智能体按照规划的时间窗及路径行驶。
一种基于点云的多智能体超视距网联协同感知动态决策装置，包括三维场景地图构建模块、原始点云处理模块、调度系统搭建模块和智能体执行模块；
所述三维场景地图构建模块，用于构建三维场景地图，对智能体进行实时定位；
所述原始点云处理模块，用于获取原始点云后，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；
所述调度系统搭建模块，用于搭建基于集中式通信框架协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；
智能体执行模块，用于接收广播的路径信息，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现上述基于点云的多智能体超视距网联协同感知动态决策方法的步骤。
一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被处理器执行时上述基于点云的多智能体超视距网联协同感知动态决策方法的步骤。
 有益效果
本发明的基于点云的多智能体超视距网联协同感知动态决策方法，通过维护智能体的障碍物列表及全局异常路径列表，将协同感知与路径规划相结合，能够实现多智能体协同感知及多智能体协同规划调度任务。引入了重规划机制，可使智能体系统根据感知信息进行动态决策，提升了对动态复杂环境的适应能力。本发明对于结构不规则、没有预定义的数据模型且动态变化的复杂非结构化场景，可保证多智能体动态、协同、高效地完成行驶任务。
进一步的，引入时间窗路径规划的路径跟随算法，该算法可根据时间条件动态调整运行速度，为协同路径规划算法提供控制支撑。
进一步的，通过添加改进的避障算法，遇到障碍物时会先判断智能体能否自主避障，若无法避障则执行重规划，降低了重规划的频率，提高智能体运行效率。
本发明的基于点云的多智能体超视距网联协同感知动态决策装置，包含完成上述动态决策方法的具体模块。
本发明提供的基于点云的多智能体超视距网联协同感知动态决策的计算机设备及存储介质，用于实现上述决策方法的具体步骤。
 附图说明
图1为本发明决策方法的框架流程图；
图2为本发明所用感知及定位框架图；
图3为本发明三维建图可视化结果；
图4为本发明智能体运动控制框架图；
图5为各智能体与调度系统运行时交互流程图；
图6为本发明调度系统路径规划框架图；
图7为本发明调度系统路径重规划框架图；
图8为本发明调度系统可执行路径维护框架图；
图9为本发明的基于点云的多智能体超视距网联协同感知动态决策装置的原理框图；
图10是本发明实施例中计算机设备的示意图。
 本发明的实施方式
为了使本技术领域的人员更好地理解本发明方案，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分的实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本发明保护的范围。
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
下面结合附图对本发明做进一步详细描述：
在一个实施例中，提供了基于点云的多智能体超视距网联协同感知动态决策方法，本发明的具体实施的框架流程图如图1所示，智能体进行三维障碍物检测、建图、定位为调度系统提供实时的环境感知、静态地图及实时定位信息支撑；调度系统实现多车全局路径规划及动态重规划的功能，并将规划结果下发给智能体；智能体获取规划结果后根据实时的定位和环境感知信息进行路径跟随和避障。主要实现过程，包括以下步骤：
步骤1：部署基于三维激光雷达点云的地图构建与实时定位框架。
基于智能体搭载的三维激光雷达以及惯性测量单元，部署现有的比较成熟的三维SLAM方法实现三维场景地图构建以及基于三维场景地图的实时定位。实施的场景建图结果如图3所示。
步骤2：基于点云地图的场景过滤及3D障碍物检测。
在本发明中，受设备算力约束，使用传统的聚类算法来实现三维障碍物的检测。主要通过获取原始点云、下采样、地面分割、坐标系变换、场景背景分离预处理操作得到障碍物点云，而后通过欧式聚类方法聚类出脱离原始场景之外的障碍物。其中的场景背景分离，通过步骤1获取的全局场景地图，进行去地面和去顶的预处理，生成三维栅格地图，然后根据三维栅格地图对实时的点云进行过滤，实现障碍物与场景的分离。并根据每个聚类的几何坐标信息，获取到每个障碍物垂直于坐标系的包围框。
步骤1和步骤2的实施框架参考图2，如图2中最右侧描述，使用开源的sc-lego-loam的SLAM框架，结合16线激光雷达和9轴IMU实现对实验场地的三维建图，得到如图3所示的建图结果；图2框图中间部分，根据建好的地图，使用开源的hdl-localization定位框架结合16线激光雷达和9轴IMU实现智能体在实验场地的实时重定位；图2框图左侧部分，根据实时重定位的结果可得到智能体自身坐标系和全局地图坐标系之间的实时转换矩阵，可根据该转换矩阵将智能体实时感知的点云信息转换到全局坐标系下，并根据图3建图结果完成上述描述的场景背景分离、障碍物检测，最终得到全局坐标系下的各个障碍物包围框。
步骤3：搭建集中式通信框架的协同感知动态决策的调度系统。
传统的4G网络难以满足数据量较大的实时通信需求，在本发明中，为满足大规模室外场景的通信需求，保证数据的稳定传输，减少丢包率，基于5G通信模块搭建了一对多集中式网联通信，基于该通信框架，搭建了一个可实现多智能体网联协同共享感知与动态决策调度系统。
关于通信框架，基于5G通信模块，建立无连接的UDP通信，其中中控计算机端接收各智能体的上传信息并广播路径信息，智能体端接收路径信息并发送实时的智能体运行状态信息、定位信息以及障碍物信息。具体的智能体与调度系统的通信交互流程参考如图5所示各智能体与调度系统运行时交互流程图，智能体上电后运行定位及障碍物检测程序，并上传相关信息，调度系统接收到状态信息、定位信息以及障碍物信息后为智能体完成注册；当调度系统中有任务添加时，会进行任务分配和路径规划，并下发给智能体；智能体接收到路径后根据环境信息执行路径跟随或避障，当路径被遮挡且无法避障时，上传状态异常信息；调度系统接收到状态异常信息后，根据当前的可执行路径列表重新规划路径，并再次下发给智能体，若无路径可规划，则提示用户清除障碍物。
关于调度系统，在本发明中，为将智能体感知与调度系统的任务分配、路径规划相结合，引入了智能体障碍物列表以及全局异常路径列表，与此同时，为适应非结构化场景、实现超视距协同感知与动态决策，将异常路径列表与全局地图结合的同时，引入了动态重规划机制，具体过程是：
1、可执行路径地图维护
为实现智能体3D感知与调度规划的结合，各智能体感知周围环境障碍物后，通过通信将障碍物信息上传给调度系统。调度系统为每个智能体分别维护障碍物列表，对于智能体将其它智能体识别为障碍物的情况，通过位置判断来消除，遵循单个智能体附近的环境以该智能体的感知情况为主的原则。如图6，调度系统参考各障碍物列表共同维护一个全局的异常路径列表，通过该异常路径列表的机制，可以实现各个智能体感知信息的共享。在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来共同维护一个可执行路径的地图，即可实现协同感知与路径规划的结合。
2、时空复用的多智能体网联协同路径规划调度
调度系统接收智能体定位及状态信息后，如图7，为每个智能体注册信息，同时根据智能体的位置及状态信息，进行任务分配、路径规划。为实现多智能体协同运行，使用空间与时间复用的路径规划算法，在最短路径规划的基础上引入以路径段为单位的时间窗机制，保证在分配规划路径时，同一时间段的同一路径只会出现一台智能体，以实现多智能体协同运作。
3、多智能体网联协同动态重规划
为适应复杂动态的非结构化场景，使智能体在遇到前方路径有障碍物遮挡且无法避让时能够进行动态的自主决策，在时空复用多智能体协同的调度系统的基础上引入重规划机制。当智能体端遇到前方路径有障碍物遮挡且无法避让时，暂停行驶并置状态异常，同时上传状态信息；如图8，中控调度系统接收到异常信息，触发重规划，遍历所有智能体的当前任务，并与异常路径列表作对比，为任务包含异常路径的智能体重新规划可行使路径，并下发路径给各智能体，以实现多车重规划的功能。此重规划功能对于距离较远的智能体，通过检查共同维护的异常路径列表来判断是否为智能体重新规划路径，若在视距外，有其它智能体检测到障碍物遮挡了本智能体的某些路径，则亦会触发本智能体的重规划，实现超视距的动态感知决策。
步骤4：实现基于目标时间的动态速度优化的路径跟随算法。
为实现时空复用的路径规划调度，需要保证智能体按照规划的时间窗运行，传统的路径跟随算法无法保证智能体在规划的固定的时间窗内运行，这会造成实际的运行与调度规划的结果有较大的误差，最终会导致各智能体运行的冲突，无法实现协同运行。在本发明中，为解决这种误差带来的冲突，在传统的路径跟随算法中引入基于目标时间的速度动态调节机制。根据调度系统下发的路径目标点，结合下一个目标点的时间点，和实时到达下一目标点的距离，计算实时所需速度，在行驶的过程中动态调节速度，保证智能体按照规划的时间窗及路径行驶。
步骤5：实现局部路径规划与全局路径规划结合的避障算法。
在本发明中，基于3D障碍物检测与路径规划相结合的动态协同的路径规划设计，当智能体前方路径遇到障碍物时就会触发重规划，调度系统为所有包含异常路径的任务重新规划分配路径，这种设计在障碍物较多的场景会导致一直出现重规划的现象，很难保证运行效率。为提高多智能体在障碍物较多的复杂场景下的运行效率，增加了避障的功能，为满足基于时间窗的路径规划设计，设计了一种局部路径规划与全局路径规划相结合的避障框架。具体如下：
智能体运行过程中，实时检测当前路径是否有障碍物遮挡，若无则正常行驶；
若当前路径有障碍物遮挡，则调用避障算法逻辑，规划出局部的避障路径，并计算以最大速度运行时该避障路径的时间代价，并加上最大速度运行下原本路径下的时间代价；
将时间代价与调度系统规划的到达目标点的时间作对比，若在规划的时间内到达，则采用避障算法，若无法在规划的时间内到达或无法避障，则暂停行驶，调整智能体状态，触发重规划。
通过在路径跟随的基础上添加以上描述的改进的避障算法，如图4所示，通过速度选择器来判断执行路径跟随还是执行避障，通过扫描实时的三维环境点云信息，根据障碍物检测的结果，前方路径无障碍物时执行路径跟随，遇到障碍物时会先判断智能体能否自主避障，若可以避障，则执行避障，若无法避障则调度系统执行重规划，智能体执行路径跟随，可以降低重规划的频率，提高智能体运行效率。
在一个实施例中，提供了基于点云的多智能体超视距网联协同感知动态决策装置，参见图9，图9为本发明的基于点云的多智能体超视距网联协同感知动态决策装置的原理框图；基于点云的多智能体超视距网联协同感知动态决策装置包括三维场景地图构建模块、原始点云处理模块、调度系统搭建模块和智能体执行模块；所述三维场景地图构建模块，用于构建三维场景地图，对智能体进行实时定位；所述原始点云处理模块，用于获取原始点云后，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；所述调度系统搭建模块，用于搭建基于集中式通信框架协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；智能体执行模块，用于接收广播的路径信息，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图10所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机程序被处理器执行时以实现基于点云的多智能体超视距网联协同感知动态决策方法。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行计算机程序时实现以下步骤：步骤1、构建三维场景地图，对智能体进行实时定位；
步骤2、获取原始点云后，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；
步骤3、搭建基于集中式通信框架的协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个可执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；
步骤4、当智能体端接收到广播的路径信息后，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：步骤1、构建三维场景地图，对智能体进行实时定位；
步骤2、获取原始点云后，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；
步骤3、搭建基于集中式通信框架的协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个可执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；
步骤4、当智能体端接收到广播的路径信息后，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本发明所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上内容仅为说明本发明的技术思想，不能以此限定本发明的保护范围，凡是按照本发明提出的技术思想，在技术方案基础上所做的任何改动，均落入本发明权利要求书的保护范围之内。

Claims (10)

  1. 一种基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，包括以下步骤：
    步骤1、构建三维场景地图，对智能体进行实时定位；
    步骤2、获取原始点云，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；
    步骤3、搭建基于集中式通信框架的协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；
    步骤4、当智能体端接收到广播的路径信息后，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
  2. 根据权利要求1所述的基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，步骤2中的场景背景分通过获取全局场景地图，进行去地面和去顶的处理，生成三维栅格地图，根据所述三维栅格地图对实时的点云进行过滤，实现障碍物与场景的分离；根据每个聚类的几何坐标信息，获取到每个障碍物垂直于坐标系的包围框。
  3. 根据权利要求1所述的基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，在步骤3中的调度系统进行路径规划前，各智能体感知周围环境障碍物后，将障碍物信息上传给调度系统；所述调度系统为每个智能体分别维护障碍物列表，所述障碍物列表存储有当前智能体实时上传的各障碍物在全局地图下的位置及包围框尺寸信息，根据各障碍物列表中的障碍物的位置及包围框尺寸信息，判断对全局拓扑路径的占用情况，将被占用的路径段添加到异常路径列表中，基于各障碍物列表维护一个实时的全局异常路径列表，所述实时的全局异常路径列表存储有全局拓扑路径中所有被障碍物占用的路径段；
    在进行任务路径规划时，调度系统通过遍历全局拓扑路径和全局异常路径来共同维护一个实时的执行路径的拓扑地图；
    当调度系统有任务起点终点输入时，首先将任务分配给距离最近的空闲的智能体，将所述实时的执行路径的拓扑地图作为地图输入，执行最短路径规划算法并进行时间窗分配，完成路径规划，并将路径下发给智能体。
  4. 根据权利要求3所述的基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，在为每个智能体添加障碍物列表的过程中，当调度系统遍历当前智能体上传的障碍物信息时，同时遍历其它在线的各智能体的定位信息，判断是否有其它智能体的位置在当前遍历的障碍物的包围框内；若有，则将当前遍历的障碍物识别为智能体，不添加到当前智能体障碍物列表中。
  5. 根据权利要求1所述的基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，当步骤3中的调度系统进行路径规划时，所述调度系统接收智能体定位及状态信息后，为每个智能体注册信息，同时根据智能体的位置及状态信息，进行任务分配和路径规划；
    在路径规划使用空间与时间复用的路径规划算法，在最短路径规划的基础上引入以路径段为单位的时间窗机制，在最短路径规划时保留多条路径，根据时间窗分配情况，将其中路径最短的一条作为规划路径。
  6. 根据权利要求1所述的基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，步骤4中的智能体进行路径跟随或避障时，当智能体端遇到前方路径有障碍物遮挡且无法避让时，暂停行驶并置状态异常，同时上传状态信息；
    当调度系统接收到异常信息时，遍历所有智能体的当前任务，并与异常路径列表作对比，基于对比结果为所有任务包含异常路径的智能体重新规划出可行使路径，同时下发任务路径给各智能体，以实现多车重规划的功能；
    当障碍物在智能体的视距外时，若有其它智能体检测到障碍物遮挡了所述智能体的路径，则调度系统对所述智能体的路径进行重规划；
    在智能体运行过程中，实时检测当前路径是否有障碍物遮挡，若无则正常行驶；
    若当前路径有障碍物遮挡，则调用避障算法计算局部运行轨迹，规划出局部的避障路径，计算以最大速度运行所述避障路径的时间代价，同时以最大速度运行下原本路径下的时间代价；将时间代价与调度系统规划的到达目标点的时间作对比，若能够在规划的时间内到达，则选择执行避障算法进行避障；若无法在规划的时间内到达或无法避障，则暂停行驶，调整智能体状态，触发重规划。
  7. 根据权利要求1所述的基于点云的多智能体超视距网联协同感知动态决策方法，其特征在于，当步骤4中智能体进行路径跟随时，智能体结合下一个目标点的时间点和实时到达下一目标点的距离，计算实时所需速度，在行驶的过程中动态调节所述智能体的速度，使得智能体在调度系统规划的时间窗内完成在当前路径上的运行任务，保证智能体按照规划的时间窗及路径行驶。
  8. 一种基于点云的多智能体超视距网联协同感知动态决策装置，其特征在于，包括三维场景地图构建模块、原始点云处理模块、调度系统搭建模块和智能体执行模块；
    所述三维场景地图构建模块，用于构建三维场景地图，对智能体进行实时定位；
    所述原始点云处理模块，用于获取原始点云后，通过下采样、地面分割、坐标系变换和场景背景分离预处理操作得到障碍物点云，之后通过欧式聚类方法聚类出脱离原始场景之外的障碍物；
    所述调度系统搭建模块，用于搭建基于集中式通信框架协同感知动态决策的调度系统，所述调度系统基于实时的智能体运行状态信息、定位信息以及障碍物信息，引入智能体障碍物列表以及全局异常路径列表，在进行任务路径规划时，通过遍历全局拓扑路径和全局异常路径来维护一个执行路径的地图，结合最短路径算法以及时间窗分配，实现多智能体协同的全局路径规划，并进行广播路径，实现协同感知与路径规划的结合；
    智能体执行模块，用于接收广播的路径信息，执行基于时间窗优化的动态调节速度的路径跟随，并根据实时的障碍物检测信息切换到避障状态或者上传异常信息执行动态重规划，同时发送实时的智能体运行状态信息、定位信息以及障碍物信息。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1-7任一项所述的基于点云的多智能体超视距网联协同感知动态决策方法的步骤。
  10. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1-7任一项所述的基于点云的多智能体超视距网联协同感知动态决策方法的步骤。
PCT/CN2022/131637 2022-04-28 2022-11-14 多智能体超视距网联协同感知动态决策方法及相关装置 WO2024098438A1 (zh)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
CN202210462524.7A CN114815832A (zh) 2022-04-28 2022-04-28 一种基于点云的多智能体超视距网联协同感知动态决策方法
CN202211388685.2 2022-11-08
CN202211388685.2A CN116048062A (zh) 2022-04-28 2022-11-08 多智能体超视距网联协同感知动态决策方法及相关装置

Publications (1)

Publication Number Publication Date
WO2024098438A1 true WO2024098438A1 (zh) 2024-05-16

Family

ID=82509944

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/131637 WO2024098438A1 (zh) 2022-04-28 2022-11-14 多智能体超视距网联协同感知动态决策方法及相关装置

Country Status (2)

Country Link
CN (2) CN114815832A (zh)
WO (1) WO2024098438A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114815832A (zh) * 2022-04-28 2022-07-29 西安交通大学 一种基于点云的多智能体超视距网联协同感知动态决策方法
CN115220461B (zh) * 2022-09-21 2023-02-17 睿羿科技(山东)有限公司 室内复杂环境下机器人单体系统及多机器人交互协同方法
CN116700298B (zh) * 2023-08-08 2023-11-21 浙江菜鸟供应链管理有限公司 路径规划方法、系统、设备及存储介质
CN117075640B (zh) * 2023-10-12 2024-01-12 江苏智慧汽车研究院有限公司 基于物联网的人工智能设备无人化运行监管系统及方法
CN117492446A (zh) * 2023-12-25 2024-02-02 北京大学 一种基于组合混合优化的多智能体合作规划方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017168423A1 (en) * 2016-03-30 2017-10-05 Aerialguard Ltd System and method for autonomous guidance of vehicles
CN110955242A (zh) * 2019-11-22 2020-04-03 深圳市优必选科技股份有限公司 机器人导航方法、系统、机器人及存储介质
CN114625140A (zh) * 2022-03-14 2022-06-14 北京大学 基于改进时间窗的多智能体网联协同调度规划方法及系统
CN114815832A (zh) * 2022-04-28 2022-07-29 西安交通大学 一种基于点云的多智能体超视距网联协同感知动态决策方法
CN115016506A (zh) * 2022-07-18 2022-09-06 北京大学 基于自主绕障的异构型多智能体网联协同调度规划方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017168423A1 (en) * 2016-03-30 2017-10-05 Aerialguard Ltd System and method for autonomous guidance of vehicles
CN110955242A (zh) * 2019-11-22 2020-04-03 深圳市优必选科技股份有限公司 机器人导航方法、系统、机器人及存储介质
CN114625140A (zh) * 2022-03-14 2022-06-14 北京大学 基于改进时间窗的多智能体网联协同调度规划方法及系统
CN114815832A (zh) * 2022-04-28 2022-07-29 西安交通大学 一种基于点云的多智能体超视距网联协同感知动态决策方法
CN115016506A (zh) * 2022-07-18 2022-09-06 北京大学 基于自主绕障的异构型多智能体网联协同调度规划方法

Also Published As

Publication number Publication date
CN114815832A (zh) 2022-07-29
CN116048062A (zh) 2023-05-02

Similar Documents

Publication Publication Date Title
WO2024098438A1 (zh) 多智能体超视距网联协同感知动态决策方法及相关装置
Cardarelli et al. Cooperative cloud robotics architecture for the coordination of multi-AGV systems in industrial warehouses
CN111426326B (zh) 一种导航方法、装置、设备、系统及存储介质
RU2589869C2 (ru) Способ и система для эффективного планирования для множества автоматизированных неголономных транспортных средств с использованием планировщика скоординированных маршрутов
CN108267149B (zh) 多移动机器人的冲突管理方法及系统
CN112698647A (zh) 跨楼层路径规划方法、装置、计算机设备及存储介质
EP2745431A1 (en) Adaptive communications network
CN110210806B (zh) 一种5g边缘计算的云基无人车架构及其控制评价方法
Gregoire et al. Priority-based intersection management with kinodynamic constraints
CN111142567B (zh) 一种无人机系统中的无人机目标位置交换方法及装置
CN114610066A (zh) 复杂未知环境下分布式集群无人机编队飞行轨迹生成方法
Montanaro et al. Cloud-assisted distributed control system architecture for platooning
Pal et al. Multi-robot exploration in wireless environments
Cao et al. Research on global optimization method for multiple AGV collision avoidance in hybrid path
CN109709985A (zh) 一种无人机任务优化方法、装置及系统
Hirata et al. Roadside-assisted cooperative planning using future path sharing for autonomous driving
Schmidt et al. Research on decentralized control strategies for automated vehicle-based in-house Transport systems–a survey
Delgado et al. OROS: Orchestrating ROS-driven collaborative connected robots in mission-critical operations
Huang et al. Multi-agent vehicle formation control based on mpc and particle swarm optimization algorithm
Indri et al. Supervised global path planning for mobile robots with obstacle avoidance
CN115220461B (zh) 室内复杂环境下机器人单体系统及多机器人交互协同方法
CN112700025A (zh) 覆盖区域的分配方法及装置、存储介质、电子设备
CN115657676A (zh) 基于优先级的集中式多agv多径通道变道决策规划方法
CN115525032A (zh) 一种基于多种类机器人智能调度处理控制系统
Rebollo et al. Collision avoidance among multiple aerial robots and other non-cooperative aircraft based on velocity planning