多机器人存取系统研究现状｜热文回顾
Research status of multi robot access system ｜ review of hot papers
文｜北京科技大学机械工程学院
Wen | School of Mechanical Engineering, Beijing University of Science and Technology
孙阳君、赵宁
Sun Yangjun, Zhao Ning
在电子商务繁荣发展的今天，利用多机器人存取系统可以节约人力成本、提高拣选效率、提高存储密度、降低碳排放，因此许多电商企业选择布局多机器人存取系统，通过“机器换人”，实现“货到人”拣眩随着多机器人存取系统在企业内的应用发展，多机器人存取系统也成为近年来的热点研究问题。本文重点介绍了多机器人存取系统中的热点研究方向，总结了现有的解决方案，最后对未来研究进行了展望。
With the prosperity and development of e-commerce, Using multi-robot access system can save labor cost, improve picking efficiency, improve storage density, reduce carbon emissions, Therefore, many e-commerce enterprises choose to lay out multi-robot access systems and realize "goods on arrival" through "machine substitution". With the application and development of multi-robot access systems in enterprises, multi-robot access systems have also become a hot research issue in recent years. This paper focuses on the hot research directions in multi-robot access systems, summarizes the existing solutions, and finally looks forward to the future research.
最早的移动机器人在1953年被美国basrrte公司开发成功，它能够沿着布置在空中的导线完成任务，但不能自主随意移动。自20世纪初，自寻址技术的出现对传统的移动机器人实现了技术革命。自寻址技术可以不预先铺设轨道，以非接触的方式实现车辆的智能寻址和定位，大幅增加了移动机器人的智能程度和应用范围。随着自寻址技术的发展，移动机器人可广泛用于车间、码头、仓库等系统中，系统内的机器人数量也从几台扩展到上百台。
The earliest mobile robot was successfully developed by Basrrte Company of the United States in 1953. It can complete tasks along wires arranged in the air, but cannot move freely. Since the beginning of the 20th century, the emergence of self-addressing technology has brought about a technological revolution to traditional mobile robots. Self-addressing technology can realize intelligent addressing and positioning of vehicles in a non-contact way without laying tracks in advance, thus greatly increasing the intelligence degree and application range of mobile robots. With the development of self-addressing technology, mobile robots can be widely used in workshops, docks, warehouses and other systems, and the number of robots in the system has also expanded from several to hundreds.
2008 年，Kiva systems公司首次将上百辆互相协作的移动机器人同时用于亚马逊的仓储作业中[1]，即最早的多机器人存取系统。2012年，亚马逊收购Kiva公司将多机器人存取系统部署在北美的各大配送中心内。国内也紧随其后，菜鸟、京东、极智嘉和快仓等公司已经成功应用多机器人存取系统完成相关业务。
In 2008, Kiva Systems used hundreds of cooperative mobile robots for Amazon's storage operations at the same time for the first time [1], the earliest multi-robot access system. In 2012, Amazon acquired Kiva and deployed multi-robot access systems in major distribution centers in North America. China is also closely followed. Cainiao, Jingdong, Jizhijia and Fast Warehouse have successfully applied multi-robot access systems to complete related businesses.
近年来，多机器人存取系统的销量也在不断增长。《2019年世界机器人报告》[2]显示，仅2018年就卖出了超过十万的多机器人系统，其中大多数都被用于电子商务仓库中，为电商公司提供存取服务。2020年，COVID-19疫情进一步推动了多机器人存取系统的市场，《2020年世界机器人报告》[3]显示，电商仓库内的移动机器人交易额增长了110％，达到19亿美元。按预测，未来的交易额增长可能达到每年40％甚至更高。
In recent years, the sales volume of multi-robot access systems has also been increasing. The 2019 World Robot Report [2] shows that more than 100,000 multi-robot systems were sold in 2018 alone, most of which were used in e-commerce warehouses to provide access services for e-commerce companies. In 2020, the COVID-19 epidemic further promoted the market for multi-robot access systems. The World Robot Report 2020 [3] showed that the turnover of mobile robots in e-commerce warehouses increased by 110% to reach 1.9 billion US dollars. According to the forecast, the future turnover growth may reach 40% or more per year.
种种迹象都表明，多机器人存取系统已经成为应用热点，对多机器人存取系统的研究也日益增多。本文首先介绍了多机器人存取系统，描述了多个热点研究方向和解决方案，最后对现有研究进行了总结和展望。
All kinds of signs show that multi-robot access system has become a hot application topic, and the research on multi-robot access system is also increasing day by day. This paper first introduces the multi-robot access system, describes many hot research directions and solutions, and finally summarizes and looks forward to the existing research.
一、系统简介
I. Introduction to the System
多机器人存取系统利用货架存储商品，通过移动机器人搬运货架，工作人员只需要在工作站等待，不用进入存储区域，工作完毕后移动机器人再将货架搬运回存储区域。这样一来，就大大减少了工作人员的劳动强度，减少人员行走距离。如图1所示，移动机器人依靠扫描地面的二维码定位，根据指令向任意方向行驶。按照任务要求，执行搬运、升降货架、等待等操作。
The multi-robot access system uses shelves to store goods, and the mobile robot transports the shelves. The staff only need to wait at the workstation without entering the storage area. After the work is completed, the mobile robot transports the shelves back to the storage area. In this way, the labor intensity of the staff is greatly reduced and the walking distance of the staff is reduced. As shown in FIG. 1, the mobile robot relies on scanning the two-dimensional code on the ground for positioning and travels in any direction according to instructions. According to the task requirements, carry out handling, lifting shelves, waiting and other operations.
图1 多机器人存取系统
Fig. 1 Multi-robot access system
图2 多机器人存取系统工作流程
Fig. 2 Workflow of Multi-Robot Access System
在电子商务的订单拣选中，系统的目的就是尽可能提升拣货效率，更快更好地完成客户的订单。如图2所示，系统的工作流程可以概况为：根据客户下达的订单，指派工作站完成订单，按照订单上的商品确定机器人需要执行的任务，通过机器人搬运货架至工作站完成拣选，再搬运货架回储区。在系统运行中，如果订单需求的货物量超过了安全库存，还需要进行补货操作。如果机器人的电量不足，需要充电，确定充电策略。除了工作流程中的策略和优化方法外，还需要确定货架和商品存放在哪些位置，并进行储位优化。此外，系统的布局模式、机器人的数量、拣货站的数量、货架的位置等都会对系统效率造成影响。因此，还需要对系统整体布局中的工作站位置、货架位置、充电站位置等进行优化。
In the order picking of e-commerce, the purpose of the system is to improve the picking efficiency as much as possible and complete the customer's orders faster and better. As shown in FIG. 2, the workflow of the system can be summarized as follows: according to the order issued by the customer, the workstation is assigned to complete the order, the tasks to be performed by the robot are determined according to the goods on the order, the shelves are transported to the workstation through the robot to complete the picking, and then the shelves are transported back to the storage area. During the operation of the system, if the quantity of goods required by the order exceeds the safety stock, replenishment operation is also required. If the robot is short of electricity and needs to be charged, determine the charging strategy. In addition to the strategies and optimization methods in the workflow, it is also necessary to determine where the shelves and commodities are stored and optimize the storage location. In addition, the layout mode of the system, the number of robots, the number of picking stations and the location of shelves will all affect the efficiency of the system. Therefore, it is also necessary to optimize the workstation location, shelf location and charging station location in the overall layout of the system.
二、研究方向
II. Research Direction
按照图2系统的工作流程，可以对系统本身和系统内的各个环节进行研究，主要包括以下研究方向：布局优化、订单指派、任务分配、路径规划、冲突消解、储位优化、充电策略、实时调度等。这些研究方向涉及到系统运行前的整体布局和系统运行中的各个流程。
According to the workflow of the system in FIG. 2, the system itself and all links in the system can be studied, mainly including the following research directions: layout optimization, order assignment, task allocation, path planning, conflict resolution, storage optimization, charging strategy, real-time scheduling, etc. These research directions involve the overall layout of the system before operation and various processes in the system operation.
1.布局优化问题
1. Layout optimization
图3 多机器人存取系统俯视图
Fig. 3 Top view of multi-robot access system
一个普通的多机器人存取系统的俯视图见图3。布局优化就是对系统内的机器人、货架、工作站等的位置、数量、比例等进行设计，确定最合适的布局，以提高拣货效率，增加系统吞吐量，更快完成订单。
A top view of an ordinary multi-robot access system is shown in Fig. 3. Layout optimization is to design the location, quantity and proportion of robots, shelves, workstations, etc. in the system to determine the most suitable layout, so as to improve picking efficiency, increase system throughput and complete orders faster.
对于系统的布局研究，包括以下几种情况：
The layout research of the system includes the following situations:
（1）工作站所处的位置和数量。图3中工作站是在货架的一侧，但在布局中，可以两侧都摆放货架，也可以四周都有工作站[4]。不同数量的工作站会对拣选效率有影响。
(1) Location and number of workstations. The workstations in FIG. 3 are on one side of the shelf, but in the layout, the shelves can be placed on both sides or there can be workstations around [4]. Different number of workstations will affect the picking efficiency.
（2）货架布局方式。a.货架长宽比，货架区域的设置影响车辆运行的路径，可以通过建立排队网模型计算不同长宽比对拣货效率的影响[5]。b.布局模式。图3展示的是传统的布局模式，还有很多新型布局模式可以考虑，特殊的布局模式可以增加拣货的效率或系统的存储密度。例如图4展示的V型布局[4]、鱼骨型布局[6]、多深布局[7]等。
(2) Shelf layout. A. The length-width ratio of shelves and the setting of shelf areas affect the route of vehicles. The influence of different length-width ratios on picking efficiency can be calculated by establishing a queuing network model [5]. B. Layout pattern. FIG. 3 shows the traditional layout mode, and there are many new layout modes to consider. Special layout modes can increase the picking efficiency or the storage density of the system. For example, the V-shaped layout [4], fishbone-shaped layout [6], multi-depth layout [7] and the like shown in FIG. 4.
图4 几种新型布局模式
Figure 4 Several New Layout Patterns
（3）仓库内道路布局。道路布局有很多需要考虑的地方。如a)道路方向。图3展示的道路都有具体的方向，即单向道。通过单向道避免发生相向冲突和死锁。除单向道外，道路也可以采用双向道布局方式，机器人可以沿任意方向运行，能够减少机器人的绕路，加快运行效率。考虑到双向道内的相向冲突较为严重，也可以将两者结合布局，部分区域采用单向道，部分区域采用双向道[8]。b）车道数量，图3中，只有在工作站和存储区域的道路是双行道，存储区域都是单行道，即只能容纳一辆车通过。双行道可以更大程度上避免机器人的冲突，但浪费了存储面积。c）是否有交叉口。交叉口同样降低了存储密度，但机器人移动时可选择的路径比无交叉口时更多，能够避免机器人间的拥堵。单向单行道和双向单行道、有交叉口和无交叉口的对比可以参考Lienert等[9]的研究。
(3) Road layout in the warehouse. There are many things to consider in road layout. Such as a) the direction of the road. The roads shown in FIG. 3 all have specific directions, i.e. One-way roads. Avoid opposite collisions and deadlocks through one-way tracks. In addition to one-way roads, roads can also adopt two-way road layout. Robots can run in any direction, which can reduce detours of robots and speed up operation efficiency. Considering the serious opposite conflicts in the two-way channels, the two can also be combined for layout, with one-way channels in some areas and two-way channels in some areas [8]. B) Number of lanes. In FIG. 3, only the roads in the workstation and storage area are two-lane, and the storage area is one-way, i.e. Only one car can pass through. Two-lane can avoid robot conflicts to a greater extent, but wastes storage area. C) Whether there are intersections. Intersections also reduce the storage density, but robots can choose more paths when moving than when there are no intersections, which can avoid congestion between robots. The comparison between one-way one-way streets and two-way one-way streets, with intersections and without intersections can refer to the research of Lienert et al. [9].
2.订单指派问题
2. Order Assignment Problem
订单是多机器人存取系统的输入，完成订单是系统运行的首要目的。单张订单每件商品按顺序拣选的完成方式会大大降低系统效率，不同订单有同一件商品时，完全可以一同拣选，即使没有一样的商品，两张订单需要拣选的商品也可能位于同一个货架上。因此，就需要对订单进行整合和指派，确立订单内商品在哪些货架上，哪些订单可以在同一个工作站内拣选[10-11]。优化订单的顺序还能减少机器人的数量，降低系统运行成本[12]。为了提高系统效率，订单内的货物甚至可以被拆分到不同的工作站完成[13]。
Order is the input of multi-robot access system, and completing order is the primary purpose of system operation. The completion method of sequential picking of each item in a single order will greatly reduce the efficiency of the system. When different orders have the same item, they can be picked together. Even if there is no same item, the items to be picked in the two orders may be located on the same shelf. Therefore, it is necessary to integrate and assign orders to establish which shelves the goods in the orders are on and which orders can be picked in the same workstation [10-11]. Optimizing the order sequence can also reduce the number of robots and the operating cost of the system [12]. In order to improve the efficiency of the system, the goods in the order can even be split into different workstations to complete [13].
3.任务分配问题
3. Assignment of tasks
任务分配是在订单指派后，将订单拆分成涉及到具体货架的任务，将任务分配给机器人进行执行。系统内同时有多个任务和多辆机器人，一辆机器人在同一时间只能搬运一个货架，任务分配情况关系到机器人未来的行动路线。在任务分配时，需要综合考虑机器人和任务要求的货架之间的距离，机器人本身的任务情况，任务间的关系等多个因素。现有的研究中，任务分配可以是采用一些策略进行分配，如基于作业速率、近似最优和最优指派策略等被用于任务分配[14]。还可以根据当前系统内任务的距离、时间、效率等状态，设计启发式规则完成任务分配[10]。此外，可以利用智能算法通过迭代优化的方式获得更好的任务分配结果，如遗传算法[15]，模拟退火算法[16]，禁忌搜索[17]等。
Task assignment is to split the order into tasks involving specific shelves after the order is assigned, and assign the tasks to robots for execution. There are multiple tasks and multiple robots in the system at the same time. A robot can only carry one shelf at the same time. The task allocation is related to the future action route of the robot. When assigning tasks, many factors such as the distance between the robot and the shelf required by the task, the task situation of the robot itself, and the relationship between tasks need to be comprehensively considered. In the existing research, task allocation can be carried out by adopting some strategies, such as based on job rate, approximate optimal and optimal assignment strategies, etc. are used for task allocation [14]. Heuristic rules can also be designed to complete task allocation according to the distance, time, efficiency and other states of tasks in the current system [10]. In addition, intelligent algorithms can be used to obtain better task allocation results through iterative optimization, such as genetic algorithm [15], simulated annealing algorithm [16], tabu search [17], etc.
4.路径规划问题
4. Path planning
图5 机器人的不同路径
Figure 5 Different Paths of Robot
机器人接取任务之后，就需要执行任务，按照任务要求对运行路径进行规划。机器人的运行路径规划是多机器人存取系统的核心。好的路径规划方案能够大幅度提升系统效率，减少机器人的能耗。如图5(a)所示，机器人完成任务需要三段路径，首先要从当前位置到货架所在位置，再从货架所在位置到工作站，之后带着货架返回储区。这期间，机器人由无数条路径可以完成任务，例如图5(a)是最短路径，但也可选择图5(b)的其他路径，但最终完成任务的路径受到多方面影响。由于仓库内不只一辆机器人，多辆机器人之间会造成冲突问题。最短路径可能有其他机器人频繁经过，产生严重的冲突问题。因此机器人运行的最短路径并非就是最好的，还需要综合考虑其他机器人的情况。常用的路径生成方法包括动作依赖图[18]、路线图生成算法[19]、A*算法[20]、Dijkstra's 算法[17]等以及对这些算法的改进。
After the robot takes the task, it needs to execute the task and plan the running path according to the task requirements. Robot path planning is the core of multi-robot access system. A good path planning scheme can greatly improve the system efficiency and reduce the energy consumption of robots. As shown in fig. 5 (a), the robot needs three paths to complete the task, first from the current position to the position of the shelf, then from the position of the shelf to the workstation, and then return to the storage area with the shelf. During this period, the robot can complete the task by numerous paths, for example, fig. 5 (a) is the shortest path, but other paths of fig. 5 (b) can also be selected, but the path that finally completes the task is affected by many factors. Since there is more than one robot in the warehouse, conflicts will occur between multiple robots. The shortest path may be frequently passed by other robots, resulting in serious conflicts. Therefore, the shortest path for robots to run is not the best, and the situation of other robots needs to be comprehensively considered. Commonly used path generation methods include action dependence graph [18], roadmap generation algorithm [19], A* algorithm [20], Dijkstra's algorithm [17] and other improvements to these algorithms.
5.冲突消解问题
5. Conflict resolution
图6 机器人间的冲突情况
Figure 6 Conflicts between Robots
多辆机器人运行时，由于运行路线有交错，必然会发生冲突。机器人间的冲突可以被归为以下几类：赶超冲突、交叉口冲突、相向冲突，如图6所示。针对这些冲突，有多种启发式规则可以采用，如设立优先级进行避让[21]，任务少的机器人等待[22]，或是采取多种方式比如离开、绕路和启动前等待以避开不同冲突[20]，利用通道协议避免一条通道内的机器人冲突[23]等。
When multiple robots are running, conflicts will inevitably occur due to the staggered running routes. Conflicts between robots can be classified into the following categories: catch-up conflicts, intersection conflicts and opposite conflicts, as shown in Figure 6. For these conflicts, there are many heuristic rules that can be adopted, such as setting priorities to avoid [21], waiting for robots with few tasks [22], or adopting various methods such as leaving, detouring and waiting before starting to avoid different conflicts [20], using channel protocols to avoid robot conflicts in a channel [23], etc.
6.储位优化问题
6. Optimization of storage location
货架的存储位置决定了机器人将其搬运至工作站时行走的距离，货架离工作站越近，搬运耗费的时间越少，系统的效率越高。但是，如果货架都在工作站附近，机器人搬运时也聚集在工作站附近，反而会造成工作站附近的拥堵。因此，储位优化也成为研究的热点问题之一。对货架存储位置设置不同的策略，对实际运行时的影响不同[24]。现有研究中，采取的策略包括：
The storage location of the shelf determines the walking distance when the robot moves it to the workstation. The closer the shelf is to the workstation, the less time it takes to move and the higher the efficiency of the system. However, if the shelves are all near the workstation, the robots will also gather near the workstation during transportation, which will cause congestion near the workstation. Therefore, reservoir optimization has also become one of the hot issues in research. Setting different strategies for shelf storage locations has different impacts on actual operation [24]. In the existing research, the strategies adopted include:
（1）固定位置存储。货架固定在某一确定位置存储；
(1) Fixed position storage. The shelf is fixed at a certain position for storage;
（2）随机存储。货架从工作站离开时，随机选择空位存放；
(2) Random storage. When the shelf leaves from the workstation, randomly select a vacant space for storage;
（3）最近存储。货架从工作站离开时，选择离工作站最近的空位存放；
(3) Recent storage. When the shelf leaves the workstation, select the vacancy nearest to the workstation for storage.
（4）分区存储[5]。按照货架上货物被拣选的频率，对储区进行分区，高频区靠近工作站，低频区远离工作站，货架按照区域存储，在区域内部也可有“固定位”、“随机”、“最近”等存储策略。
(4) Partitioned storage [5]. According to the frequency of goods being picked on the shelves, the storage area is divided into areas. The high-frequency area is close to the workstation and the low-frequency area is far away from the workstation. The shelves are stored according to the area. There can also be storage strategies such as "fixed position", "random" and "nearest" inside the area.
7.充电策略问题
7. Charging Strategy Issues
机器人虽然不用像人类一样需要休息，但机器人的运行需要依靠电能，图3右下角就是机器人的充电区。机器人的充电策略涉及到以下两部分：
Although robots do not need to rest like human beings, the operation of robots depends on electric energy. The charging area of robots is in the lower right corner of FIG. 3. The charging strategy of the robot involves the following two parts:
（1）充电时机，即机器人还有多少电量时需要充电[25]。如果充电时机选择的过晚，就会出现有机器人因为电量耗尽而停留在储区或工作区内的情况，会直接影响任务的完成和其他机器人的运行。过早的充电虽然可以避免机器人没电停止，但机器人频繁的充电影响系统效率，同时也影响电池寿命，而且充电区的容量也是有限的，频繁充电可能导致充电区发生堵塞。
(1) Charging timing, that is, how much power the robot still has, needs to be charged [25]. If the charging timing is chosen too late, there will be robots staying in the storage area or the working area due to power depletion, which will directly affect the completion of tasks and the operation of other robots. Although premature charging can prevent the robot from stopping without electricity, frequent charging of the robot affects the system efficiency and battery life, and the capacity of the charging area is also limited. Frequent charging may lead to blockage of the charging area.
（2）充电模式。机器人可以选择在充电区充电，也可以直接换电池。现有研究的实验表明，换电池比充电更能提升系统性能[26]。
(2) Charging mode. The robot can choose to charge in the charging area or directly change the battery. Existing research experiments show that battery replacement can improve system performance more than charging [26].
8.实时调度问题
8. Real-time scheduling issues
实时调度是在系统实际运行时进行调度，对调度的时效性要求较高。因此，多采用一些规则处理系统运行中遇到的各种问题，如利用分配任务极为快速的令牌传递算法[27]；通过得到的当前机器人的状态，进行动态路径规划，避免机器人间的冲突[7,28]；实时调整机器人的优先级以协调冲突[29]。还有一些和系统输入相关的问题，例如紧急订单的出现，紧急订单优于其他订单进行指派，优于其他任务被分配给机器人完成[30]。此外，在实际运行中还会出现机器人故障、货物跌落等突发事件。
Real-time scheduling is to schedule when the system is actually running, which requires high timeliness of scheduling. Therefore, some rules are often used to deal with various problems encountered in the operation of the system, such as using token passing algorithm that assigns tasks very quickly [27]; According to the obtained current robot state, dynamic path planning is carried out to avoid conflicts between robots [7, 28]; Adjust the priority of the robot in real time to coordinate conflicts [29]. There are also some problems related to system input, such as the emergence of emergency orders, which are better than other orders for assignment and better than other tasks assigned to robots to complete [30]. In addition, in actual operation, there will be unexpected events such as robot failure and goods falling.
三、总结和展望
III. Summary and Prospect
多机器人存取系统具有更高的拣货效率、更好的系统可扩展性和柔性，因此成为近年来备受关注的研究领域之一。本文描述了多机器人存取系统中的布局优化、订单指派、任务分配、路径规划、冲突消解、储位优化、充电策略、实时调度等问题。除此之外，还有多个方面有待关注。
Multi-robot access system has higher picking efficiency, better system scalability and flexibility, so it has become one of the research fields that have attracted much attention in recent years. This paper describes the layout optimization, order assignment, task allocation, path planning, conflict resolution, storage optimization, charging strategy, real-time scheduling and other issues in multi-robot access systems. In addition, there are still many aspects to be paid attention to.
1.协同优化
1. Collaborative optimization
现有的仓库规模不断扩大，机器人数量逐渐增多，系统运行流程复杂，问题之间联系紧密，单个机器人的最佳路线、单个问题的最优策略在系统全局运行时并非最优。因此，需要进行协同优化，找到系统运行时最合适的策略组合，确定同时考虑多辆机器人多个问题的优化方法，共同提升系统效率。
The existing warehouse scale is continuously expanding, the number of robots is gradually increasing, the system operation process is complex, and the problems are closely linked. The optimal route of a single robot and the optimal strategy of a single problem are not optimal when the system is running globally. Therefore, it is necessary to carry out collaborative optimization, find the most suitable strategy combination when the system is running, and determine the optimization method that simultaneously considers multiple problems of multiple robots, so as to jointly improve the system efficiency.
2.数字孪生系统
2. Digital twin system
数字孪生是近年来的热点方向，数字孪生可以将虚拟系统和现实系统结合起来，利用虚拟系统模拟物理系统运行。多机器人存取系统内人类活动少，机器人行为可以预测，进而实现对整个系统运行情况的预测。通过对物理系统状态的预测，可以提前处理未来可能发生的问题，从而指导物理系统运行。
Digital twinning is a hot topic in recent years. Digital twinning can combine virtual system with real system and use virtual system to simulate the operation of physical system. There are few human activities in the multi-robot access system, and the robot behavior can be predicted, thus realizing the prediction of the operation of the whole system. By predicting the state of the physical system, the problems that may occur in the future can be dealt with in advance, thus guiding the operation of the physical system.
3.机器学习
3. Machine Learning
多机器人存取系统本身是一个复杂系统，系统内的问题很多。机器学习和人工智能的发展为系统运行提供了新思路。通过对大量经验数据的学习，将复杂问题简单化，快速做出决策，将是未来研究的新方向。
Multi-robot access system itself is a complex system, and there are many problems in the system. The development of machine learning and artificial intelligence provides new ideas for system operation. Through the study of a large number of empirical data, simplifying complex problems and making quick decisions will be the new direction of future research.
赞助本站
Sponsor this site
