Under the international background of "industry 4.0" proposed by Germany, the strategic policy of "made in China 2025" has been issued in China, aiming to promote the intelligent process of domestic manufacturing industry. Intelligent factory based on intelligent manufacturing has become the innovation goal of all walks of life, among which the transformation of unmanned processing production line has become the key to the transformation of intelligent factory . In recent years, most of the solid wood plate processing and production process has been automated, but the transportation and connection between various processes still need a lot of labor, and the continuous increase of labor cost restricts the development of enterprises. In order to improve the production efficiency of solid wood plate processing and realize the intelligent transformation of solid wood plate processing industry, this paper developed an automatic handling system of solid wood plate processing production line. By using automated guided vehicle (AGV) for unmanned transportation and loading and unloading, the problem of high handling cost between various processing procedures of solid wood plate can be effectively solved.
1. The overall design of the automatic handling system of the 1 solid wood plate processing production line
1.1 design requirements of automatic handling system
The design of automatic handling system is based on the current situation of solid wood plate processing of Jiangsu Jiangjia Machinery Co., Ltd. the production line processes logs into solid wood plates. Although the production line has basically realized automatic processing, in order to improve the utilization rate of timber, the factory splices short timber into specification timber, that is to say, comb and tenon the small size solid wood plates through the finger joint production line[ 2] The method of "finger joint" is used to combine with the long solid wood plate, and then the plate meeting the production specifications is sawed. In order to improve the intelligent level of solid wood plate processing, the enterprise improves the traditional processing production line, and adds image recognition device to the traditional production line for intelligent scanning recognition and selecting the defective plates. The intelligent cutting production line based on image recognition uses image recognition device to determine the position information of the defects (such as insect eye, scab, etc.) of solid wood plate , and then cuts the defect part through the intelligent cutting device, so that the defect free part can be extended for use.
The automatic handling system is applied to the connection between the traditional automatic processing production line of solid wood plate, the intelligent cutting production line of plate and the intelligent finger joint processing production line, so as to realize the intelligent handling and intelligent loading and unloading of solid wood plate. Through image recognition device and pop-up device, the traditional processing production line is reformed, the traditional production line is upgraded intelligently, the intelligent selection function of defective materials is added, the loading and stacking operation of plates in the production function area is realized by using the rectangular coordinate robot, and then all kinds of plates are transported to the corresponding specification material area by AGV, so as to realize the unmanned production and moving of the solid wood plate processing plant Yun. Figure 1 shows the intelligent cutting production line of sheet metal, and Figure 2 shows the intelligent finger joint processing production line.
The specification of the solid wood plate processed by the traditional automatic processing line and the intelligent finger joint processing line is 1200mmx150mmx18mm. In order to keep up with the production rhythm and ensure the efficiency, the tray capacity should be able to accommodate 5 rows of plates. In order to ensure the stability of the plate handling process of AGV, the four corners of the tray should be equipped with a support. In order to facilitate handling, the lifting and shifting AGV shall be selected. At the same time, the AGV shall be equipped with intelligent navigation and positioning device to ensure the position accuracy of handling. Its structure is shown in Figure 1.
1.2 environmental layout of automatic handling system
The application environment of the automatic handling system includes the traditional production line, the intelligent cutting production line of the board, the intelligent finger processing production line, the defective wood stacking area, the AGV rest area, the feeding area, the storage entrance sorting platform and the office control center, among which the AGV rest area is the place for AGV charging and fault maintenance, while in the actual production process, the traditional production line has the defective solid wood board The quantity is relatively small. In order to give full play to the role of the intelligent cutting production line and the intelligent finger joint processing production line, the factory additionally purchases low-cost and defective plates for processing, stacks these plates in the feeding area, and uses multiple AGVs to connect the various functional areas.
Because the traditional production line mainly processes the plates with few defects, there are not many plates to be finger jointed. Most of the handling tasks are carried out in other functional areas, among which the largest handling volume is between the feeding area and the sorting platform and other functional areas. Considering that the processing time of the intelligent cutting production line and the intelligent finger jointing production line is long, the plate output and stacking can complete a pallet unit The time is slow and the handling demand is relatively small. In addition, in order to reduce the AGV travel, the incomplete wood stacking area is designed near the feeding area. The number of AGV needed in each work area is calculated by analytic method, which is defined as the total time (in seconds) for the average AGV to complete one plate handling process.
𝑇r = 𝑇w + 𝑇k + 𝑇m + 𝑇z (1)
In equation (1), T w is the average AGV waiting time, 𝑇 k is the average no-load travel time, and 𝑇 m ; is the average full load Travel time, 𝑇 z is the average sheet material conversion time, and defines ω as the sheet material demand time, that is, a handling task will be generated every ω seconds, and the number of times per hour is C = 3600 / ω, Then the total time required for the AGV to complete all tasks C per hour is T a = C × T r , and let the average working time per hour excluding the charging time of the AGV be μs , You can get the basic formula of the number of AGVs: N = T a / μ, set the working time of the AGV to 50 minutes per hour, and the average time for the AGV to complete one handling is 200 seconds, calculated according to the amount of tasks in each work area Between the traditional production line, the feeding area, the plate intelligent cutting production line and the defective wood accumulation area, 4 AGVs are required to carry the plate back and forth. When the defective material output of the traditional production line reaches a pallet capacity, 1 of the 4 vehicles will be transferred Transportation to the intelligent cutting production line, 3 AGVs are used to carry the plate between the intelligent cutting production line and the intelligent finger joint processing production line , Intelligent finger joint processing production line, traditional production line and warehouse entrance sorting table are handled by 4 AGVs, a total of 11 AGVs. The environmental layout of the automatic handling system is based on a production area of 80mx50m, of which the traditional production line partially occupies 65mx15m. On this traditional production line, mainly processed relatively good plates, only a small number of these plates have defects, and are selected by robots and placed on the defective materials. Stacking port, the finished non-defective finished sheet is placed in the finished product stacking port, and the occupied area and position of other functional areas are shown in Figure 3.
2 Design of automatic navigation and transportation system (AGVS)
The Automated Guided Vehicle System (AGVS) mainly includes a central control module, an AGV monitoring module, a positioning module, a communication module, and an automatic charging module. Through this system, the distribution and scheduling of handling tasks , AGV automatic positioning And navigation, AGV path planning and motion control  are the core parts of the unmanned production handling system.
2.1 Design of navigation system
At present, the commonly used AGV navigation methods on the market mainly include magnetic navigation, laser navigation , inertial navigation, and visual navigation . In visual navigation, QR code (Quick Response Code) navigation is favored by many AGV manufacturers because of its efficient and stable characteristics. In the environment of solid wood sheet processing factories, AGV has certain requirements for handling speed and accuracy, so this article uses QR code navigation, each code has a unique number, and the AGV car camera recognizes and reads the information to achieve AGV navigation control. The code label material uses wear-resistant labels and is regularly replaced to ensure effective identification. In addition, in order to reduce the impact of dust in the factory on the identification, it is required to equip the AGV with a dust blowing device to identify the possible dust before identifying the two-dimensional code get rid of. The recognition calculation process of QR codes is shown in Figure 4 (b), where 𝑁 1 , 𝑁 2 , 𝑁 3 are the positioning pattern outlines, ⊿𝑁 1 𝑁 2 𝑁 3 is an isosceles right triangle; the origin 𝑂 is the center of the requested QR code, vector 𝑁 1 𝑁 2 is the rotation axis vector.
et the error angle of the AGV car camera to scan the QR code as 𝜃, the value range is [-180 °,+180 °], and the counterclockwise rotation is positive. If the actual rotation vector is 𝑁1′𝑁2 ′, the error angle can be calculated by formula (2):
The horizontal and vertical position errors are:
In equation (3), ∆𝑥 and ∆y are horizontal and vertical position errors, respectively, the actual coordinate of the QR code collected is 𝑂 (𝑂𝑥, 𝑂𝑦), the AGV car camera resolution is (2𝑊) × (2𝐿), 𝑊, 𝐿 Horizontal pixels and vertical pixels respectively 𝑆 𝑥 is the pixel area of the QR code positioning pattern, 𝑆 0 is the real area of the QR code, and D is the side length of the positioning pattern. The position of the AGV relative to the QR code is represented as P (𝑥, 𝑦, 𝜇), where 𝑥 = 𝑂 𝑥 + ∆ 𝑥 , 𝑦 = 𝑂 𝑦 + ∆ 𝑦 , 𝜇 = 𝜃, the coordinates of the AGV in the entire electronic map are set to (𝑋 𝑖 , 𝑌 𝑖 ), From scanning to the current QR code until the AGV reaches the next QR code, the position expression (𝑋, 𝑌, 𝛾) of the AGV in the electronic map coordinate system at any time can be obtained by equation (4), 𝛾 indicates that the AGV position and pose are positive and Angle deviation in the direction of the travel path. The placement of the two-dimensional code needs to ensure that the AGV can continuously recognize the two-dimensional code, so as to determine its position information in real time and achieve the effect of traveling in any direction.
2.2 Establishment of electronic map
The electronic map is the basis of the AGVS visual interactive interface, which can display the running status of the AGV in real time. There are three main types of electronic maps used for AGVS: feature maps , topological maps , and grid maps . Among them, the grid map divides the actual map into grids of the same size, which is convenient for creation, representation and maintenance. It is convenient for short path planning, and in the solid wood plate automatic handling system designed in this paper, the position of each device is fixed and unchanged, so it is used. Raster maps express the environment. The actual map is provided by the company and verified by the actual. After the work environment is reasonably transformed, the working area and the AGV drivable area are determined. Only the AGV drivable area is rasterized. According to the plate specification length of 1.2m, the grid size is set to 2x2m, to ensure enough AGV safe driving space to avoid congestion and collision. The electronic map created according to the planned system environment layout is shown in Figure 5. In order to simplify the task distribution of AGV, the map is divided into three main work areas: sheet intelligent cutting work area A, intelligent finger joint processing work area B, Finished sheet handling work area C. When the system is operating normally, AGVs are only moved within each work area. When there is a large difference in the amount of handling tasks between work areas, the system reallocates the number of AGVs in each work area. The non-travelable areas such as pillars in the working area of the actual plant are represented by a black grid with a standard shape on the map, and the grid outline contains the safe distance between the AGV and the obstacle.
2.3 path planning and design
2.3.1 path planning algorithm
The commonly used AGV path planning algorithms include Dijkstra algorithm , a * algorithm , genetic algorithm , ant colony algorithm , etc. Jiang Kang  mainly considers the interference between the cable and the component when arranging the cable, and uses the rigidity factor to evaluate the cable diameter when bending continuously to improve the a * algorithm to complete the reasonable routing. Yu Bixiu  Based on the need of unmanned vessel to return to the preset route after avoiding obstacles, uses different evaluation functions to solve the route planning for different positions of the vessel, so that the vessel can return to the preset route faster. Based on the real wood production environment, this paper puts forward the basic rules of path algorithm: to ensure that AGV needs the least time in actual driving; @ path information is as simple as possible. On the basis of the initial shortest path, on the one hand, reduce the number of turns as much as possible, and remove the deceleration and re acceleration time needed by AGV to shorten the actual driving time, on the other hand, remove the intermediate nodes in the straight path, and only record and transfer the key node information to AGV. The overall formula of the algorithm is:
𝐹(𝑛) = 𝐺(𝑛) + 𝐻(𝑛) + 𝑇(𝐾𝑛) + 𝑆(𝜔) (5)
Equation (5) is the solution function of the initial shortest path, which is the cost function, which represents the true generation value from the initial node to 𝐺 (𝑛) + 𝐻 (𝑛) 𝐺 (𝑛) current node 𝑛, 𝐻 (𝑛): heuristic function , Which represents the evaluation value of the current node to the target node. Commonly used heuristic functions are Manhattan distance function , diagonal distance function and Euclidean distance function . In the map environment of the design system in this paper, The heuristic function uses the Manhattan distance function, that is, the sum of the distance difference between the current node and the target node in the X-axis direction and the Y-axis distance. The calculation formula is:
𝐻(𝑛) = |𝑥𝑛 ― 𝑥𝑔𝑜𝑎𝑙| + |𝑦𝑛 ― 𝑦𝑔𝑜𝑎𝑙| (6)
Equation (5) is the collinear node optimization function to determine whether the current node is a collinear node. The method is: the known node p i + 1 is the current position of the AGV, and the coordinates in the map are （x𝑖 + 1, 𝑦𝑖 + 1), the node 𝑃𝑖 is the node that the AGV just drove, the coordinate is (x𝑖, 𝑦𝑖), the node p i + 2 is the next node to be reached, the coordinate is (x i + 2 , y i + 2 ).
According to formula (7) (8), the slopes of line P pipi+1 and line P pi+1pi+2 are Kpipi+1 and Kpi+1pi+2，if Kpipi+1=Kpi+1pi+2，Then nodes pi、pi+1、pi+2 are redundant collinear nodes in tihe same driving direction, so they are not added to the path node calculation list,At this time,if Kpipi+1≠Kpi+1pi+2, then the node is an inflection point, at this time 𝑇 (𝐾𝑛) = 1.
In formula (5), 𝑆 (𝜔) is the optimization function of the excess inflection point, to determine whether the corner node is an excess inflection point, the method is: known nodes pi、pi+1、pi+2 are three consecutive nodes at the corner of the path, the node pi+1 is the corner node, pi, pi+2 are the two nodes before and after, if there is no obstacle between the connection between node pi and node pi+2, then directly delete node pi+1; if there is an obstacle on the connection path, determine whether the connection intersects with one of the diagonals of the obstacle, if it intersects, the node pi+1 Save, otherwise delete the node pi+1.
As shown in Fig. 6, the AGV wants to reach the node pi+2 from the node pi, the nodes pi, pi+1, pi+2 are (xi,yi)、(xi+1,yi+1)、(xi+2,yi+2), nodes 𝑍1, 𝑍2, 𝑍3, 𝑍4 are the four vertices of an obstacle in the map, the coordinates are expressed as (xZ1,yZ1)、(xZ2,yZ2)、(xZ3,yZ3)、(xZ4,yZ4).
In formulas (9)-(12), 𝜔1, 𝜔2, 𝜔3, 𝜔4 are vector cross multiplication coefficients, and 𝜎 is a unit vector.If 𝜔1 ∙ 𝜔2 ＜ 0 and 𝜔3 ∙ 𝜔4 ＜ 0, the line segments pipi+2 and Z1Z3 Intersect, as shown in FIG. 6 (b), otherwise do not intersect, as shown in 6 (a), using the same method to determine the line segment pipi+2Intersect with Z2Z4. If the line segment pipi+2 intersects with any diagonal line segment of the obstacle, the inflection point pi+1 is added to the path node list, At this time 𝑆 (𝜔) = 1, the AGV driving path node is pi-pi+1-pi+2, if not Intersect, the turning point pi+1 is not added to the list, at this time 𝑆 (𝜔) = 0, the AGV driving path node is pi-pi+2. The flowchart of the node reduction algorithm is shown in Figure 7:
As shown in Figure 8, the results of node optimization algorithm and a * algorithm are compared, which are the working paths of area a and area C respectively. In the electronic map, the border between the working area and the obstacle already contains the safe distance from the AGV. If the path removal cannot intersect with its border, the map can reach any desired position. According to the side length of each grid as 1, the specific data is shown in Table 1. From the start point (2.5,4.5) to the end point (14.5,15.5), compared with the a * algorithm, the node optimization algorithm reduces the number of calculation path nodes by 13 (72%), the path length by about 1.17 (6%), and the number of turns by 2 (40%), which can effectively shorten the handling time of AGV.
2.4 coordination rules of multiple AGVS
The optimal path of a single AGV is determined by the path planning algorithm, but when multiple AGVs are running at the same time, it is inevitable that the optimal path will cross, resulting in road congestion or even AGV collision. Unreasonable scheduling will reduce the working efficiency of AGV and affect the production beat . Therefore, it is necessary to carry out collaborative planning for the simultaneous operation of multiple AGVs to prevent congestion and collision. According to the driving situation of AGV, there are three possible congestion and collision situations: ① rear end collision; ② opposite collision; ③ side collision, as shown in Figure 9:
Xie Yongliang  designed an obstacle avoidance algorithm based on the principle of ultrasound, which enables AGV to bypass the static obstacles in front, slow down or stop to avoid the dynamic obstacles when it goes straight, but it can not effectively deal with the complex dynamic congestion and collision environment of multiple AGVs. He Changzheng  proposed a hybrid genetic algorithm based on the time window and Dijkstra algorithm. Through the method of speed adjustment and new path replacement, aiming at the complex congestion situation of multiple AGVs, it can accurately plan a collision free and conflict free shortest path for AGVs. However, the resulting calculation will affect the operation efficiency of the whole system, and the path transformation will increase the number of turns of AGVs, and increase the actual situation Travel time. In this paper, the system is designed from the aspects of software and hardware to prevent and deal with problems effectively. In the aspect of software, by establishing mathematical model function to optimize the path solution, AGV will not travel in the same straight line path in the opposite direction, and based on the solution path to drive, reduce the possibility of further increase of conflict. Under the rule of minimum actual driving time, based on the planning path to predict the collision nodes in advance, AGV can prevent the possible collision while driving.
The objective function of collision prevention processing is as follows:
𝑞: AGV collection, 𝑞 = (1,2,3, ···, 𝑞𝑘), the number of AGVs is 𝑘;
𝑅: A collection of all nodes in the map, 𝑖, 𝑗 ∈ 𝑅 to indicate the position of the AGV in 𝑅;
𝑑𝑖, 𝑗: Distance between node 𝑖 and node 𝑗;
Equation (14) indicates that after the AGV 𝑞𝑘 passes the node 𝑖, the next driving target has only a unique node 𝑗; Equation (15) indicates that the AGV at the node 𝑗 𝑞𝑘 has passed the node and there is only a unique node 𝑖; Equation (16) constrains the AGV to move from the beginning of the path 𝑂𝑠 indicates the starting point of the path; formula (17) constrains the AGV to reach the end of the path; 𝑂𝑓 indicates the end of the path; formula (18) indicates that the path where the AGV is traveling can only pass in one direction.
The AGV applies for occupying the forward path node in real time while driving. If the path overlaps, the coincident node is occupied by one of the AGVs, and the other AGV cannot apply, which is a conflict. If the AGV continues to move, a collision may occur, as shown in the figure As shown in 10, AGV No. 1 has occupied the node, and AGV No. 2 performs conflict prevention.
In this case, the safety distance is used to limit the AGV's driving speed to prevent collisions. In equation (19), Plock represents the length of the next section of the path that the AGV 2 applies for occupation, Vmaxis the maximum traveling speed of AGV, V1 is the minimum traveling speed of AGV, namely 0m / 𝑠, Vm0 is the current traveling speed of AGV, a is the acceleration of AGV , Rs represents the safety distance required for the AGV to decelerate from the maximum speed to stop. It can be seen in Figure 10 that the range of this distance is a circle, representing the safety distance at which the AGV from different directions stops at the origin. Ls indicates that after the No. 2 AGV failed to apply for the occupation of the node, a pre-deceleration distance. During the No. 2 AGV pre-deceleration, if the No. 1 AGV has drove away and the conflicting node is released, the No. 2 AGV gradually accelerates from the current speed. The conflicting node was not released, and AGV 2 stopped when it reached a safe distance, and gradually accelerated through when AGV 1 moved out of the safe distance.
Formula (20) according to the conflicting nodes in the two AGV driving paths obtained by the path planning algorithm to determine the possibility of AGV conflict during driving, 𝑈(path1,path2) represents the driving path path1 and path2 The set of conflicting nodes; Equation (21) is based on the queuing algorithm and determines the priority of the AGV on a first-come-first-served basis. The AGV that applies to the node first passes first, and the AGV that fails to apply slows down and waits for the AGV with higher priority to pass Driving, tpath1k1uh represents the time required for the k1th AGV in the driving path path1 to reach the ℎ conflicting node, and Ppath1k1 represents the priority of the first AGV in the driving path path1 level.
In terms of hardware, first of all, ensure that each AGV is equipped with a real-time distance measuring device, so that there is a sufficient safety distance between the AGV and other vehicles and obstacles, and secondly, to prevent system failure, each AGV is equipped with a bumper for AGV Buffer the impact force in case of collision to protect the equipment.
2.5 system example
The system monitoring and control interface is shown in Figure 12. 11 AGVs are displayed on the electronic map in real time. Work area A contains AGVs 1, 2, 3, and 4, area B contains AGVs 5, 6, and 7, and area C contains 8. No. 9, 10, 11 AGV, the path of AGV is displayed as the path from the current position of AGV to the end point. The starting and ending points of the path in the console are automatically entered by the system by default, and can be changed by the operator in special circumstances. When the communication status of the AGV is displayed as a fault, the system will alarm and activate the emergency stop button to stop the AGV immediately to avoid accidents. occur. According to the AGV specifications, the linear speed of the unloaded AGV is 1 m / s, the turning speed is 0.5 m / s, the linear speed of the cargo AGV is 0.6 m / s, the turning speed is 0.3 m / s, and the AGV safety distance is calculated at the maximum speed 𝑅s = 1.6m, driving at a turning speed within a safe distance, so it takes about 5 seconds longer for the AGV to pass through the safe area of the turning than to travel straight. According to the handling speed and turning time-consuming, the time required for the AGV's limit working distance in each work area is shown in Table 2, which is calculated based on the maximum travel speed of the cargo AGV to obtain the shortest AGV in each work area Time required for working distance and maximum working distance.
In this paper, the automatic handling system of solid wood plate processing line is designed to solve the problem of high labor cost in material handling between production lines. Based on the design requirements of the automatic handling system, the main components of the system are determined. According to the specifications and handling requirements of the plate, each production line is modified and the type of handling AGV is selected to ensure the intelligent and smooth operation of the system. The design of system environment layout maintains the balance between production efficiency and handling efficiency, improves the stability of the system while effectively using the space, and improves the production efficiency. Through the design of AGV navigation system, the establishment of electronic map, the improvement of path algorithm, the formulation of system task scheduling rules and multi AGV cooperative operation system, the automatic handling system can realize a series of functions such as task intelligent distribution, AGV intelligent driving and control. The research of the automatic handling system of the solid wood plate processing production line makes the material connection between the production lines in the plate factory realize the unmanned operation, saves the labor cost, improves the production efficiency, and plays a positive role in promoting the intelligent transformation and development of the solid wood plate processing enterprise.
 ZHOU Jiajun, YAO Xinfan, LIU Min, et al. State-of-Art review on new emerging intelligent manufacturing paradigms[J]. Jisuanji Jicheng Zhizao Xitong/Computer Integrated Manufacturing Systems, CIMS, 2017, 23(3):624-639(in Chinese).[周佳军,姚锡凡,刘敏,张剑铭,陶韬.几种新兴智能制造模式研究评述[J].计算机集成制造系统,2017,23(3):624-639.]
 XIE Lisheng, LIU Jiaquan, QIAO Luting, et al. Finger-Jointing Technique of Larch Structural Glued Laminated Timber[J]. Scientia Silvae Sinicae. 2016, 52(9): 107-112(in Chinese).[谢力生,刘佳权,乔鹭婷,李贤军,周先雁.结构用集成材落叶松层板指接工艺[J].林业科学,2016,52(9):107-112.]
 ZHAO Peng, ZHAO Yun, CHEN Guangsheng. Quantitative analysis of wood defect based on 3D scanning technique[J]. Nongye Gongcheng Xuebao/Transactions of the Chinese Society of Agricultural Engineering, 2017, 33(7):171-176(in Chinese).[赵鹏,赵匀,陈广胜.基于3D扫描技术的木材缺陷定量化分析[J].农业工程学报,2017,33(7):171-176.]
 HUO Kaige, ZHANG Yaqi, HU Zhihua. Research on scheduling problem of multi-load AGV at automated container terminal[J]. Journal of Dalian University of Technology, 2016,56(3):244-251(in Chinese).[霍凯歌,张亚琦,胡志华.自动化集装箱码头多载AGV调度问题研究[J].大连理工大学学报,2016,56(3):244-251.]
 Rundong Y , Dunnett S J , Jackson L M . Novel methodology for optimising the design, operation and maintenance of a multi-AGV system[J]. Reliability Engineering & System Safety, 2018, 178:130-139.
 HE Zhen, LOU Peihuang, QIAN Xiaoming, et al. Research on precise positioning technology for AGV based on multi-object vision and laser integrated navigation[J]. Chinese Journal of Scientific Instrument,2017,38(11):2830-2838(in Chinese).[何珍,楼佩煌,钱晓明,武星,朱立群.多目视觉与激光组合导航 AGV 精确定位技术研究[J].仪器仪表学报,2017,38(11):2830-2838.]
 Shi C , Lou P , Wu X , et al. Collaborative calibration of initiative multi-camera guiding system for transport automatic guided vehicle[J]. Yi Qi Yi Biao Xue Bao/Chinese Journal of Scientific Instrument, 2014,35(11):2589-2599(in Chinese).[石陈陈,楼佩煌,武星,钱晓明.自动导引车多摄像机主动导引系统的协同标定[J].仪器仪表学报,2014,35(11):2589-2599.]
 ZHANG Junning, SU Qunxing, LIU Pengyuan, et al. An Improved VSLAM Algorithm Based on Adaptive Feature Map[J]. Acta Automatica Sinica. 2019, 45(3): 553-565(in Chinese).[张峻宁,苏群星,刘鹏远,朱庆,张凯.一种自适应特征地图匹配的改进 VSLAM 算法[J].自动化报,2019,45(3):553-565.]
 Blazquez C A, Ries J, Leon R J, et al. An Instance-Specific Parameter Tuning Approach Using Fuzzy Logic for a Post Processing Topological Map-Matching Algorithm[J]. IEEE INTELLIGENT TRANSPORTATION SYSTEMS MAGAZINE. 2018, 10(4SI): 87-97.
 Nuss D, Reuter S, Thom M, et al. A random finite set approach for dynamic occupancy grid maps with real-time application[J]. INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH. 2018, 37(8): 841-866.
 Ciesielski K C, Falcao A X, Miranda P. Path-Value Functions for Which Dijkstra's Algorithm Returns Optimal Mapping[J]. JOURNAL OF MATHEMATICAL IMAGING AND VISION. 2018, 60(7): 1025-1036.
 Kang N K, Son H J, Lee S H. Modified A-star algorithm for modular plant land transportation[J]. JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY. 2018, 32(12): 5563-5571.
 Mokshin A V, Mokshin V V, Sharnin L M. Adaptive genetic algorithms used to analyze behavior of complex system[J]. COMMUNICATIONS IN NONLINEAR SCIENCE AND NUMERICAL SIMULATION. 2019, 71: 174-186.
 Xiong C K, Chen D F, Lu D, et al. Path planning of multiple autonomous marine vehicles for adaptive sampling using Voronoi-based ant colony optimization[J]. ROBOTICS AND AUTONOMOUS SYSTEMS. 2019, 115: 90-103.
 Kang J, Shiji M. A Cable Path Planning Method Based on Improved A* Algorithm[J]. China Mechanical Engineering, 2019, 30(06): 699-708(in Chinese).[姜康,马世纪.基于改进 A~*算法的线缆路径规划方法[J].中国机械工程,2019,30(06):699-708.]
Bixiu Y, Xiumin C, Chenguang L, et al.A Path Planning Method for Unmanned Waterway Survey Ships Based on Improved A* Algorithm[J]. Geomatics and Information Science of Wuhan University,2019,44(08):1258-1264(in Chinese).[余必秀,初秀民,柳晨光,张豪,毛庆洲.基于改进A~*算法的无人航道测量船路径规划方法[J].武汉大学学报(信息科学版),2019,44(08):1258-1264.]
 Clempner J B, Poznyak A S. Using the Manhattan distance for computing the multiobjective Markov chains problem[J]. INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS. 2018, 95(11): 2269-2286.
 Alencar J, Lavor C, Liberti L. Realizing Euclidean distance matrices by sphere intersection[J]. DISCRETE APPLIED MATHEMATICS. 2019, 256(SI): 5-10.
Zhiyong L, Shuhai J. Review of mobile robot path planning based on reinforcement learning[J]. Manufacturing Automation, 2019,41(03):90-92(in Chinese).[刘志荣,姜树海.基于强化学习的移动机器人路径规划研究综述[J].制造业自动化,2019,41(03):90-92.]
 Yongliang X , Jianjun Y , Chengchao Y U , et al. Obstacle Avoidance Navigation Algorithm and Analog Experiment for Wheeled AGV Running along Vineyard Ridge Road[J]. Transactions of the Chinese Society for Agricultural Machinery, 2018(in Chinese).[谢永良,尹建军,余承超,贺坤,胡旭东,李仁旺.轮式 AGV 沿葡萄园垄道行驶避障导航算法与模拟试验[J].农业机械学报,2018,49(07):13-22.]
 He C , Song Y , Lei Q , et al. Integrated Scheduling of Multiple AGVs and Machines in Flexible Job Shops[J]. Zhongguo Jixie Gongcheng/China Mechanical Engineering, 2019, 30(4):438-447(in Chinese).[贺长征,宋豫川,雷琦,吕向飞,刘软香,陈进.柔性作业车间多自动导引小车和机器的集成调度[J].中国机械工程,2019,30(04):438-447.]