An AGVS dynamic collision-free path planning method based on multi-agent simulation

Abstract

The application discloses a kind of AGVS dynamic non-collision path planning methods based on multi-agent simulation, specifically: the hierarchical hybrid multi-agent system including task management Agent, AGV management Agent, road network management Agent, AGV Agent, node Agent, section Agent is constructed, various kinds of agents have perception, decision, judge, interactive ability, utilize knowledge method to mutually cooperate and carry out dynamic non-collision path planning: the improved A* path planning algorithm is used to provide shortest path planning for AGV Agent, conflict prediction and resolution strategy based on six categories of conflict situation is used to provide prediction and elimination method for AGV dynamic real-time conflict in road network, road network congestion degree calculation method is used to update the congestion state of path network in real time, cooperate A* path planning algorithm selects non-collision shortest path.The application makes AGV logistics system can smoothly and efficiently run and no conflict deadlock occurs, has better transport efficiency and collision avoidance performance.

Description

Technical Field

[0001] This invention belongs to the field of AGV intelligent control technology, and in particular relates to a dynamic collision-free path planning method for AGVS based on multi-agent simulation. Background Technology

[0002] Optimizing AGVS parameters only improves the expected performance at the design and planning level. However, it doesn't guarantee efficient and low-cost operation after deployment in production. In reality, path planning and dynamic scheduling during operation are also crucial. In actual manufacturing, multiple AGVs often need to be deployed simultaneously to complete material transport tasks. Clearly, AGV path planning is the foundation for efficient logistics in manufacturing systems. AGV path planning involves planning the route from the starting point to the destination for multiple AGVs based on the transport task, considering conflicts, waiting, avoidance, line switching, and charging during transport, with the goal of minimizing task completion time, maximizing task completion volume, and minimizing total transport distance or total energy consumption. With the widespread application of AGVs in manufacturing systems, the use of multiple AGVs has expanded from single AGVs to multiple AGVs. The congestion, conflicts, and deadlocks caused by multiple AGVs operating in the path network are increasing, easily leading to collisions and blockages, and even system paralysis. Therefore, collision avoidance and deadlock prevention have become critical, requiring solutions to conflicts, collisions, and deadlocks during AGV operation. Furthermore, because AGVS systems and their associated manufacturing systems are typical complex discrete event systems, characterized by flexibility, randomness, dynamism, and complexity, their real-time state is often difficult to plan and predict as the system runs and a series of events progress. In addition, dynamic events such as task changes, AGV malfunctions, machine tool failures, and AGV charging can also affect or disrupt the entire system's operation. Therefore, path planning must consider dynamic, real-time, and rapid responses under various conditions. Consequently, current AGV path planning has evolved from single-AGV path planning and static path planning for multiple AGVs to dynamic path planning for multiple AGVs that considers collision avoidance. The problem of dynamic collision-free path planning for multiple AGVs is a key issue in improving the logistics efficiency of manufacturing systems and thus enhancing their operational effectiveness.

[0003] Dynamic collision-free path planning for multiple AGVs mainly incorporates dynamic strategies into static path planning methods. However, current research exhibits significant differences in problem description and expression, with varying perspectives and a lack of a unified mathematical model, resulting in limited applicability. Most solution methods, such as intelligent optimization algorithms, rely on centralized scheduling and can achieve satisfactory solutions, but their accuracy and efficiency are limited as the problem scales up or becomes more complex. Other methods are primarily designed for specific models, lacking broad adaptability or real-time performance, and their collision avoidance strategies are not flexible enough to adapt to various dynamic events. Therefore, there is a need to find a low-cost, high-efficiency, and widely applicable method to solve the dynamic collision-free path planning problem for multiple AGVs. Summary of the Invention

[0004] To address the above problems, this invention provides a dynamic collision-free path planning method for AGVS based on multi-agent simulation.

[0005] The present invention provides a dynamic collision-free path planning method for AGVS based on multi-agent simulation, specifically as follows:

[0006] A. Analyze the abstraction methods, expression mechanisms, and operation control of the AGVS logistics transportation system. Use object-oriented methods and graphical modeling and visualization techniques to model the AGVS logistics transportation system and construct an AGVS logistics and road network model, including logistics task management, material transportation process management, as well as AGV body model, path network model, node model, road segment model, AGV docking station, various loading and unloading station models, and workpiece model.

[0007] B. Based on the AGVS logistics and road network model, analyze the logical relationships between various entities in AGVS, the operation mechanism of system events, propose the types and functions of multi-agent systems, and construct a hybrid architecture multi-agent system (MAS) composed of various types of agents.

[0008] The hybrid architecture multi-agent system (MAS) has a top layer of shop floor management agents, a middle layer of task management agents, AGV management agents, and road network management agents, and a bottom layer of agents corresponding to various system elements, including workpiece agents, AGV agents, node agents, and road segment agents. The shop floor management agent, as the highest manager of the system, possesses global information and the highest authority. The bottom layer of various element agents are the executors or collaborators for completing transportation tasks. The middle layer management agents, as intermediaries for communication between the upper and lower layers, not only convey the decisions of the shop floor management agents to the bottom layer but also collect, summarize, and feed back the information from the bottom layer agents to the top layer agents.

[0009] C. Use the improved A* algorithm to solve for the shortest path of a single AGV.

[0010] The improved A* algorithm's cost function f(n) is the sum of the actual cost and the heuristically estimated cost, i.e.:

[0011] f(n)=g(n)+h(n) (1)

[0012] Where n represents the current node, g(n) represents the actual cost of the AGV from the starting node to the current node, and h(n) represents the estimated cost of the AGV from the current node to the target node, i.e., the heuristic function.

[0013] Compared to introducing an acceleration factor λ into the traditional heuristic function to reduce the number of nodes traversed by the A* algorithm, considering the aspect ratio of the Manhattan distance and introducing a scaling factor α to coordinate the balance in the two-dimensional search direction, the heuristic function is:

[0014] h(n)=λ(|x n -x end |+α|y n -y end |) (2)

[0015] Among them, (x n ,y n (x) represents the x-coordinate and y-coordinate of the current point, respectively. end ,y end ) represent the x-coordinate and y-coordinate of the target point, respectively.

[0016] D. AGV conflict prediction and handling based on multi-agent system.

[0017] Conflict detection between AGVs is achieved through comprehensive prediction and judgment based on information exchange among AGV Agents, node Agents, road segment Agents, and road network management agents. Conflict elimination is achieved by the AGV management agent controlling the AGV Agents' normal driving, deceleration / stop to avoid obstacles, and path switching commands. Solutions for predicting and eliminating each type of conflict are provided, and the principle of conflict elimination is:

[0018] (1) Try to maintain the optimal path planning of the original AGV to minimize the task time of each AGV.

[0019] (2) Prioritize the AGVs that are closest to the node to pass through in order to improve passage efficiency.

[0020] (3) When three or more AGVs conflict, one AGV should be prioritized to pass through as soon as possible in order to reduce congestion at nodes and road sections and simplify the conflict situation.

[0021] Furthermore, the definitions and functions of each agent in the hybrid architecture multi-agent system (MAS) are discussed.

[0022] (1) Workshop Management Agent

[0023] The workshop management agent is the highest manager of the system, possessing global information and the highest authority of the entire AGV system, and serves as the system management agent. The workshop management agent communicates with the task management agent, AGV management agent, and road network management agent to coordinate the initiation and task execution processes of the intermediate management agents and collect and summarize overall transportation information.

[0024] (2) Task Management Agent

[0025] The task management agent manages all material / workpiece agents in the system, summarizing the status information of all workpieces in real time, including: workpiece number, AGV number carrying the workpiece, and location information, and updating the status data in a timely manner for aggregation and real-time monitoring. At the same time, the task management agent communicates and negotiates with the AGV management agent, sending workpiece transportation requests to the AGV management agent, negotiating the assignment and allocation between workpieces and AGVs, assigning AGVs to undertake transportation tasks, and binding the AGV agent with the workpiece agent being transported.

[0026] (3) Workpiece Agent

[0027] The workpiece agent is a mapping of transported materials / workpieces in the system. Each workpiece corresponds to one workpiece agent, which carries origin and destination information. Workpiece agents are uniformly managed by the task management agent. The task management agent and AGV management agent negotiate to determine the order of transport tasks and negotiate the allocation and assignment between workpieces and AGVs. Once the assignment is completed, the workpiece is bound to the AGV, and the AGV completes the subsequent transport task of the workpiece. The workpiece agent promptly uploads its own status information, including the AGV number carrying the workpiece and the workpiece's location information in the road network, to the task management agent for aggregation and real-time monitoring.

[0028] (4) AGV Management Agent

[0029] The AGV management agent manages all AGV agents in the system, summarizing real-time status information for all AGVs, including AGV number, speed, no-load status, load status, and fault information, and updating the status data promptly. As the central layer of the intermediate management layer, the AGV management agent communicates and interacts with the workpiece management agent and the road network management agent. It receives transportation requests from the workpiece management agent, assigns AGVs to undertake transportation tasks, and binds the AGV agents to the workpiece agents being transported. Furthermore, the AGV management agent interacts with the road network management agent, receiving conflict resolution control commands from the road network management agent and transmitting them to the corresponding AGV agents for execution. Simultaneously, the AGV management agent sends the AGV's position and speed information to the road network management agent for aggregation and real-time monitoring.

[0030] (5) AGV Agent

[0031] An AGV Agent is a mapping of AGVs in the system. Each AGV corresponds to one AGV Agent, which completes a series of specified transportation tasks, moving workpieces from the starting point to the destination through a path network. The AGV Agent is a hybrid intelligent agent structure, capable of both planning autonomous behavior and responding to events. The AGV Agent can perceive the external environment and interact with other Agents. It updates the current state data of the AGV, including its position, speed, empty load, loaded status, and transportation task information. The AGV Agent's own decision-making and calculation functions include: a path planning module using an improved A* algorithm to plan the optimal driving path; a motion calculation module calculating the AGV's kinematics; receiving workpiece transfer task instructions from the AGV management Agent; and controlling the workpiece transfer process. Simultaneously, the AGV Agent promptly uploads its own state information, such as its position, speed, and task transfer status, to the AGV management Agent. The AGV management Agent provides the AGV Agent with node and road segment congestion information when optimizing the path.

[0032] (6) Node Agent

[0033] A node agent is a mapping of path network nodes in the system. Each node corresponds to a node agent, which is used to determine the probability of AGV conflict, provide conflict warnings, and calculate the congestion of the node. The node agent can sense the approach, entry, and departure of AGVs on the node, and update the current status data of the node, including the number of AGVs near the node, their location, and distance information. The node agent uses conflict prediction methods to predict AGV conflicts, uses node congestion calculation methods to update the congestion of the node, and transmits the current status data, conflict prediction information, and congestion information to the path network agent for aggregation and real-time monitoring.

[0034] (7) Road segment Agent

[0035] A segment agent is a mapping of road segments in the system's path network. Each road segment corresponds to a segment agent, which is used to detect AGV conflicts on the road segment and calculate the segment's congestion level. The segment agent can sense AGVs entering and leaving the road segment, update the current status data of the road segment, including the number, location, direction of travel, and distance information of AGVs on the road segment. The segment agent uses conflict prediction methods to predict AGV conflicts, uses road segment congestion calculation methods to update the road segment's congestion level, and transmits the current status data, conflict prediction information, and congestion level information to the path network agent for aggregation and real-time monitoring.

[0036] (8) Road Network Management Agent

[0037] The road network management agent maps the entire path network in the system, summarizing information reported by all nodes and road segments, including conflict node information, conflict road segment information, node congestion, and road segment congestion information, and updating status data in a timely manner. It interacts with the AGV management agent, processing conflict warning information in the road network using conflict resolution strategies and notifying the AGV management agent for further processing. On the other hand, it obtains and updates the AGV's position and speed information in real time from the AGV management agent, summarizing and monitoring it in real time.

[0038] Furthermore, the specific process of improving the A* algorithm is as follows:

[0039] Step 1: Create a grid map and convert the AGV path network into a grid network.

[0040] Step 2: Determine the coefficients of the heuristic function based on the starting point and the target point, and obtain the AGVUsed table of other AGV occupied nodes.

[0041] Step 3: Create an Open table and a Close table, and empty them. The Open table is used to store the points to be tested, and the Close table is used to store the points that have been tested. Put the starting node S into the Open table.

[0042] Step 4: Traverse the Open table, check all points in the table, calculate the f value for all points, find the point with the smallest f value, and set it as the current node CurNode.

[0043] Step 5: Remove the current node CurNode from the Open list and move it into the Close list. Then search for the adjacent node AdjNode of the current node CurNode.

[0044] Step 6: Check all adjacent nodes AdjNode of the current node CurNode one by one. For adjacent nodes not in the Close list, check if they are in the Open list. If they are, go to Step 8; otherwise, go to Step 7.

[0045] Step 7: Check if the adjacent point is in the AGVUsed table. If it is not in the AGVUsed table, go to Step 9. If it is in the AGVUsed table, go to Step 10 to continue execution.

[0046] Step 8: If the neighboring node AdjNode of the current node CurNode is in the Open table, calculate a new path for this neighboring node. That is, check if the g value of the path from the current node CurNode to the neighboring node AdjNode is smaller than the g value of the old path of the neighboring node AdjNode. If it is smaller, change the parent node of the neighboring node AdjNode to the current node CurNode and recalculate the g and f values.

[0047] Step 9: If it is not in the AGVUsed table, add it to the Open table, record the values ​​of f, g, and h of this adjacent node, set the current node CurNode as its parent node, and go to Step 10 to continue execution.

[0048] Step 10: Determine if there are any adjacent points that have not been checked. If yes, go to Step 6; otherwise, go to Step 11.

[0049] Step 11: Determine if the target node E has been found: Check if the newly added point in the Open table has the target node E. If the target node E is found, proceed to Step 12. If the target node E is not found, continue to the next step, Step 13.

[0050] Step 12: If the newly added point in the Open list contains the target node E, it means that the shortest path has been successfully found. Move the target node E into the Close list, and sequentially search for the parent node from the target node E to the starting node S, and store it as the path. This is the shortest path found, and the program ends.

[0051] Step 13: Check if the Open table is empty: if it is not empty, proceed to Step 4; if it is empty, the path search fails and the program ends.

[0052] Furthermore, AGV conflicts are classified into six types, and different prediction and elimination methods are used for different types of conflicts;

[0053] (1) First type of basic conflict

[0054] Conflict description: If two AGVs simultaneously travel from different road segments to the same node, a conflict may occur at the node. However, the next paths of the two AGVs do not conflict with each other, meaning that the next paths of the two AGVs do not overlap in either direction.

[0055] Conflict prediction method: AGV1 and AGV2 may collide at nodes. Conflict prediction is based on the following formula:

[0056]

[0057] Where TA1 is the estimated arrival time of AGV1 at the node, TA2 is the estimated arrival time of AGV2 at the node, D is the monitoring diameter of the node, and V A The AGV's operating speed is such that a conflict will occur if certain conditions are met.

[0058] Conflict resolution method: If a conflict is predicted, the AGV that arrives later at the node is controlled to slow down or stop and wait before the node, and then resume normal speed and pass through the node after the other AGV has completely passed through the node.

[0059] (2) Second type of fundamental conflict

[0060] Conflict Description: Two AGVs simultaneously travel from different road segments to the same node, and a conflict may occur at the node. The next paths of the two AGVs overlap in the forward direction. In addition to the potential conflict at the node, there may also be a potential overtaking conflict on the forward overlapping road segments.

[0061] Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is carried out in the same way as the first type of conflict, that is, the prediction is made by formula (3); for the chasing conflict that occurs on the road segment, the chasing conflict is predicted in the same way as the fourth type of conflict.

[0062] Conflict resolution methods: If a node conflict is predicted, the AGV that arrives later at the node will slow down or stop and wait before the node, and resume normal speed and pass through the node after the other AGV has completely passed through the node; If a chasing conflict is predicted on a road segment, the AGVs behind will slow down so that the distance between them is greater than the minimum safe distance.

[0063] (3) The third type of basic conflict

[0064] Conflict Description: Two AGVs simultaneously travel from different road segments to the same node, and a conflict may occur at the node. The next path of one AGV is the current path of the other AGV, which is a reverse overlapping road segment. In addition to the possible conflict at the node, there is a potential encounter conflict in the reverse overlapping road segment.

[0065] Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is performed in the same way as the first type of conflict, i.e., using formula (3) for prediction; for encounter conflicts on the road segment, the following formula is used for prediction:

[0066]

[0067] Conflict elimination method: If a conflict is predicted, whether it is a node conflict or a road segment encounter conflict, the AGV2 is directly controlled to slow down or stop and wait before reaching the node. After the AGV1 has completely passed the node, it resumes normal speed and passes through the node.

[0068] (4) The fourth type of basic conflict

[0069] Conflict Description: Two AGVs simultaneously travel from different road segments to the same node. The next path of each AGV is the current path of the other AGV. Since the road segments are in reverse overlap, there will inevitably be node conflicts or road segment encounter conflicts. Conflicts may occur at the node or encounter conflicts may occur in the reverse overlapping road segments, resulting in mutual locking.

[0070] Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is performed in the same way as the first type of conflict, that is, node conflict is predicted using equation (3); for encounter conflicts on the road segment, the following formula is used for prediction:

[0071]

[0072] Conflict elimination method: In this case, node conflict or road section meeting conflict will inevitably occur. Since the conflict cannot be avoided, it is considered to change the path, one of the AGVs is selected to change its path to avoid the other AGV, the conflict is converted into a third-type conflict, and conflict prediction and elimination processing are performed according to the third-type conflict; the principle for selecting the AGV to change the path is to select the AGV closer to the node or select the AGV with lower path change cost.

[0073] (5) The fifth type of basic conflict

[0074] Conflict description: Conflicts occurring between two AGVs on a road section are divided into three categories. The first category is the chasing conflict occurring on the same road section: the rear AGV has a higher speed than the front AGV, thus a chasing conflict will occur; the second category is the meeting conflict occurring on the same road section: two AGVs drive towards each other, thus a meeting conflict occurs; the third category is that for two AGVs, a chasing conflict or a meeting conflict exists on the next road section of the AGVs.

[0075] Conflict prediction method: For the first and second situations, it is necessary to monitor the separation distance between the two AGVs in real time. If the distance is less than the safe distance, a conflict will occur, which is expressed by the formula as follows:

[0076] Dis 12 <D safe +D res (6)

[0077] Wherein, Dis 12 is the distance between two AGVs, D safe is the minimum safe distance of the AGV, D res is the reserved distance value. For the third situation, when a meeting conflict occurs on the next road section of the two AGVs, the following formula is used for prediction:

[0078] TAS2<TAS1<TAE2 or TAS1<TAS2<TAE1 (7)

[0079] Wherein, TAS1 and TAE1 are the expected start and end time of AGV1 entering the conflict road section, TAS2 and TAE2 are the expected start and end time of AGV2 entering the conflict road section, if the condition is satisfied, a road section meeting conflict will occur.

[0080] Conflict resolution methods: For the first scenario, when a conflict is predicted, the following vehicle slows down or stops to maintain a distance not exceeding the preset safe distance; for the second scenario, when a conflict is predicted and it is impossible to take measures to avoid it in advance, the vehicle slows down and stops for manual handling; for the third scenario, when two AGVs meet and conflict on the next road segment, since there is an overlap of reverse road segments, it is necessary to consider changing the path to avoid the conflict. One AGV is selected to change its path to avoid the other AGV. The principle for selecting the AGV to change its path is to select the AGV that is closer to the node or the AGV with the lower cost of changing its path.

[0081] (6) The sixth type of basic conflict

[0082] Conflict Description: The sixth type of conflict is a conflict involving three AGVs. Node conflicts and road segment conflicts are mixed together, and there is a possibility of multiple consecutive conflicts. At a node, there may be conflicts between two or three AGVs, and on a road segment, there may also be encounter conflicts or chasing conflicts between two AGVs.

[0083] Conflict prediction method: The conflict between the three AGVs consists of the basic conflicts of type (1), type (2), type (3) and type (4) mentioned above. The conflict detection method is to perform conflict detection on every two of the three AGVs, and then perform conflict detection on AGV1 and AGV2, AGV1 and AGV3, and AGV2 and AGV3 in sequence. The type and number of conflicts can be obtained. The conflicts are sorted in sequence according to type (4), type (3), type (2) and type (1) conflicts, and the existing node conflicts and road segment conflicts are cleared.

[0084] Conflict resolution method: Conflict detection begins, and the conflict situation of the three AGVs is analyzed. Since the three AGVs occupy the three road segments of the node respectively, the remaining empty road segment is not occupied. In order to simplify the conflict situation of the three AGVs and quickly alleviate congestion and conflict, the solution to the conflict of the three AGVs is to prioritize ensuring that one AGV can pass through the empty road segment as soon as possible. That is, prioritize the conflict that can utilize the empty road segment. The remaining two AGVs can then be handled according to the conflict of types (1)-(4). After each conflict handling is completed, the next conflict detection and handling will begin. Conflict events will continue to be handled according to the actual operation of the AGVs.

[0085] The beneficial technical effects of this invention are as follows:

[0086] This invention integrates an improved A* algorithm, multi-agent modeling and simulation methods, road network congestion calculation, and conflict prediction and resolution strategies to construct a hierarchical hybrid multi-agent system. This system includes a task management agent, an AGV management agent, a road network management agent, an AGV agent, a node agent, and a road segment agent. Each agent possesses perception, decision-making, judgment, and interaction capabilities, and collaborates using knowledge-based methods for dynamic collision-free path planning. The improved A* path planning algorithm provides the shortest path for the AGV agent. Conflict prediction and resolution strategies based on six conflict scenarios provide prediction and elimination methods for dynamic real-time conflicts within the road network. The road network congestion calculation method updates the congestion status of the path network in real time, working in conjunction with the A* path planning algorithm to select the shortest collision-free path. This multi-agent simulation-based dynamic collision-free path planning method for AGVs enables the AGV logistics system to operate smoothly and efficiently without conflict or deadlock. Examples demonstrate the effectiveness of this method. Attached Figure Description

[0087] Figure 1 This is a schematic diagram of the dynamic collision-free path planning method for AGVS based on multi-agent simulation of the present invention.

[0088] Figure 2 A hierarchical hybrid multi-agent system architecture for AGVS dynamic collision-free path planning.

[0089] Figure 3 A schematic diagram of the AGV management agent function.

[0090] Figure 4 This is a schematic diagram of the AGV Agent function.

[0091] Figure 5 This is a schematic diagram of the node agent function.

[0092] Figure 6 This is a schematic diagram of the Agent function for a road segment.

[0093] Figure 7 This is a schematic diagram of the road network management agent function.

[0094] Figure 8 A flowchart illustrating the improvement of the A* algorithm.

[0095] Figure 9 This is a schematic diagram of the first type of basic conflict.

[0096] Figure 10 This is a schematic diagram of the second type of basic conflict.

[0097] Figure 11 This is a schematic diagram of the third type of basic conflict.

[0098] Figure 12 This is a diagram illustrating the fourth type of basic conflict.

[0099] Figure 13 This is a diagram illustrating the fifth type of basic conflict.

[0100] Figure 14 This is a diagram illustrating the sixth type of basic conflict.

[0101] Figure 15 This is the sixth type of basic conflict resolution process.

[0102] Figure 16 This describes the operational scenario for AGVS's dynamic collision-free path planning.

[0103] Figure 17 A simulation scenario for AGVS dynamic path planning.

[0104] Figure 18 Example of AGV transportation efficiency comparison (fixed transportation time).

[0105] Figure 19 This is a comparison of AGV collision avoidance performance (fixed transport time) for an example.

[0106] Figure 20 Example of AGV transportation efficiency comparison (fixed number of transported workpieces).

[0107] Figure 21 This example compares the collision avoidance performance of AGVs (with a fixed number of transported workpieces).

[0108] Figure 22 The collision avoidance performance is shown in the example with different numbers of AGVs (fixed number of transported workpieces).

[0109] Figure 23 This example demonstrates the transportation efficiency (fixed transportation time) under different numbers of AGVs. Detailed Implementation

[0110] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0111] The dynamic collision-free path planning problem for AGVS needs to consider the following: assigning tasks to AGVs, selecting the optimal AGV travel route, avoiding conflicts, collisions and deadlocks, and the ultimate goal is to complete the material transportation task in the shortest time.

[0112] This invention presents a dynamic collision-free path planning method for AGVS based on multi-agent simulation. It combines an improved A* algorithm with multi-agent simulation for dynamic collision-free path planning of AGVS. The improved A* algorithm is used to find the shortest collision-free path for a single AGV, and a multi-agent conflict prediction and elimination strategy is used to solve the dynamic real-time conflict problem of AGVs. For ease of description, the dynamic collision-free path planning method for AGVS based on multi-agent simulation in this paper is named CF-MAS (Conflict-Free Multi-Agent System).

[0113] The present invention provides a dynamic collision-free path planning method for AGVS based on multi-agent simulation, as follows: Figure 1 As shown, in order to solve the dynamic collision-free path planning problem of AGVS, a comprehensive AGVS multi-agent simulation system was established based on the Anylogic simulation software platform and combined with Java secondary development. It consists of AGVS logistics and road network model, AGVS agent elements, and AGVS knowledge methods.

[0114] First, the abstraction methods, expression mechanisms, and operational control of the AGVS logistics transportation system are analyzed. Object-oriented methods and graphical modeling and visualization techniques are used to model the AGVS logistics transportation system, constructing an AGVS logistics and road network model. This model includes logistics task management, material transportation process management, as well as AGV ontology models, path network models, node models, road segment models, AGV docking stations, various loading and unloading station models, and workpiece models. The model features reusability, flexibility, and scalability.

[0115] Secondly, based on the AGVS logistics and road network model, the logical relationships between various entities in AGVS and the operation mechanism of system events are analyzed. The types and functions of multi-agent systems are proposed, and a hybrid architecture multi-agent system (MAS) composed of various types of agents is constructed.

[0116] Hybrid multi-agent system (MAS) architecture, such as Figure 2As shown, the top layer consists of the workshop management agent, the middle layer consists of the task management agent, AGV management agent, and road network management agent, and the bottom layer consists of the agents corresponding to each element of the system, including workpiece agents, AGV agents, node agents, and road segment agents. The workshop management agent, as the highest manager of the system, possesses global system information and the highest authority. The bottom-level element agents are the executors or collaborators in completing transportation tasks, possessing a certain degree of autonomy and intelligence. The middle-layer management agent acts as the intermediary for communication between the upper and lower layers, not only conveying the decisions of the workshop management agent to the bottom layer but also collecting and summarizing information from the bottom-level agents and feeding it back to the top-level agent. The bottom-level agents are managed by the middle-layer agents and include multiple workpiece agents, multiple AGV agents, multiple node agents, and multiple road segment agents. The task management agent manages its group of workpiece agents, the AGV management agent manages its group of AGV agents, and the road network management agent is responsible for managing the group of node agents and the group of road segment agents within the path network. The workshop management agent can communicate with the task management agent, AGV management agent, and road network management agent. The task management agent and AGV management agent can communicate with each other, as can the AGV management agent and road network management agent. Each AGV agent can communicate with related workpiece agents, node agents, and road segment agents during dynamic operation. Utilizing this hierarchical hybrid multi-agent system architecture, the multi-agent system can manage the communication, data perception, and proactive decision-making functions of various agents during dynamic real-time operation, enabling all agents to form a unified and collaborative whole to complete the transportation task.

[0117] GVS's dynamic collision-free path planning, which employs a hierarchical hybrid multi-agent architecture, offers the following advantages:

[0118] (1) The hybrid multi-agent system architecture has the advantages of both distributed and centralized systems, and has the advantages of high system communication efficiency, good stability and strong autonomy.

[0119] (2) It overcomes the inherent defects of the distributed structure, namely, the large and complex information storage of Agents in the system, the frequent switching of system communication routes, the lack of global information, and the complexity of collaborative solution.

[0120] (3) It has strong adaptability and scalability. Whether adding transportation tasks, workpieces, AGVs, or changing road network nodes and road segments, the system can simply add or delete new Agents and can be configured quickly.

[0121] (4) Complex transportation tasks can be distributed to various agents, and local problems in the transportation process can be solved by a single agent. The solution is simple and can be computed in parallel, thereby reducing the complexity of the overall problem computation.

[0122] Definitions and functions of each agent in the hybrid architecture multi-agent system (MAS).

[0123] (1) Workshop Management Agent

[0124] The workshop management agent is the highest manager of the system, possessing global information and the highest authority of the entire AGV system, and serves as the system management agent. The workshop management agent communicates with the task management agent, AGV management agent, and road network management agent to coordinate the initiation and task execution processes of the intermediate management agents and collect and summarize overall transportation information.

[0125] (2) Task Management Agent

[0126] The task management agent manages all material / workpiece agents in the system, summarizing the status information of all workpieces in real time, including: workpiece number, AGV number carrying the workpiece, and location information, and updating the status data in a timely manner for aggregation and real-time monitoring. At the same time, the task management agent communicates and negotiates with the AGV management agent, sending workpiece transportation requests to the AGV management agent, negotiating the assignment and allocation between workpieces and AGVs, assigning AGVs to undertake transportation tasks, and binding the AGV agent with the workpiece agent being transported.

[0127] (3) Workpiece Agent

[0128] The workpiece agent is a mapping of transported materials / workpieces in the system. Each workpiece corresponds to one workpiece agent, which carries origin and destination information. Workpiece agents are uniformly managed by the task management agent. The task management agent and AGV management agent negotiate to determine the order of transport tasks and negotiate the allocation and assignment between workpieces and AGVs. Once the assignment is completed, the workpiece is bound to the AGV, and the AGV completes the subsequent transport task of the workpiece. The workpiece agent promptly uploads its own status information, including the AGV number carrying the workpiece and the workpiece's location information in the road network, to the task management agent for aggregation and real-time monitoring.

[0129] (4) AGV Management Agent

[0130] like Figure 3As shown, the AGV management agent manages all AGV agents in the system, summarizing the status information of all AGVs in real time, including AGV number, speed, no load, load, and fault information, and updating the status data promptly. As the central layer of the intermediate management layer, the AGV management agent communicates and interacts with the workpiece management agent and the road network management agent. It receives transportation requests from the workpiece management agent and assigns AGVs to undertake transportation tasks, binding the AGV agents to the workpiece agents being transported. Furthermore, the AGV management agent interacts with the road network management agent, receiving conflict resolution control commands from the road network management agent and transmitting them to the corresponding AGV agents for execution. Simultaneously, the AGV management agent sends the AGV's position and speed information to the road network management agent for aggregation and real-time monitoring.

[0131] (5) AGV Agent

[0132] like Figure 4 As shown, the AGV Agent is a mapping of AGVs in the system. Each AGV corresponds to one AGV Agent, which completes a series of specified transportation tasks, moving workpieces from the starting point to the destination through a path network. The AGV Agent is a hybrid intelligent agent structure, capable of both planning autonomous behavior and responding to events. The AGV Agent can perceive the external environment and interact with other Agents. It updates the current state data of the AGV, including its position, speed, empty load, loaded load, and transportation task information. The AGV Agent's own decision-making and calculation functions include: a path planning module using an improved A* algorithm to plan the optimal driving path; a motion calculation module calculating the AGV's kinematics; receiving workpiece transfer task instructions from the AGV management Agent; and completing the workpiece transfer process control. At the same time, the AGV Agent promptly uploads its own state information, such as the AGV's position, speed, and task transfer status information, to the AGV management Agent. The AGV management Agent provides the AGV Agent with node and road segment congestion information when optimizing the path.

[0133] (6) Node Agent

[0134] like Figure 5As shown, a node agent is a mapping of path network nodes in the system. Each node corresponds to a node agent, which is used to determine the probability of AGV conflict, provide conflict warnings, and calculate the congestion level of the node. The node agent can sense the approach, entry, and departure of AGVs on the node, and update the current status data of the node, including the number of AGVs near the node, their location, and distance information. The node agent uses conflict prediction methods to predict AGV conflicts, uses node congestion calculation methods to update the congestion level of the node, and transmits the current status data, conflict prediction information, and congestion information to the path network agent for aggregation and real-time monitoring.

[0135] (7) Road segment Agent

[0136] like Figure 6 As shown, the segment agent is a mapping of the path network segments in the system. Each segment corresponds to a segment agent, which is used to detect AGV conflicts on the segment and calculate the segment's congestion level. The segment agent can sense AGVs entering and leaving the segment and update the current status data of the segment, including the number, location, direction of travel, and distance information of AGVs on the segment. The segment agent uses conflict prediction methods to predict AGV conflicts, uses segment congestion calculation methods to update the segment's congestion level, and transmits the current status data, conflict prediction information, and congestion level information to the path network agent for aggregation and real-time monitoring.

[0137] (8) Road Network Management Agent

[0138] like Figure 7 As shown, the road network management agent is a mapping of the entire path network in the system. It is used to summarize the information reported by all nodes and road segments, including conflict node information, conflict road segment information, node congestion, road segment congestion information, and update the status data in a timely manner. It interacts with the AGV management agent. On the one hand, it processes the conflict warning information in the road network using conflict elimination strategies and notifies the AGV management agent for further processing. On the other hand, it obtains and updates the position and speed information of AGVs in real time from the AGV management agent, and summarizes and monitors them in real time.

[0139] An improved A* algorithm is used to solve the shortest path for a single AGV.

[0140] The improved A* algorithm's cost function f(n) is the sum of the actual cost and the heuristically estimated cost, i.e.:

[0141] f(n)=g(n)+h(n) (1)

[0142] Where n represents the current node, g(n) represents the actual cost of the AGV from the starting node to the current node, and h(n) represents the estimated cost of the AGV from the current node to the target node, i.e., the heuristic function.

[0143] Compared to introducing an acceleration factor λ into the traditional heuristic function to reduce the number of nodes traversed by the A* algorithm, considering the aspect ratio of the Manhattan distance and introducing a scaling factor α to coordinate the balance in the two-dimensional search direction, the heuristic function is:

[0144] h(n)=λ(|x n -x end |+α|y n -y end |) (2)

[0145] Among them, (x n ,y n (x) represents the x-coordinate and y-coordinate of the current point, respectively. end ,y end ) represent the x-coordinate and y-coordinate of the target point, respectively.

[0146] Improved A* algorithm process as follows Figure 8 As shown, specifically:

[0147] Step 1: Create a grid map and convert the AGV path network into a grid network.

[0148] Step 2: Determine the coefficients of the heuristic function based on the starting point and the target point, and obtain the AGVUsed table of other AGV occupied nodes.

[0149] Step 3: Create an Open table and a Close table, and empty them. The Open table is used to store the points to be tested, and the Close table is used to store the points that have been tested. Put the starting node S into the Open table.

[0150] Step 4: Traverse the Open table, check all points in the table, calculate the f value for all points, find the point with the smallest f value, and set it as the current node CurNode.

[0151] Step 5: Remove the current node CurNode from the Open list and move it into the Close list. Then search for the adjacent node AdjNode of the current node CurNode.

[0152] Step 6: Check all adjacent nodes AdjNode of the current node CurNode one by one. For adjacent nodes not in the Close list, check if they are in the Open list. If they are, go to Step 8; otherwise, go to Step 7.

[0153] Step 7: Check if the adjacent point is in the AGVUsed table. If it is not in the AGVUsed table, go to Step 9. If it is in the AGVUsed table, go to Step 10 to continue execution.

[0154] Step 8: If the neighboring node AdjNode of the current node CurNode is in the Open table, calculate a new path for this neighboring node. That is, check if the g value of the path from the current node CurNode to the neighboring node AdjNode is smaller than the g value of the old path of the neighboring node AdjNode. If it is smaller, change the parent node of the neighboring node AdjNode to the current node CurNode and recalculate the g and f values.

[0155] Step 9: If it is not in the AGVUsed table, add it to the Open table, record the values ​​of f, g, and h of this adjacent node, set the current node CurNode as its parent node, and go to Step 10 to continue execution.

[0156] Step 10: Determine if there are any adjacent points that have not been checked. If yes, go to Step 6; otherwise, go to Step 11.

[0157] Step 11: Determine if the target node E has been found: Check if the newly added point in the Open table has the target node E. If the target node E is found, proceed to Step 12. If the target node E is not found, continue to the next step, Step 13.

[0158] Step 12: If the newly added point in the Open list contains the target node E, it means that the shortest path has been successfully found. Move the target node E into the Close list, and sequentially search for the parent node from the target node E to the starting node S, and store it as the path. This is the shortest path found, and the program ends.

[0159] Step 13: Check if the Open table is empty: if it is not empty, proceed to Step 4; if it is empty, the path search fails and the program ends.

[0160] AGV Conflict Prediction and Handling Based on Multi-Agent System.

[0161] Conflict detection between AGVs is achieved through comprehensive prediction and judgment based on information exchange among AGV Agents, node Agents, road segment Agents, and road network management agents. Conflict elimination is achieved by the AGV management agent controlling the AGV Agents' normal driving, deceleration / stop to avoid obstacles, and path switching commands. Solutions for predicting and eliminating each type of conflict are provided, and the principle of conflict elimination is:

[0162] (1) Try to maintain the optimal path planning of the original AGV to minimize the task time of each AGV.

[0163] (2) Prioritize the AGVs that are closest to the node to pass through in order to improve passage efficiency.

[0164] (3) When three or more AGVs conflict, one AGV should be prioritized to pass through as soon as possible in order to reduce congestion at nodes and road sections and simplify the conflict situation.

[0165] AGV conflicts are classified into six types, and different prediction and elimination methods are used for different types of conflicts;

[0166] (1) First type of basic conflict

[0167] Conflict description: such as Figure 9 As shown, if two AGVs simultaneously travel from different road segments to the same node, a conflict may occur at the node. However, the next paths of the two AGVs do not conflict with each other, meaning that the next paths of the two AGVs do not overlap in either direction.

[0168] Conflict prediction method: AGV1 and AGV2 may collide at nodes. Conflict prediction is based on the following formula:

[0169]

[0170] Where TA1 is the estimated arrival time of AGV1 at the node, TA2 is the estimated arrival time of AGV2 at the node, D is the monitoring diameter of the node, and V A The AGV's operating speed is such that a conflict will occur if certain conditions are met.

[0171] Conflict resolution method: If a conflict is predicted, the AGV that arrives later at the node is controlled to slow down or stop and wait before the node, and then resume normal speed and pass through the node after the other AGV has completely passed through the node.

[0172] (2) Second type of fundamental conflict

[0173] Conflict description: such as Figure 10 As shown, two AGVs simultaneously travel from different road segments to the same node, where a conflict may occur. The next paths of the two AGVs overlap in the forward direction. In addition to the potential conflict at the node, there may also be a potential overtaking conflict on the overlapping road segments.

[0174] Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is carried out in the same way as the first type of conflict, that is, the prediction is made by formula (3); for the chasing conflict that occurs on the road segment, the chasing conflict is predicted in the same way as the fourth type of conflict.

[0175] Conflict resolution methods: If a node conflict is predicted, the AGV that arrives later at the node will slow down or stop and wait before the node, and resume normal speed and pass through the node after the other AGV has completely passed through the node; If a chasing conflict is predicted on a road segment, the AGVs behind will slow down so that the distance between them is greater than the minimum safe distance.

[0176] (3) The third type of basic conflict

[0177] Conflict description: such as Figure 11 As shown, two AGVs simultaneously travel from different road segments to the same node. Conflicts may occur at the node. The next path of one AGV is the current path of the other AGV, which is a reverse overlapping road segment. In addition to the possible conflict at the node, there is a potential encounter conflict in the reverse overlapping road segment.

[0178] Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is performed in the same way as the first type of conflict, i.e., using formula (3) for prediction; for encounter conflicts on the road segment, the following formula is used for prediction:

[0179]

[0180] Conflict elimination method: If a conflict is predicted, whether it is a node conflict or a road segment encounter conflict, the AGV2 is directly controlled to slow down or stop and wait before reaching the node. After the AGV1 has completely passed the node, it resumes normal speed and passes through the node.

[0181] (4) The fourth type of basic conflict

[0182] Conflict description: such as Figure 12 As shown, two AGVs simultaneously travel from different road segments to the same node. The next path of each AGV is the current path of the other AGV. Since the road segments are in reverse overlap, there will inevitably be node conflicts or road segment encounter conflicts. Conflicts may occur at the node or encounter conflicts may occur in the reverse overlap road segments, resulting in mutual locking.

[0183] Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is performed in the same way as the first type of conflict, that is, node conflict is predicted using equation (3); for encounter conflicts on the road segment, the following formula is used for prediction:

[0184]

[0185] Conflict elimination method: In this case, node conflicts or road section meeting conflicts will inevitably occur. Since conflicts cannot be avoided, it is considered to change the path, one of the AGVs is selected to change its path to avoid the other AGV, and the conflict is transformed into a third-type conflict, and conflict prediction and elimination processing are performed according to the third-type conflict; the principle for selecting an AGV to change the path is to select the AGV closer to the node or select the AGV with lower path change cost.

[0186] (5) The fifth type of basic conflict

[0187] Conflict description: as shown in Figure 13 , conflicts occurring between two AGVs on a road section are divided into three categories. The first category is the chasing conflict occurring on the same road section: the rear AGV is faster than the front AGV, and a chasing conflict will occur; the second category is the meeting conflict occurring on the same road section: two AGVs drive towards each other, and a meeting conflict occurs; the third category is that two AGVs have a chasing conflict or a meeting conflict on the next road section of the AGVs.

[0188] Conflict prediction method: for the first and second situations, it is necessary to monitor the interval distance between the two AGVs in real time. If the distance is less than the safety distance, a conflict will occur, which is expressed by the formula as follows:

[0189] Dis 12 <D safe +D res (6)

[0190] Wherein, Dis 12 is the distance between the two AGVs, D safe is the minimum safety distance of the AGV, and D res is the reserved distance value. For the third situation, a meeting conflict occurs on the next road section of the two AGVs, and the following formula is used for prediction:

[0191] TAS2<TAS1<TAE2 or TAS1<TAS2<TAE1 (7)

[0192] Wherein, TAS1 and TAE1 are the expected start and end time of AGV1 entering the conflict road section, TAS2 and TAE2 are the expected start and end time of AGV2 entering the conflict road section, if the condition is satisfied, a road section meeting conflict will occur.

[0193] Conflict resolution methods: For the first scenario, when a conflict is predicted, the following vehicle slows down or stops to maintain a distance not exceeding the preset safe distance; for the second scenario, when a conflict is predicted and it is impossible to take measures to avoid it in advance, the vehicle slows down and stops for manual handling; for the third scenario, when two AGVs meet and conflict on the next road segment, since there is an overlap of reverse road segments, it is necessary to consider changing the path to avoid the conflict. One AGV is selected to change its path to avoid the other AGV. The principle for selecting the AGV to change its path is to select the AGV that is closer to the node or the AGV with the lower cost of changing its path.

[0194] (6) The sixth type of basic conflict

[0195] Conflict description: such as Figure 14 As shown, the sixth type of conflict is a conflict between three AGVs, where node conflicts and road segment conflicts are mixed together, and there is a possibility of multiple consecutive conflicts. At the node, there may be a conflict between two or three AGVs, and on the road segment, there may also be a conflict between two AGVs meeting or chasing each other.

[0196] Conflict prediction method: The conflict between the three AGVs consists of the basic conflicts of type (1), type (2), type (3) and type (4) mentioned above. The conflict detection method is to perform conflict detection on every two of the three AGVs, and then perform conflict detection on AGV1 and AGV2, AGV1 and AGV3, and AGV2 and AGV3 in sequence. The type and number of conflicts can be obtained. The conflicts are sorted in sequence according to type (4), type (3), type (2) and type (1) conflicts, and the existing node conflicts and road segment conflicts are cleared.

[0197] Methods for resolving conflicts: such as Figure 15 As shown, conflict detection begins, and the conflict situation of the three AGVs is analyzed. Since the three AGVs occupy the three road segments of the node respectively, the remaining empty road segment is not occupied. In order to simplify the conflict situation of the three AGVs and quickly alleviate congestion and conflict, the solution to the conflict of the three AGVs is to prioritize ensuring that one AGV can pass through the empty road segment as soon as possible, that is, to prioritize the conflict that can utilize the empty road segment. The remaining two AGVs can then be handled according to the conflict of types (1)-(4). After each conflict handling is completed, the next conflict detection and handling begins, and the conflict events are continuously handled according to the actual operation of the AGVs.

[0198] Simulation Examples and Verification

[0199] To verify the correctness and practicality of the CF-MAS dynamic collision-free path planning method for AGVS based on multi-agent simulation, Anylogic multi-agent simulation software was used as the development environment. Secondary development was carried out using Java language. By combining multi-agent state diagrams and flowcharts, a hierarchical hybrid multi-agent architecture was constructed. An improved A* path algorithm and conflict prediction and handling mechanism were developed and implemented. The development of the AGVS multi-agent simulation prototype system was completed, and the results were verified using an example model.

[0200] The instance model is described as follows: Figure 16 The system is a grid-like AGVS path network, consisting of several nodes and connected paths. Various stations are distributed around the path network, including AGV docking stations H, workpiece starting stations A1, A2, A3, and A4, and workpiece destination stations B1, B2, B3, and B4. Workpiece transport tasks are generated as follows: four workpieces arrive at station A1 every 2 minutes, with destinations at B1, B2, B3, and B4 respectively. Stations A2, A3, and A4 generate workpieces and transport tasks in the same way as A1, ensuring the system has a saturated number of workpiece transport tasks. Workpieces are transported by AGVs from the starting station to the destination station via the path network. The initial position of the AGV is at the designated AGV docking station H. The initial number of AGVs is 4-10, the maximum speed of the AGV is 1 m / s, and the acceleration / deceleration is 1 m / s. 2 The AGV loading and unloading times are 5 seconds, the obstacle detection distance is 0.2 meters, and the unlocking time in case of AGV deadlock is 40 seconds. Each AGV can only carry one workpiece at a time and must complete the transport of the current workpiece before accepting the next task; it cannot stop midway.

[0201] Two simulation methods were used for comprehensive testing: first, a specific time range was set, and the number of workpieces transported and the number of AGV collisions and deadlocks were recorded during the simulation experiment; second, with a fixed number of workpieces to be transported, the transport time required to complete the task and the number of AGV collisions and deadlocks were recorded during the simulation experiment. These two methods were used to evaluate the transport efficiency and collision avoidance performance of the AGV system. To further test the effectiveness of the CF-MAS method presented in this paper, it was compared with the Ant-agent method. Furthermore, to analyze the impact of the parameter factors of the number of AGVs and unlocking time on system performance, simulation tests were conducted on the system with 4-10 AGVs respectively, and the results were analyzed.

[0202] The established simulation model is as follows: Figure 17 Show.

[0203] Simulation experiments were conducted on the CF-MAS method and the Ant-Agent method presented in this paper. First, a simulation scenario was run with a fixed transportation duration of 25 hours, recording the total number of collisions and the total number of transported workpieces. The AGV transportation efficiency was described using the number of workpieces transported per hour. Compared to the Ant-Agent method, the CF-MAS method improved transportation efficiency by an average of 5.27%. (See...) Figure 18 As shown.

[0204] The collision avoidance performance of the AGV system is described by the number of collisions per hour. Compared with the Ant-Agent method, CF-MAS reduces the average number of collisions per hour by 32.52%. (See...) Figure 19 As shown.

[0205] Then, a simulation scenario was run with a fixed number of 4000 transported workpieces, recording the number of collisions and the transport completion time. The transport time per workpiece was used to describe the AGV transport efficiency. Compared to the CF-MAS and Ant-Agent methods, the average transport time per workpiece was reduced by 4.88%. (See...) Figure 20 As shown.

[0206] The collision avoidance performance of the AGV system is described by the number of collisions per 100 workpieces. Compared with the Ant-Agent method, CF-MAS reduces the number of collisions per 100 workpieces by an average of 35.33%. (See...) Figure 21 As shown.

[0207] The above analysis shows that the CF-MAS method has better transportation efficiency and collision avoidance performance compared to the Ant-Agent method. During the operation phase of the manufacturing system, the configuration of various parameters of the AGV system is fixed, but the number of AGVs used can be increased or decreased according to the system's task requirements. To further analyze the impact of the number of AGVs on the system's transportation efficiency and collision avoidance performance in the CF-MAS method, simulation experiments were conducted for two scenarios: a fixed number of transported workpieces and a fixed transportation time, with AGV numbers of 4, 5, 6, 7, 8, 9, and 10. With a fixed number of transported workpieces of 4000, the system's collision avoidance performance was measured by the total number of collisions and the number of collisions per vehicle. Figure 22 As shown in the diagram, the analysis reveals that as the number of AGVs increases, the total number of collisions also increases, as does the number of collisions per vehicle. This indicates that increasing the number of AGVs leads to an increase in the number of system collisions, thereby reducing the system's collision avoidance performance.

[0208] With a fixed transport duration of 20 hours, the system transport efficiency is measured by the hourly transport volume and the hourly transport volume per vehicle for different numbers of AGVs. For example... Figure 23As shown in the diagram, the analysis reveals that while the hourly transport volume increases linearly with the number of AGVs, the hourly transport volume per vehicle decreases. This indicates that increasing the number of AGVs improves the system's transport efficiency, but the transport efficiency of a single AGV decreases. This is because as the number of AGVs increases, the average number of collisions per vehicle also increases. Since each collision and deadlock requires a certain amount of unlocking time, this reduces the transport efficiency of a single AGV.

[0209] The above case studies validate the CF-MAS method for dynamic collision-free path planning in AGVS based on multi-agent simulation, leading to the following conclusions:

[0210] (1) The CF-MAS method is a comprehensive simulation optimization method for dynamic path planning of AGVS. The CF-MAS method is used to construct AGVS simulation cases based on multi-agent modeling and simulation, and the improved A* algorithm is integrated to realize AGV conflict prediction and elimination strategies in six major categories.

[0211] (2) Simulation experiments were conducted for two cases: fixed transportation time and fixed number of transported workpieces. The results show that the CF-MAS method of the present invention has better transportation efficiency and collision avoidance performance than the Ant-Agent method.

[0212] This invention proposes a dynamic collision-free path planning method for AGVS based on multi-agent simulation. It integrates an improved A* algorithm, multi-agent modeling and simulation methods, road network congestion calculation, and conflict prediction and resolution strategies. A hierarchical hybrid multi-agent system is constructed, comprising a task management agent, an AGV management agent, a road network management agent, an AGV agent, a node agent, and a road segment agent. Each agent possesses perception, decision-making, judgment, and interaction capabilities, and collaborates using knowledge-based methods for dynamic collision-free path planning. The improved A* path planning algorithm provides the shortest path for the AGV agent. Conflict prediction and resolution strategies based on six conflict scenarios provide prediction and elimination methods for dynamic real-time conflicts within the road network. The road network congestion calculation method updates the congestion status of the path network in real time, working in conjunction with the A* path planning algorithm to select the shortest collision-free path. This multi-agent simulation-based dynamic collision-free path planning method for AGVS enables the AGV logistics system to operate smoothly and efficiently without conflict or deadlock. Examples demonstrate the effectiveness of this method.

Claims

1. A dynamic collision-free path planning method for AGVS based on multi-agent simulation, characterized in that, Specifically: A. Analyze the abstraction methods, expression mechanisms, and operation control of the AGVS logistics transportation system. Use object-oriented methods and graphical modeling and visualization technologies to model the AGVS logistics transportation system and construct an AGVS logistics and road network model, including logistics task management, material transportation process management, as well as AGV body model, path network model, node model, road segment model, AGV docking station, various loading and unloading station models, and workpiece model. B. Based on the AGVS logistics and road network model, analyze the logical relationships between various entities in AGVS, the operation mechanism of system events, propose the types and functions of multi-agent systems, and construct a hybrid architecture multi-agent system (MAS) composed of various types of agents. The hybrid architecture multi-agent system (MAS) has a top layer of workshop management agents, a middle layer of task management agents, AGV management agents, and road network management agents, and a bottom layer of agents corresponding to various system elements, including workpiece agents, AGV agents, node agents, and road segment agents. The workshop management agent, as the highest manager of the system, has global information and the highest authority. The bottom layer of various element agents are the executors or collaborators for completing transportation tasks. The middle layer management agents act as intermediaries for communication between the upper and lower layers, not only conveying the decisions of the workshop management agents to the bottom layer, but also collecting, summarizing, and feeding back the information from the bottom layer agents to the top layer agents. C. Solve the shortest path for a single AGV using the improved A* algorithm; The improved A* algorithm's cost function f(n) is the sum of the actual cost and the heuristically estimated cost, i.e.: f(n)=g(n)+h(n) (1) Where n represents the current node, g(n) represents the actual cost of the AGV from the starting node to the current node, and h(n) represents the estimated cost of the AGV from the current node to the target node, i.e., the heuristic function; Compared to introducing an acceleration factor λ into the traditional heuristic function to reduce the number of nodes traversed by the A* algorithm, considering the aspect ratio of the Manhattan distance and introducing a scaling factor α to coordinate the balance in the two-dimensional search direction, the heuristic function is: h(n)=λ(|x n -x end |+α|y n -y end |) (2) Among them, (x n ,y n (x) represents the x-coordinate and y-coordinate of the current point, respectively. end ,y end () represent the x-coordinate and y-coordinate of the target point, respectively; D. AGV conflict prediction and handling based on multi-agent systems; Conflict detection between AGVs is achieved through comprehensive prediction and judgment based on information exchange among AGV Agents, node Agents, road segment Agents, and road network management agents. Conflict elimination is achieved by the AGV management agent controlling the AGV Agents' normal driving, deceleration / stop to avoid obstacles, and path switching commands. Solutions for predicting and eliminating each type of conflict are provided, and the principle of conflict elimination is: (1) Try to maintain the optimal path planning of the original AGV to minimize the task time of each AGV. (2) Prioritize the AGVs that are closest to the node to pass through in order to improve passage efficiency; (3) When three or more AGVs conflict, one AGV should be prioritized to pass through as soon as possible in order to reduce congestion at nodes and road sections and simplify the conflict situation.

2. The AGVS dynamic collision-free path planning method based on multi-agent simulation according to claim 1, characterized in that, The definition and function of each agent in the hybrid architecture multi-agent system (MAS); (1) Workshop Management Agent The workshop management agent is the highest manager of the system, possessing global information and the highest authority of the entire AGV system, and serves as the system management agent. The workshop management agent communicates with the task management agent, AGV management agent, and road network management agent to coordinate the initiation and task execution processes of the intermediate management agents and collect and summarize overall transportation information. (2) Task Management Agent The task management agent manages all material / workpiece agents in the system, summarizing the status information of all workpieces in real time, including: workpiece number, AGV number carrying the workpiece, and location information, and updating the status data in a timely manner for summary and real-time monitoring; at the same time, the task management agent communicates and negotiates with the AGV management agent, sending workpiece transportation requests to the AGV management agent, negotiating the assignment and allocation between workpieces and AGVs, assigning AGVs to undertake transportation tasks, and binding the AGV intelligent agent with the workpiece agent being transported; (3) Workpiece Agent The workpiece Agent is a mapping of transported materials / workpieces in the system. Each workpiece corresponds to one workpiece Agent, which carries origin and destination information. Workpiece Agents are uniformly managed by the task management Agent. The task management Agent and AGV management Agent negotiate to determine the order of transport tasks and negotiate the allocation and assignment between workpieces and AGVs. Once the assignment is completed, the workpiece is bound to the AGV, and the AGV completes the subsequent transport task of the workpiece. The workpiece Agent promptly uploads its own status information, including the AGV number carrying the workpiece and the workpiece's location information in the road network, to the task management Agent for aggregation and real-time monitoring. (4) AGV Management Agent The AGV management agent manages all AGV agents in the system, summarizing real-time status information for all AGVs, including AGV number, speed, no-load status, load status, and fault information, and updating the status data promptly. As the central layer of the intermediate management system, the AGV management agent communicates and interacts with the workpiece management agent and the road network management agent. It receives transportation requests from the workpiece management agent, assigns AGVs to undertake transportation tasks, and binds the AGV agents to the workpiece agents being transported. Furthermore, the AGV management agent interacts with the road network management agent, receiving conflict resolution control commands from the road network management agent and transmitting them to the corresponding AGV agents for execution. Simultaneously, the AGV management agent sends the AGV's position and speed information to the road network management agent for aggregation and real-time monitoring. (5) AGV Agent An AGV Agent is a mapping of AGVs in the system. Each AGV corresponds to one AGV Agent, which completes a series of specified transportation tasks, moving workpieces from the starting point to the destination through a path network. The AGV Agent is a hybrid intelligent agent structure, capable of both planning autonomous behavior and responding to events. The AGV Agent can perceive the external environment and interact with other Agents; it updates the current state data of the AGV, including its position, speed, empty load, loaded status, and transportation task information. The AGV Agent's own decision-making and calculation functions include: a path planning module using an improved A* algorithm to plan the optimal driving path; a motion calculation module calculating the AGV's kinematics; receiving workpiece transfer task instructions from the AGV management Agent; and controlling the workpiece transfer process. Simultaneously, the AGV Agent promptly uploads its own state information, such as its position, speed, and task transfer status, to the AGV management Agent. The AGV management Agent provides the AGV Agent with node and road segment congestion information when optimizing the path. (6) Node Agent A node agent is a mapping of path network nodes in the system. Each node corresponds to a node agent, which is used to determine the probability of AGV conflict, provide conflict warnings, and calculate the congestion of the node. The node agent can sense the approach, entry, and departure of AGVs on the node, update the current status data of the node, including the number of AGVs near the node, their location, and distance information. The node agent uses conflict prediction methods to predict AGV conflicts, uses node congestion calculation methods to update the congestion of the node, and transmits the current status data, conflict prediction information, and congestion information to the path network agent for aggregation and real-time monitoring. (7) Road segment Agent A segment agent is a mapping of road segments in the system's path network. Each road segment corresponds to a segment agent, which is used to detect AGV conflicts on the road segment and calculate the congestion level of the road segment. The segment agent can sense AGVs entering and leaving the road segment, update the current status data of the road segment, including the number, location, direction of travel, and distance information of AGVs on the road segment. The segment agent uses conflict prediction methods to predict AGV conflicts, uses road segment congestion calculation methods to update the road segment congestion level, and transmits the current status data, conflict prediction information, and congestion level information to the path network agent for aggregation and real-time monitoring. (8) Road Network Management Agent The road network management agent maps the entire path network in the system, summarizing information reported by all nodes and road segments, including conflict node information, conflict road segment information, node congestion, and road segment congestion information, and updating status data in a timely manner. It interacts with the AGV management agent, processing conflict warning information in the road network using conflict resolution strategies and notifying the AGV management agent for further processing. On the other hand, it obtains and updates the AGV's position and speed information in real time from the AGV management agent, summarizing and monitoring it in real time.

3. The AGVS dynamic collision-free path planning method based on multi-agent simulation according to claim 1, characterized in that, The specific process of the improved A* algorithm is as follows: Step 1: Create a grid map and convert the AGV path network into a grid network; Step 2: Determine the coefficients of the heuristic function based on the starting point and the target point, and obtain the AGVUsed table of other AGV occupied nodes; Step 3: Create an Open table and a Close table, and empty them. The Open table is used to store the points to be tested, and the Close table is used to store the points that have been tested. Put the starting node S into the Open table. Step 4: Traverse the Open table, check all points in the table, calculate the f value for all points, find the point with the smallest f value, and set it as the current node CurNode; Step 5: Remove the current node CurNode from the Open list and move it into the Close list. Search for the adjacent node AdjNode of the current node CurNode. Step 6: Check all adjacent nodes AdjNode of the current node CurNode one by one. For adjacent nodes that are not in the Close list, check if they are in the Open list. If they are, go to Step 8; otherwise, go to Step 7. Step 7: Check if the adjacent point is in the AGVUsed list. If it is not in the AGVUsed list, go to Step 9. If it is in the AGVUsed list, go to Step 10 to continue execution. Step 8: If the neighboring node AdjNode of the current node CurNode is in the Open list, then calculate a new path for this neighboring node, that is, whether the g value of the path from the current node CurNode to the neighboring node AdjNode is smaller than the g value of the old path of the neighboring node AdjNode. If it is smaller, then change the parent node of the neighboring node AdjNode to the current node CurNode and recalculate the g value and f value. Step 9; If it is not in the AGVUsed table, add it to the Open table, record the values ​​of f, g, and h of this adjacent node, set the current node CurNode as its parent node, and go to Step 10 to continue execution; Step 10: Determine if there are still any adjacent points that have not been detected. If yes, go to Step 6; otherwise, go to Step 11. Step 11: Determine if the target node E has been found: Check if the newly added point in the Open table has the target node E. If the target node E is found, proceed to Step 12; otherwise, continue to the next step, Step 13. Step 12: If the newly added point in the Open list contains the target node E, it means that the shortest path has been successfully found. Move the target node E into the Close list, and search for the parent node from the target node E to the starting node S in sequence. Store the path as the path, which is the shortest path found. The program ends. Step 13: Check if the Open table is empty: If it is not empty, proceed to Step 4. If the value is empty, the path search fails and the program terminates.

4. The AGVS dynamic collision-free path planning method based on multi-agent simulation according to claim 1, characterized in that, The AGV conflicts are classified into six types, and different prediction and elimination methods are used for different types of conflicts; (1) First type of basic conflict Conflict description: If two AGVs simultaneously travel from different road segments to the same node, a conflict may occur at the node. However, the next paths of the two AGVs do not conflict with each other, that is, the next paths of the two AGVs do not overlap in the forward or reverse direction. Conflict prediction method: AGV1 and AGV2 may collide at nodes. Conflict prediction is based on the following formula: Where TA1 is the estimated arrival time of AGV1 at the node, TA2 is the estimated arrival time of AGV2 at the node, D is the monitoring diameter of the node, and V A The AGV's operating speed is such that a conflict will occur if certain conditions are met. Conflict elimination method: If a conflict is predicted, control the AGV that arrives at the node later to slow down or stop and wait before the node, and resume normal speed to pass through the node after the other AGV has completely passed through the node. (2) Second type of fundamental conflict Conflict Description: Two AGVs simultaneously travel from different road segments to the same node, and a conflict may occur at the node. The next paths of the two AGVs overlap in the forward direction. In addition to the possible conflict at the node, there may also be potential overtaking conflicts on the forward overlapping road segments. Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is carried out in the same way as the first type of conflict, that is, the prediction is made by formula (3); for chasing conflicts that occur on the road segment, the chasing conflict is predicted in the same way as the fourth type of conflict. Conflict resolution methods: If a node conflict is predicted, the AGV that arrives later at the node will slow down or stop and wait before the node, and resume normal speed and pass through the node after the other AGV has completely passed through the node; If a chasing conflict is predicted on the road segment, the AGVs behind will slow down so that the distance between them is greater than the minimum safe distance. (3) The third type of basic conflict Conflict Description: Two AGVs simultaneously travel from different road segments to the same node, and a conflict may occur at the node. The next path of one AGV is the current path of the other AGV, which is a reverse overlapping road segment. In addition to the possible conflict at the node, there is a potential encounter conflict in the reverse overlapping road segment. Conflict prediction method: AGV1 and AGV2 may conflict at the node. Conflict prediction is performed in the same way as the first type of conflict, i.e., using formula (3) for prediction; for encounter conflicts on the road segment, the following formula is used for prediction: Conflict elimination method: If a conflict is predicted, whether it is a node conflict or a road segment encounter conflict, directly control AGV2 to slow down or stop and wait before reaching the node, and resume normal speed to pass through the node after AGV1 has completely passed through the node. (4) The fourth type of basic conflict Conflict description: Two automated guided vehicles (AGVs) drive toward the same node simultaneously from different road sections, and the next path of each AGV is the current path of the other AGV, which constitutes a reverse overlapping road section. Node conflicts or encounter conflicts on the road section are inevitable, conflicts may occur at the node or encounter conflicts may occur on the reverse overlapping road section, resulting in mutual deadlock; Conflict prediction method: AGV1 and AGV2 may conflict at the node, and conflict prediction is performed in the same way as the first type of conflict, that is, formula (3) is used to predict node conflict; for encounter conflicts occurring on the road section, the following formula is used for prediction: Conflict elimination method: In this case, node conflict or road section encounter conflict is bound to occur. Since the conflict cannot be avoided, it is necessary to change the path, one of the AGVs is selected to change its path to avoid the other AGV, the conflict is converted into a third-type conflict, and conflict prediction and elimination processing are performed according to the third-type conflict; the principle for selecting an AGV to change path is to select the AGV closer to the node or the AGV with lower path change cost; (5) The fifth basic type of conflict Conflict description: Conflicts occurring between two AGVs on road sections, which are divided into three categories: the first category is catching-up conflict occurring on the same road section, the rear AGV is faster than the front AGV, and a catching-up conflict will occur; the second category is encounter conflict occurring on the same road section, two AGVs drive towards each other, and an encounter conflict occurs; the third category is that for two AGVs, a catching-up conflict or an encounter conflict exists on the next road section of the AGVs; Conflict prediction method: For the first and second situations, it is necessary to monitor the separation distance between the two AGVs in real time, a conflict will occur if the distance is less than the safety distance, which is expressed by the formula as follows: Dis 12 <D safe +D res (6) Among them, Dis 12 D is the distance between the two AGVs. safe D is the minimum safe distance for AGVs. res To reserve distance values; For the third situation, when an encounter conflict occurs on the next road section of the two AGVs, the following formula is used for prediction: TAS2<TAS1<TAE2 or TAS1<TAS2<TAE1 (7) Wherein, TAS1 and TAE1 are the estimated start and end times of AGV1 entering the conflict road section, TAS2 and TAE2 are the estimated start and end times of AGV2 entering the conflict road section, if the condition is satisfied, a road section encounter conflict will occur; Conflict elimination method: For the first situation, when a conflict is predicted, the rear AGV decelerates or stops to keep the spacing no less than the preset safety distance; for the second situation, when a conflict is predicted, no measures can be taken to avoid it in advance, and the vehicle decelerates and stops waiting for manual processing; for the third situation where two AGVs have an encounter conflict on the next road section, since there is overlapping reverse road section, it is necessary to consider changing the path to avoid the conflict, one of the AGVs is selected to change its path to avoid the other AGV, the principle for selecting an AGV to change path is to select the AGV closer to the node or the AGV with lower path change cost; (6) The sixth basic type of conflict Conflict Description: The sixth type of conflict is a conflict between three AGVs, where node conflicts and road segment conflicts are mixed together, and there is a possibility of multiple consecutive conflicts; at a node, there may be a conflict between two or three AGVs, and on a road segment, there may also be a conflict between two AGVs meeting or chasing each other. Conflict prediction method: The conflict between the three AGVs is composed of the basic conflicts of the above-mentioned (1), (2), (3) and (4) types. The conflict detection method is to perform conflict detection on every two of the three AGVs, and then perform conflict detection on AGV1 and AGV2, AGV1 and AGV3, and AGV2 and AGV3 in sequence. The type and number of conflicts can be obtained. The conflicts are sorted in sequence according to the (4), (3), (2) and (1) types of conflicts, and the existing node conflicts and road segment conflicts are cleared. Conflict resolution method: Conflict detection begins, and the conflict situation of the three AGVs is analyzed. Since the three AGVs occupy the three road segments of the node respectively, the remaining empty road segment is not occupied. In order to simplify the conflict situation of the three AGVs and quickly alleviate congestion and conflict, the solution to the conflict of the three AGVs is to prioritize ensuring that one AGV can pass through the empty road segment as soon as possible. That is, prioritize the conflict that can utilize the empty road segment. The remaining two AGVs can then be handled according to the conflict of types (1)-(4). After each conflict handling is completed, the next conflict detection and handling will begin. Conflict events will continue to be handled according to the actual operation of the AGVs.

Images