Multi-AGV continuous task path planning method and system based on improved ECBS algorithm

Abstract

The invention provides a multi-AGV continuous task path planning method and system based on an improved ECBS algorithm. According to the method, tasks of a plurality of AGVs are decomposed into sub-tasks, and path planning is carried out through an improved enhanced conflict decomposition search algorithm. According to the improved algorithm, an online learning strategy is adopted to dynamically estimate the cost of nodes, and the solution of path conflicts is optimized through a conflict priority mechanism and geometric symmetry. Furthermore, a heuristic weight map is constructed for calculating path dependency, thereby improving planning efficiency and path quality. The system has a real-time dynamic adjustment function, can adjust the AGV path according to the environment change in the task execution process, and adapts to the complex scheduling requirement in a multi-AGV continuous task scene. The task execution efficiency and the path planning stability of the multi-AGV system are improved, and the method is suitable for the fields of intelligent storage, unmanned driving, logistics transportation and the like.

Description

Technical Field

[0001] The present application relates to the technical field of automatic guided vehicles, and in particular to a multi-AGV continuous task path planning method and system based on an improved ECBS algorithm. Background Art

[0002] With the rapid development of the warehousing and logistics industry, automation technology has played a vital role in improving warehousing efficiency and reducing labor costs. Among them, AGV, as an important part of the intelligent warehousing and logistics system, has gradually been widely used in warehouse transportation. The main task of AGV is to automatically transport goods in the warehouse according to the predetermined path, and its efficient operation depends on accurate path planning technology. Therefore, the AGV path planning problem has become a focus of research and application in warehousing and logistics automation.

[0003] The core of AGV path planning technology is to find the optimal path from the starting point to the end point for the AGV, while ensuring that the path avoids obstacles and conflicts with other vehicles during execution, and ensures the efficient operation of the AGV. AGV path planning problems usually involve multiple goals, such as the shortest path, minimum time, minimum energy consumption, etc. These goals require the path planning algorithm to quickly and accurately generate driving routes in a dynamic and changing warehouse environment, while adapting to emergencies such as road congestion and equipment failure.

[0004] Traditional path planning methods are mainly based on graph search algorithms, such as the A* algorithm and the Dijkstra algorithm. These algorithms can search for the shortest path in a known environment and accelerate the search process through heuristic functions. The A* algorithm is one of the commonly used classic algorithms. By combining the path length with the heuristic value, it can quickly find a solution while ensuring the shortest path. However, the A* algorithm has a high computational complexity when dealing with dynamic environments or large-scale warehousing environments, and it is difficult to cope with application scenarios with high real-time requirements.

[0005] In recent years, intelligent optimization algorithms and machine learning techniques have gradually been introduced into AGV path planning. Heuristic and meta-heuristic algorithms such as genetic algorithms, particle swarm algorithms, and ant colony algorithms can search for global optimal solutions under complex and multi-objective conditions. Compared with traditional graph search algorithms, intelligent optimization algorithms have stronger global search capabilities, especially in solving path planning problems, they can avoid falling into local optimality and ensure the flexibility and adaptability of path planning. In addition, machine learning-based technologies (such as deep learning and reinforcement learning) are gradually being applied to autonomous navigation and path planning of AGVs. These technologies can plan reasonable paths for AGVs in dynamic and unknown environments by learning motion patterns in complex environments.

[0006] However, in practical applications, AGV path planning algorithms are roughly divided into traditional algorithms and population-based intelligent optimization algorithms. Traditional path planning methods, such as graph search algorithms (A* and Dijkstra), have the disadvantages of high computational complexity, low path search efficiency, especially in environments with many obstacles, and difficulty in coping with changes in dynamic environments. When obstacles or AGV positions change, the path needs to be recalculated; it is not possible to handle path conflicts and coordination problems in multi-AGV systems well. The computational complexity is high, especially in environments with many obstacles. The path search efficiency is low; it is difficult to cope with changes in dynamic environments, and the path needs to be recalculated when obstacles or AGV positions change. In complex environments, population-based intelligent optimization algorithms, such as reinforcement learning and neural networks, have certain advantages and can adapt to dynamic environments to a certain extent, but their training time is long, requiring a large amount of data and computing resources. At the same time, the learning process is unstable and easily falls into local optimality, making it difficult to ensure the global optimality of the path. In practical applications, AGV path planning technology faces many challenges. For example, the warehouse environment is highly dynamic and uncertain, and factors such as cargo accumulation and interference from other vehicles can cause path blockage, which in turn affects the driving efficiency of AGVs. To address these challenges, researchers introduced multi-vehicle collaborative scheduling and dynamic obstacle avoidance strategies, allowing multiple AGVs to work together in the same environment and dynamically adjust driving paths through communication mechanisms to avoid collisions. Summary of the invention

[0007] The present application provides a multi-AGV continuous task path planning method and system based on an improved ECBS algorithm, which can be used to solve the technical problem of high computational complexity of the inability to handle path conflicts and coordination problems in a multi-AGV system well.

[0008] A multi-AGV continuous task path planning method based on an improved ECBS algorithm, the method comprising the following steps:

[0009] Step 1: Flatten the task list of each AGV; before flattening, the task list is represented by a three-dimensional list, with the second dimension being the task and the third dimension being the target point; after flattening, the task list is represented by a two-dimensional list, with the second dimension being the target point; that is, all tasks of each AGV are regarded as an ordered list of a series of target points; finally, convert the task list of each AGV into a stack;

[0010] Step 2: Iterate all AGVs, pop a target point from the AGV task stack each time as the end point of the AGV in this path planning, and the starting point is the previous target point or the initial position; after the iteration is completed, a set of corresponding starting and ending lists are obtained;

[0011] Step 3: Preprocess the start and end points: For AGVs with the same end point, only one of the end points is kept unchanged, and the remaining end points are changed. The change method includes two forms: the first form is to change to the waypoint of the path from the start point to the end point corresponding to the end point; the second form is to change to the corresponding starting point; after the change is completed, the end point that cannot be reached in this planning due to the change is pushed back to the top of the task stack of the corresponding AGV to ensure that no end point is missed. Remove the start-end point pairs with the same start and end points from the start and end point list, because they do not need to be planned, they only need to wait (or occupy) the corresponding time step at this location, and the time step is related to the path planning result. Take the second floor of a certain scene one as an example. Since there are only two raw material upstairs areas, and they are connected to the workstations through only one path, and there are 20 AGVs in the raw material area, according to the task ratio, nearly 10 AGVs go to these two upstairs areas at the same time, and it is easy to cause a deadlock that is difficult to resolve on the only connecting path. Therefore, the second change strategy is more suitable. The list of start and end points after preprocessing is as follows Figure 3 As shown, the key is the AGV number (starting from 0), and there are 20+16 AGVs in the first and second floors of the scene, numbered from 0 to 35.

[0012] Step 4: Path planning; Path planning results are obtained based on ECBS; ECBS is a conflict-based planning algorithm that greatly reduces search events by searching for suboptimal solutions. It is suitable for path planning problems with large maps (over 100X100) or many agents (over 50). Since the current scenarios generally have the problem of large maps, and there are too many AGVs in scenario one, the ECBS algorithm is used in all scenarios. Some path planning results are shown in Figure 2. Figure 4 As shown, the coordinate values ​​are processed linearly; if the planning fails and there is an avoidance point, the deadlock release algorithm is called; if the planning fails and there is no avoidance point, the program ends in failure and re-enters step 1 for task planning.

[0013] Step 5: Path truncation: In order to connect to the next task in time and prevent the AGV that completes the task ahead of time from waiting unnecessarily, the path planning result needs to be truncated. Calculate the minimum length of the path list of all AGVs, that is, the time step required for the AGV to complete the task the fastest; to fully consider the loading and unloading time of the AGV at the target point (i.e., the workstation), support truncation from the minimum value or any time step in the future; if the time is ignored, truncate from the minimum value; the time step at the truncation is the time step of this path planning; finally, append the truncated path to the path list of the corresponding AGV.

[0014] Step 6: Task connection: Update the starting point and maintain the end point; after the path is truncate, for the AGV that has not reached the end point (because the path list was truncated from the way and did not reach the end point), push the end point back to the top of the corresponding task stack; for all AGVs, use the last position of the truncated path list as its new starting point.

[0015] Step 7: Repeat steps 2 to 6 to complete the path planning of all tasks and output the path list for each AGV.

[0016] Furthermore, the ECBS algorithm adopts a heuristic optimization strategy and uses a heuristic weight graph to globally optimize the paths of multiple AGVs to reduce path conflicts and time consumption during AGV task execution.

[0017] Furthermore, the method estimates the path costs of low-level nodes based on historical planning data and updates the cost function in the planning process in real time;

[0018] The cost function is as follows:

[0019]

[0020] Among them, p i (t) is the position of agent i at time t, w(p i (t)) is the weight of the bit (in most cases, it may be 1, indicating that the cost of each step is equal);

[0021] The conflict cost is defined as:

[0022] C conflict (C) = λ·|C|

[0023] Where C is the set of conflicts that occur, |C| represents the number of conflicts, and λ is an adjustment factor used to adjust the impact of the severity of the conflict on the total cost;

[0024] In ECBS, the total cost is the weighted sum of the path cost and the conflict cost; assuming that the path cost of each agent is C path (p i ), then the total cost of the multi-agent system is expressed as:

[0025]

[0026] Where N is the number of agents, C represents all conflicts, and C total is the total cost of the system.

[0027] Furthermore, conflicts between AGVs are divided into fatal conflicts, minor conflicts, and non-fatal conflicts according to the number of conflicts caused by AGVs. Fatal conflicts are resolved first to ensure task continuity and execution efficiency.

[0028] Furthermore, the method further includes: Step 8, if a new task is added or other AGV paths are changed, the path planning of the AGV is adjusted according to the real-time monitored environmental changes, including the following steps:

[0029] Step 81: Calculate the number of deadlocks or conflicts of all AGVs, and reorder the AGVs in descending order of the number of conflicts, so that the AGV with the largest number of conflicts goes to the avoidance point first;

[0030] Step 82: Iterate all re-sorted AGVs, for each AGV, estimate the distance to all avoidance points using Manhattan distance, and sort the avoidance points in ascending order of distance;

[0031] In the Manhattan distance formula, suppose there are two points (x 1 ,y 1 ) and (x 2 ,y 2 ), Manhattan distance d can be expressed as:

[0032] d=|x 1 -x 2 |+|y 1 -y 2 |

[0033] Here, |x 1 -x 2 | is the distance in the horizontal direction, |y 1 -y 2 | is the distance in the vertical direction.

[0034] Step 83: Utilize A * The algorithm iterates all avoidance points, A * The algorithm is a heuristic algorithm for graph search, which aims to find the shortest path from the starting point to the target point. It combines the greedy strategy and Dijkstra's algorithm to optimize the path selection by evaluating the cost of each node during the search process; * The algorithm uses the following cost function to evaluate the priority of each node:

[0035] f(n)=g(n)+h(n)

[0036] Among them: g(n) is the actual cost from the starting point to the current node n, that is, the path length; h(n) is the estimated cost from the current node n to the target node, that is, the heuristic function; the A* algorithm ensures that the path is optimal by continuously expanding the node with the smallest f(n) value until the target node is found. The path from the AGV to the avoidance point is solved by the above A* algorithm. At this time, the positions of other AGVs are set to inaccessible, which means that the AGV occupies the corresponding position; if the avoidance point is unreachable, continue to iterate until the first reachable avoidance point with the shortest distance is found; if all avoidance points are unreachable, return to step 82 to iterate the next AGV;

[0037] Step 84: Let the AGV of the current iteration in step 82 go to the avoidance point found in step 83, and the path has been solved; during this period, other AGVs wait in place;

[0038] Step 85: Try to plan the path of other AGVs. At this time, the avoidance point in step 83 is not passable (occupied); if the planning is successful, the deadlock is successfully released, and during this period, the AGV at the avoidance point will keep waiting; if the planning fails, return to step 83 to iterate the next avoidance point;

[0039] Step 86: After the deadlock is successfully resolved, continue to plan subsequent tasks based on the path planning algorithm; the AGV at the avoidance point pushes its endpoint to the top of the corresponding task stack (that is, the AGV does not complete the task in this planning, but leaves it to the next planning to complete automatically, avoiding the other AGVs from waiting for a long time); return the path planning result of this deadlock resolution and enter step 85 of the path planning to connect the task stage with subsequent tasks.

[0040] The present application also provides a multi-AGV continuous task path planning system based on an improved ECBS algorithm, the system comprising:

[0041] M1, task decomposition module, used to decompose the tasks of multiple AGVs into continuous subtask sequences;

[0042] M2, path planning module, plans the path for each AGV based on the improved ECBS algorithm;

[0043] M3, conflict detection and resolution module, used to detect and resolve path conflicts between multiple AGVs;

[0044] M4, dynamic adjustment module, is used to adjust the path of the AGV according to real-time environmental changes during task execution.

[0045] The present invention provides a multi-AGV continuous task path planning method and system based on an improved ECBS algorithm. An improved ECBS algorithm is proposed, which focuses on processing the path planning problem of multiple AGVs when performing continuous tasks. This method not only retains the high efficiency of the traditional ECBS algorithm in conflict detection and resolution, but also further improves the overall operating efficiency and robustness of the AGV by optimizing task allocation and dynamic obstacle avoidance strategies. In practical applications, this system can quickly generate reasonable paths, reduce task conflicts, and optimize vehicle scheduling when dealing with complex and dynamic environments in warehouses. At the same time, the system also combines a multi-vehicle collaborative mechanism to achieve energy consumption optimization and reasonable resource scheduling while ensuring efficient AGV driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flow chart of the AGV path planning method of the present invention;

[0047] Figure 2 A schematic diagram of conflict detection in the path planning process of the present invention;

[0048] Figure 3 A schematic diagram of a scenario 1 provided in an embodiment of the present application;

[0049] Figure 4 A flow chart of path planning provided for an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0051] The present invention regards the task list of each AGV as a stack structure, pops out the destination nodes in the task in turn, and uses each path node in the task list as a destination to stay, such as an operating station, a charging station, etc. For each AGV, firstly, an initial path to the corresponding destination is independently generated based on its initial conditions to ensure the optimality and safety of path planning, and detect conflicts in real time during the path planning process.

[0052] The generated initial paths are checked for conflicts to determine whether the paths of multiple AGVs collide at the same time point. If a path conflict is found, candidate solutions are generated through a binary branch structure. In each branch, constraints are imposed on the conflicting AGVs to avoid the conflict area and replan the path.

[0053] A focus search strategy based on suboptimal search is adopted, with suboptimal coefficients as constraints, to quickly generate candidate paths that meet the shortest path conditions and ensure that the path cost is within the allowable range. When the AGV completes the path of the current task node, it pops the next destination from the stack and repeats the above path planning steps until the task list is empty.

[0054] The present application is further described below in conjunction with two embodiments.

[0055] Example 1

[0056] In a multi-AGV task scenario, each AGV usually needs to complete multiple continuous transportation tasks in different time periods. The present invention first decomposes the task of each AGV into multiple subtasks in chronological order. Each subtask includes a starting position and a target position, representing the process that the AGV needs to move from the starting point to the target point.

[0057] After receiving the task information of multiple AGVs, the system first parses the task sequence of each AGV based on the task decomposition module and converts it into multiple subtasks, each of which corresponds to an independent path planning requirement.

[0058] Then, the system initiates the path planning process, including:

[0059] S1. Assign an initial path to each AGV based on the complexity of the task. The initial path is the shortest path for each task.

[0060] S2, initialize global constraints, such as avoiding collisions, safe distances between AGVs, etc.;

[0061] S3. Initialize the parameters of the improved ECBS algorithm, such as the weight of the heuristic function, the tolerance of the sub-optimal solution, etc.

[0062] By decomposing the tasks and planning the initial path, the system provides basic data input for subsequent conflict detection and resolution.

[0063] Example 2

[0064] In practical applications, the working environment of AGVs often changes dynamically. For example, a new transport task may be assigned to an AGV before the original task is completed. To cope with this dynamic environment, the present invention provides a real-time path adjustment mechanism. The steps are as follows:

[0065] S1. Dynamic access to tasks: When a new task enters the system, the system first checks the task status of all current AGVs. For AGVs that are close to completing their current tasks, the system will directly assign them a new task starting point and perform path planning. For AGVs that are currently executing, the system will determine whether their current path is affected by the new task and adjust the path as needed to avoid conflicts.

[0066] S2. Real-time path update: The improved ECBS algorithm supports real-time path updates. When a new conflict or environmental change is detected, the system will partially update the currently planned path without having to replan the paths of all AGVs from scratch. By adjusting the local planning of specific conflict areas, the system can quickly respond to dynamic changes in the environment and ensure that the AGV can continue to perform its tasks smoothly.

[0067] S3. Priority adjustment of AGV paths: The system adjusts path planning according to the urgency and priority of AGV tasks. For example, for important AGV tasks, the system can assign a higher priority path, and other AGVs need to adjust their paths without affecting important tasks. This priority adjustment mechanism enables multi-task, multi-AGV systems to operate efficiently even when facing urgent tasks.

[0068] The above-described embodiments of the present application do not constitute a limitation on the protection scope of the present application.

Claims

1. A multi-AGV continuous task path planning method based on an improved ECBS algorithm, characterized in that: The multi-AGV continuous task path planning method based on the improved ECBS algorithm comprises the following steps: Step 1: Flatten the task list of each AGV; before flattening, the task list is represented by a three-dimensional list, with the second dimension being the task and the third dimension being the target point; after flattening, the task list is represented by a two-dimensional list, with the second dimension being the target point; that is, all tasks of each AGV are regarded as an ordered list of a series of target points; finally, convert the task list of each AGV into a stack; Step 2: Iterate all AGVs, pop a target point from the AGV task stack each time as the end point of the AGV in this path planning, and the starting point is the previous target point or the initial position; after the iteration is completed, a set of corresponding starting and ending lists are obtained; Step 3: Preprocess the starting point and end point; Step 4: Path planning: Obtain the path planning result based on ECBS; the coordinate values ​​are processed linearly; if the planning fails and there is an avoidance point, call the deadlock release algorithm; if the planning fails and there is no avoidance point, the program ends with failure and re-enters step 1 for task planning; Step 5: Path truncation, calculate the minimum length of the path list of all AGVs, that is, the time step required for the AGV to complete the task the fastest; support truncation from the minimum value or any time step in the future; if the time length is ignored, truncate from the minimum value; the time step at the truncation is the time step of this path planning; finally, append the truncated path to the path list of the corresponding AGV. ; Step 6: Task connection, update the starting point and maintain the end point; Step 7: Repeat steps 2 to 6 to complete the path planning of all tasks and output the path list for each AGV.

2. The method according to claim 1, characterized in that The improved ECBS algorithm adopts a heuristic optimization strategy and uses a heuristic weight graph to globally optimize the paths of multiple AGVs.

3. The method according to claim 1, characterized in that The method estimates the path cost of low-level nodes based on historical planning data and updates the cost function in the planning process in real time; The cost function is as follows: Among them, p i (t) is the position of agent i at time t, w(p i (t)) is the weight of the bit; The conflict cost is defined as: C conflict (C ) =λ·|C| Where C is the set of conflicts that occur, |C| represents the number of conflicts, and λ is an adjustment factor used to adjust the impact of the severity of the conflict on the total cost; In ECBS, the total cost is the weighted sum of the path cost and the conflict cost; assuming that the path cost of each agent is C path (p i ), then the total cost of the multi-agent system is expressed as: Where N is the number of agents, C represents all conflicts, and C total is the total cost of the system.

4. The method according to claim 1, characterized in that: The method divides conflicts between AGVs into fatal conflicts, minor conflicts and non-fatal conflicts according to the number of conflicts caused by AGVs, and resolves fatal conflicts first.

5. The method according to claim 1, characterized in that The method further includes: step 8, if a new task is added or the path of other AGVs is changed, the path planning of the AGV is adjusted according to the environmental changes monitored in real time.

6. The method according to claim 5, characterized in that Step 8: If a new task is added or other AGV paths are changed, the AGV path planning is adjusted based on the real-time monitored environmental changes, including the following steps: Step 81: Calculate the number of deadlocks or conflicts of all AGVs, and reorder the AGVs in descending order of the number of conflicts; Step 82: Iterate all re-sorted AGVs, for each AGV, estimate the distance to all avoidance points using Manhattan distance, and sort the avoidance points in ascending order of distance; Step 83: Utilize A * The algorithm iterates over all avoidance points and optimizes the path selection by evaluating the cost of each node during the search process; A * The algorithm uses the following cost function to evaluate the priority of each node: f(n)=g(n)+h(n) Among them: g(n) is the actual cost from the starting point to the current node n, that is, the path length; h(n) is the estimated cost from the current node n to the target node, that is, the heuristic function; the A* algorithm continuously expands the node with the smallest f(n) value until the target node is found; the A* algorithm solves the path from the AGV to the avoidance point, and at this time sets the position of other AGVs to inaccessible, which means that the AGV occupies the corresponding position; if the avoidance point is unreachable, continue to iterate until the first reachable avoidance point with the shortest distance is found; if all avoidance points are unreachable, return to step 82 to iterate the next AGV; Step 84: Let the AGV of the current iteration in step 82 go to the avoidance point found in step 83, and the path has been solved; during this period, other AGVs wait in place; Step 85: Try to plan the path of other AGVs. At this time, the avoidance point in step 83 is not passable. If the planning is successful, the deadlock is successfully released, and during this period, the AGV at the avoidance point will keep waiting. If the planning fails, return to step 83 to iterate the next avoidance point. Step 86: After the deadlock is successfully resolved, continue to plan subsequent tasks based on the path planning algorithm; the AGV at the avoidance point pushes its end point to the top of the corresponding task stack; returns the path planning result of this deadlock resolution and enters step 85 of the path planning to connect the task stage and subsequent tasks.

7. A multi-AGV continuous task path planning system based on an improved ECBS algorithm, characterized in that: The system comprises: Task decomposition module, used to decompose the tasks of multiple AGVs into continuous subtask sequences; Path planning module, which plans the path for each AGV based on the improved ECBS algorithm; Conflict detection and resolution module, used to detect and resolve path conflicts between multiple AGVs; The dynamic adjustment module is used to adjust the path of the AGV according to real-time environmental changes during task execution.

Images