A multi-agent parallel scheduling method

Abstract

The application provides a kind of multi-agent parallel scheduling method.It relates to computer path planning and distributed control technical field, steps include: constructing the perception domain that completely surrounds target agent and the buffer domain outside side;In the process of planning promotion, read the space-time occupation information of the space region covered by perception domain;If there is pre-occupation, calculate whether there is space-time overlap in a certain time interval;If there is space-time overlap, determine the highest authority agent in the competition area;If target agent is determined as the highest authority agent, write or update the authority flag of target agent in the competition area, and continue to plan candidate trajectory node towards the competition area;If target agent is determined as non-highest authority agent, suspend the expansion of candidate trajectory node;When the perception domain of target agent passes along the planned trajectory and departs from the competition area after planning promotion, reset the authority flag in the competition area.

Description

Technical Field

[0001] This invention relates to the field of computer path planning and distributed control technology, and in particular to a multi-agent parallel scheduling method. Background Technology

[0002] Multi-agent path scheduling typically requires handling multiple tasks simultaneously within a shared space. Each agent must not only satisfy static obstacle constraints but also avoid entering the same spatial region at the same or adjacent times as other agents. As the number of tasks, the size of the space, and the amount of parallel planning increase, relying solely on global conflict checks after task completion can easily lead to numerous path backtracking and redundant planning.

[0003] In existing technical solutions, one type of method checks for trajectory conflicts and replans after planning is completed; another type of method reads the occupancy information of other agents using a fixed neighborhood during planning. The former detects conflicts relatively late, and the latter, if it only uses a detection range the same size as the safety buffer, can only react when approaching the conflict location, which can easily lead to multiple agents entering the competing area almost simultaneously and modifying or reading the shared state at the same time.

[0004] Furthermore, in high-concurrency scenarios, even if priorities are set for contested regions, read / write propagation latency or execution order differences may still exist. For example, a low-privilege agent may not read the permission flag just written by a high-privilege agent in a planning cycle, and may still continue to advance into the contested region. Therefore, a scheduling synchronization mechanism that allows for "missed reads this time can be corrected next time" is needed, and this mechanism cannot come at the expense of safe distance.

[0005] In summary, the existing technical solutions have the following drawbacks:

[0006] 1) Conflict perception is too late: If the perception range is the same as or close to the safety buffer range, the agent will only know about the competition when it reaches the vicinity of the buffer boundary, leaving insufficient margin for permission determination, pausing and replanning.

[0007] 2) Coupling of security constraints and contention scheduling: If only permission judgment is relied upon without an independent buffer domain, the brief asynchrony of permission information may directly affect security; if only the buffer domain is relied upon, conflict detection and rollback will frequently occur at close range.

[0008] 3) Shared flags may be temporarily out of sync: Concurrent agents may execute on different processing threads, processors or devices, and a single read cannot guarantee that the highest privilege flag that was just written will be observed.

[0009] 4) A low-privilege agent may miss a highest-privilege flag: If a low-privilege agent completes the read just before the flag is written, the current iteration may continue planning. Without a subsequent correction mechanism, this can lead to a push conflict in a contested region.

[0010] 5) Unclear timing of permission release: If the highest-privilege agent releases its permission immediately after passing through the buffer, its tail safe area may still overlap with the advance area of ​​other agents; if it does not release its permission for a long time, it will cause unnecessary waiting. Summary of the Invention

[0011] To address the technical problems existing in the prior art, the present invention aims to provide a multi-agent parallel scheduling method that can improve the foresight of concurrent scheduling by setting the perception domain to be greater than the buffer domain and discovering the competition relationship in advance before actually entering the safety conflict boundary.

[0012] To achieve the above-mentioned objectives, this invention provides a multi-agent parallel scheduling method, comprising the following steps:

[0013] Construct a buffer domain that completely surrounds the target agent and moves synchronously with the target agent's planning and progress, as well as a perception domain outside the buffer domain; the perception domain is greater than the distance corresponding to at least two preset step sizes of the target agent in both the positive and negative directions of each spatial dimension.

[0014] During the planning and implementation process, the spatiotemporal occupancy information of the spatial area covered by the sensing domain is read;

[0015] If the spatiotemporal occupancy information indicates that there is pre-occupancy, then calculate whether there is spatiotemporal overlap between the target agent's buffer domain and the buffer domains of other agents within a specific time interval.

[0016] If there is spatiotemporal overlap, the corresponding spatial region is identified as the competition region. The recorded permission flags in the competition region are read, and the permission level of the target agent itself is compared with the permission level associated with the recorded permission flags to determine the agent with the highest permission.

[0017] If the target agent is determined to be the agent with the highest authority, then write or update the authority flag in the competition area and continue to plan candidate trajectory nodes towards the competition area; otherwise, pause the expansion of candidate trajectory nodes towards the competition area.

[0018] Once the target agent's perception domain has crossed and left the competition area along the planned trajectory as the planning progresses, the permission flags within the competition area are reset.

[0019] According to one technical solution of the present invention, the spatiotemporal occupancy information includes a spatial location identifier of a corresponding spatial region and a specific time interval; the specific time interval is composed of the start time of the other intelligent agent's buffer domain entering the spatial region and the end time of leaving the spatial region;

[0020] The calculation of whether the buffer domain of the target agent overlaps with the buffer domains of other agents within the specific time interval includes:

[0021] When the target intelligent agent plans to advance towards the spatial region, it is estimated that the buffer zone of the target intelligent agent will pass through the expected time interval of the spatial region; if the expected time interval is compared with the specific time interval, it is determined that there is a spatiotemporal overlap.

[0022] According to one technical solution of the present invention, constructing a buffer domain that completely surrounds the target intelligent agent includes:

[0023] Obtain the preset geometric envelope size, maximum moving speed, and preset safety factor of the target intelligent agent;

[0024] Based on the product of the safety factor and the maximum moving speed, and in conjunction with the preset geometric envelope size, the boundary distance of the buffer domain of the target agent is calculated and determined.

[0025] According to one technical solution of the present invention, the calculation rule for the permission level includes:

[0026] The permission level of each agent is determined by weighting calculation based on at least one of the following: the preset task priority, the remaining path length cost, the estimated arrival time to the competition area, the task deadline, and the agent's unique ID.

[0027] The agent with the highest permission level is determined as the agent with the highest permission; and the permission flag contains or is associated with the permission level of the agent with the highest permission.

[0028] According to one technical solution of the present invention, when the target agent is determined to be an agent without the highest authority, pausing the expansion of candidate trajectory nodes towards the competition region includes:

[0029] At the current candidate trajectory node, plan an in-situ waiting node with a fixed spatial location and an increasing time step, and repeatedly read the permission flag of the competing area in subsequent planning cycles.

[0030] Once the permission flag within the competition area is detected to be reset, the permission levels of each agent are re-compared to determine the agent with the highest permission within the competition area; this continues until the target agent is determined to be the agent with the highest permission.

[0031] According to one technical solution of the present invention, when the target agent is determined to be an agent without the highest authority, pausing the expansion of candidate trajectory nodes towards the competition region includes:

[0032] Using the current candidate trajectory node of the target agent as the planning starting point, an alternative candidate trajectory that bypasses the competing region is replanned.

[0033] According to one technical solution of the present invention, writing or updating a permission flag in the contested area includes:

[0034] Obtain the spatial center coordinates, competition region identifier, or shared spatiotemporal unit of the competing region;

[0035] Write the permission flag into a shared storage structure that is bound to the spatial center coordinates, the competition area identifier, or the shared spatiotemporal unit.

[0036] According to one technical solution of the present invention, the planning of candidate trajectory nodes is performed by a preset planner; the preset planner includes at least one of Dijkstra planner, heuristic search planner, sampling-based planner, local path planner or global path planner.

[0037] This invention provides a multi-agent parallel scheduling method, which has the following beneficial effects:

[0038] 1. By setting the perception domain to be greater than the buffer domain, the competition relationship can be detected in advance before actually entering the security conflict boundary, thus improving the foresight of concurrent scheduling.

[0039] 2. Decouple permission scheduling from hard security constraints: Permission flags are used for ordering passage, and buffer domains are used for final security fallback, so that short-term synchronization delays will not directly translate into collision risks.

[0040] 3. It allows a read to fail to see the latest permission flag, and then corrects the error by repeating the read in the next iteration. Therefore, it reduces the reliance on global strong synchronization, global locks, or full synchronization barriers for all threads, making it more suitable for high-concurrency planning.

[0041] 4. By using the method of "releasing the highest-authority agent's perception domain only after it leaves the center of the flag", we can avoid prematurely releasing permissions and causing the tail safety area to still overlap with other agents.

[0042] 5. Agents without the highest authority only pause or make local adjustments near the contention area, without having to perform a global rollback on all tasks, which can reduce redundant planning and scheduling jitter. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0044] Figure 1 The diagram illustrates the cooperative structure of agents, buffer domains, and perception domains in a multi-agent parallel scheduling method according to an embodiment of the present invention; wherein 1 represents the perception domain, 2 represents the buffer domain, and the black dot represents the center of the agent. Detailed Implementation

[0045] The description of the embodiments in this specification should be taken in conjunction with the accompanying drawings, which should form part of the complete specification. In the drawings, the shape or thickness of the embodiments may be exaggerated and may be indicated in a simplified or convenient manner. Furthermore, parts of the various structures in the drawings will be described separately; it is worth noting that elements not shown in the figures or not described in words are in a form known to those skilled in the art.

[0046] The descriptions of the embodiments herein, including any references to directions and orientations, are for ease of description only and should not be construed as limiting the scope of the invention. The following description of preferred embodiments involves combinations of features, which may exist independently or in combination; the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims. Figure 1 As shown. Specific Implementation Method 1

[0048] This embodiment provides a multi-agent parallel scheduling method, the steps of which include:

[0049] Construct a buffer domain that completely surrounds the target agent and moves synchronously with the target agent's planning and progress, as well as a perception domain outside the buffer domain; the perception domain is greater than the distance corresponding to at least two preset step sizes of the target agent in both the positive and negative directions of each spatial dimension.

[0050] During the planning and implementation process, the spatiotemporal occupancy information of the spatial area covered by the sensing domain is read;

[0051] If the spatiotemporal occupancy information indicates that there is pre-occupancy, then calculate whether there is spatiotemporal overlap between the target agent's buffer domain and the buffer domains of other agents within a specific time interval.

[0052] If there is spatiotemporal overlap, the corresponding spatial region is identified as the competition region. The recorded permission flags in the competition region are read, and the permission level of the target agent itself is compared with the permission level associated with the recorded permission flags to determine the agent with the highest permission.

[0053] If the target agent is determined to be the agent with the highest authority, then write or update the authority flag in the competition area and continue to plan candidate trajectory nodes towards the competition area; otherwise, pause the expansion of candidate trajectory nodes towards the competition area.

[0054] Once the target agent's perception domain has crossed and left the competition area along the planned trajectory as the planning progresses, the permission flags within the competition area are reset.

[0055] Specifically, the overall idea of ​​the multi-agent parallel scheduling method in this embodiment is to divide "safety" and "competition coordination" in multi-agent scheduling into two levels:

[0056] The inner layer uses a buffer domain to enforce hard security constraints, while the outer layer uses a larger perceptual domain to pre-read the spatiotemporal occupancy and permission flags of the contention area. Compared to the buffer domain, the perceptual domain has at least two more discrete units in both the positive and negative directions of each spatial dimension, enabling the agent to complete the identification of the contention relationship, determination of permissions, and necessary waiting before entering the security conflict area.

[0057] The specific implementation steps include:

[0058] I. Construction of the perceptual domain and buffer domain:

[0059] S1, such as Figure 1 As shown, a buffer zone centered on the agent's current position, candidate next position, or path node is established.

[0060] S2. Construct a receptive field outside the buffer domain. For three-dimensional discrete coordinates (x, y, z), if the expansion distance (or boundary distance) of the buffer domain in the x, y, and z directions are Bx+, Bx-, By+, By-, Bz+, and Bz- respectively, then the expansion distance (or boundary distance) of the receptive field in the positive and negative directions of the corresponding coordinate axes are Px+, Px-, Py+, Py-, Pz+, and Pz-, and satisfy the following relationships: Px+≥Bx++2, Px-≥Bx-+2, Py+≥By++2, Py-≥By-+2, Pz+≥Bz++2, Pz-≥Bz-+2.

[0061] Here, "2" represents the preferred minimum number of discrete units, which can be replaced by a physical distance of not less than two basic step sizes in continuous space implementation. Alternatively, it can be dynamically determined based on the maximum single-cycle displacement and the worst-case synchronization delay, as long as the sensing domain has sufficient advance sensing margin relative to the buffer domain.

[0062] Furthermore, the perception domain is not limited to a cuboid; it can also be a sphere, an ellipsoid, a Manhattan distance neighborhood, a forward extension region related to the velocity direction, or an asymmetric region that is dynamically adjusted according to velocity, braking distance, and communication delay.

[0063] S3, the buffer domain is only responsible for judging security conflicts; the perception domain is responsible for reading spatiotemporal occupancy information, competition relationships, and permission flags in advance. Even if the permission information is not synchronized temporarily, the buffer domain still acts as the final security constraint to prevent mutual intrusion.

[0064] II. Permission flags and synchronization mechanism for competing regions:

[0065] S4. When a certain intelligent agent (target intelligent agent) enters the perception area of ​​a certain competitive area for the first time, read the spatiotemporal occupancy information in the perception area (the spatiotemporal occupancy information is used to indicate whether the corresponding spatial area has been pre-occupied by the buffer area of ​​other intelligent agents within a specific time interval) and determine whether there is occupancy by other intelligent agents.

[0066] S5. If there are no other intelligent agents within the perception domain, the current intelligent agent continues to proceed according to the original plan; if there are other intelligent agents, the permission flags are read for all locations in the current perception domain center index.

[0067] The permission flag can store agent ID, permission level, timestamp, version number, lease expiration time, or a combination thereof. It can be maintained by a central scheduler or by multiple agents through a distributed consensus mechanism.

[0068] S6. Determine the agent with the highest authority within the competing region based on the permission rules.

[0069] S7. After reading the permission flag of the highest-authority agent, other agents that do not have the highest authority will no longer advance into the contested area. Instead, they will enter a pause, wait, or partial replanning state and will repeatedly read the permission flag in subsequent planning cycles.

[0070] S8, the highest-authority agent continues execution or planning to traverse the contested area. When the highest-authority agent's perception domain has crossed, no longer covers, or has moved away from the center of the contested area corresponding to the authorization flag according to preset release conditions, the highest-authority agent, scheduler, or shared state maintenance module resets, releases, or updates the authorization flag. Other agents resume synchronously.

[0071] The reset, release, or update of the permission flag can be triggered by the perception domain leaving the center of the contention area corresponding to the permission flag, or by conditions such as distance threshold, time threshold, leaving the tail buffer domain, explicit completion message, or lease expiration.

[0072] Without altering the basic logic of contention awareness, permission determination, waiting / correction, permission release, and security buffer verification, the above steps can have some read order swapped or be executed in parallel.

[0073] The specific technical features are explained below:

[0074] 1. Perception Domain: A pre-perception spatial region constructed with the current agent or candidate planning node as the center, used to read in advance the spatiotemporal occupancy status and permission flags of other agents in the region.

[0075] 2. Buffer Zone: A safety constraint region constructed around the safe outline or trajectory nodes of an agent. When any other agent's protected territory enters this region, it is determined that there is a conflict risk, thus prohibiting the corresponding path from proceeding.

[0076] 3. Competing region: The region where the perceptual domains of two or more agents overlap in space and time.

[0077] 4. Permission Flag: Shared state information associated with a spatial center, discrete grid point, or equivalent region index in the competition region, used to record the agent with the highest scheduling authority and / or its permission level in the current competition region.

[0078] 5. The agent with the highest authority: The agent that obtains the highest scheduling authority within the same competitive area according to the pre-set or dynamically calculated authority rules.

[0079] 6. Spatiotemporal Occupancy Information: Information describing whether a certain spatial location, spatial voxel, or safe area has been occupied or reserved by an intelligent agent within a specific time period.

[0080] 7. First entry into the perception zone: The state in which a certain intelligent agent's planning and advancement process first causes the competing area to fall into its perception domain, which is used to trigger permission competition and flag writing / reading processes.

[0081] 8. Synchronous Recovery: When the perception domain of the highest-authority agent has crossed or no longer covers the center corresponding to the authorization flag, the authorization flag is reset or released, allowing other waiting agents to reread the state of the competing area and continue planning. Specific Implementation Method Two

[0083] In this embodiment, when the target agent fails to read the permission flag in the competitive area through its own perception domain during the current planning cycle, the target agent does not perform a write operation and continues to advance the current planning to generate a corresponding candidate trajectory node.

[0084] When the target agent reads the recorded permission flag in the competition area through its own perception domain in the next planning cycle, and determines that it is not the agent with the highest permission, it triggers the execution pause to expand the candidate trajectory node to the competition area.

[0085] Specifically, this implementation method addresses an exception to S7, namely, considering the existence of short-term data synchronization issues in a concurrent environment:

[0086] If a non-highest-authority agent fails to read the flag recently written by the highest-authority agent in a planning cycle, it is allowed to complete the current calculation; in the next planning cycle, it reads the flag again, and pauses upon reading it. Because the outer perception domain is larger than the inner buffer domain, and reserves at least two discrete units of space difference, there is still a correction window for a missed read of the authorization flag. Specific Implementation Method 3

[0088] In this embodiment, the spatiotemporal occupancy information includes the spatial location identifier of the corresponding spatial region and a specific time interval; the specific time interval is composed of the start time when the buffer domain of other intelligent agents enters the spatial region and the end time when they leave the spatial region;

[0089] Calculating whether the buffer domain of the target agent overlaps with the buffer domains of other agents within a specific time interval includes:

[0090] When the target intelligent agent plans to advance towards the spatial region, it is estimated that the target intelligent agent's domain will pass through the expected time interval of the spatial region. If the expected time interval is compared with the specific time interval, it is determined that there is a spatiotemporal overlap.

[0091] Specifically, the aforementioned spatiotemporal occupancy information can be organized in the manner of "spatial location - height / level - time interval", or it can be expressed using a four-dimensional grid. The data can be structured as a time interval table, bitmap, hash table, tree structure, or database table. Permission flags can be bound to spatial centers, contested region IDs, or shared spatiotemporal units. Specific Implementation Method Four

[0093] In this embodiment, constructing a buffer domain that completely surrounds the target agent includes:

[0094] Obtain the target agent's preset geometric envelope size, maximum movement speed, and preset safety factor;

[0095] Based on the product of the safety factor and the maximum moving speed, and combined with the preset geometric envelope size, the boundary distance of the buffer domain of the target intelligent agent is calculated and determined.

[0096] Specifically, this involves the process of creating the buffer domain in S1:

[0097] S1. Based on the agent's current position, candidate next positions, or path nodes, establish a buffer zone centered on that position. The buffer zone can be represented by a cuboid, cube, sphere, ellipsoid, grid neighborhood, or a discrete set matching the shape of the agent.

[0098]

[0099] Among them, v x v y v z k represents the agent's maximum (or average) movement speed in the x, y, and z components, respectively. x k y k z is the safety factor (all values ​​are greater than or equal to 1), and a, b, and c are the corresponding side lengths.

[0100] Furthermore, before each path expansion, trajectory submission, or execution advancement, a spatiotemporal occupancy check is performed on the buffer domain corresponding to the candidate position. If any protected position within the buffer domain has been occupied or reserved by another agent during the target time period, the candidate position is unavailable, and the current agent waits or selects another candidate path. Thus, by using the perception domain to read the spatiotemporal occupancy status of other agents and the permission flags of the competing areas in advance, the scheduling problem of "who passes through the competing area first" is solved; and by using the buffer domain to independently perform security conflict determination, the security problem of "no collision regardless of whether the permission flags are temporarily out of sync" is solved. Detailed Implementation Method Five

[0102] In this embodiment, the rules for calculating permission levels include:

[0103] The authority level of each agent is determined by weighting at least one of the following factors: the preset task priority, the remaining path length cost, the estimated arrival time to the competition area, the task deadline, and the agent's unique ID.

[0104] The agent with the highest permission level is determined to be the agent with the highest permission; and the permission flag contains or is associated with the permission level of the agent with the highest permission.

[0105] Specifically, this involves the implementation of the permission rules in step S6.

[0106] Permission rules can be based on task priority, order of entry into the perception domain (first-come, first-served), remaining path cost (remaining path cost takes precedence), task deadline (urgent tasks take precedence, deadline takes precedence), agent ID, random backoff, or a combination of the above factors. The agent with the highest authority writes or updates its permission flag at the center of the corresponding contested area.

[0107] Taking the grid sequence where the drone is located as an example, the permission rules can be scored using the total number of rows (Rows) and columns (Cols) of the grid sequence, i.e., row*Cols+col. Here, row and col are the row number and column number of the drone, respectively. Specific Implementation Method Six

[0109] In this embodiment, when the target agent is determined to be an agent without the highest authority, the expansion of candidate trajectory nodes towards the competition area is paused, including:

[0110] At the current candidate trajectory node, plan a waiting node with an unchanged spatial location and an increasing time step, and repeatedly read the permission flag of the competing area in subsequent planning cycles.

[0111] Once the permission flag within the contested area is detected to have been reset, the permission levels of each agent are re-compared to determine the agent with the highest permission within the contested area; this continues until the target agent is determined to be the agent with the highest permission. Detailed Implementation Method Seven

[0113] In this embodiment, when the target agent is determined to be an agent without the highest authority, the expansion of candidate trajectory nodes towards the competition area is paused, including:

[0114] Starting from the current candidate trajectory node of the target agent, a new alternative candidate trajectory is planned to bypass the competing area.

[0115] Specifically, when an agent without the highest authority reads the flag, it can choose to completely pause, reduce speed, wait in place, partially detour, or replan its route; these methods are all equivalent implementations of "avoiding further entry into the competitive area". Detailed Implementation Method Eight

[0117] In this embodiment, writing or updating permission flags in the contested region includes:

[0118] Obtain the spatial center coordinates of the competing region, the identifier of the competing region, or the shared spatiotemporal unit;

[0119] Write the permission flag to the shared storage structure that is bound to the spatial center coordinates, the competition area identifier, or the shared spatiotemporal unit.

[0120] Specifically, in multi-threaded, multi-process, CPU / GPU heterogeneous, or distributed implementations, permission flags can be implemented through atomic variables, lock-protected shared memory, version numbers, message broadcasting, database transactions, or consistent storage. It does not require that the latest write be observed immediately upon a single read; instead, eventual synchronization and a safety fallback are achieved through perceptual domain redundancy distance and subsequent iterative repeated reads. Detailed Implementation Method Nine

[0122] In this embodiment, the planning of candidate trajectory nodes is performed by a preset planner; the preset planner includes at least one of Dijkstra planner, heuristic search planner, sampling-based planner, local path planner, or global path planner.

[0123] Specifically, this invention can embed Dijkstra's algorithm, heuristic search, sampling planning, path planning, or other local / global planners. For each candidate node, buffer domain security constraints are checked first, and then the perceptual domain contention information is read; when contention exists, a permission synchronization process is executed.

[0124] Table 1: The complete timing process of concurrent scheduling and collision avoidance arbitration when agent A and agent B execute the multi-agent parallel scheduling method of the present invention in a two-dimensional grid scenario.

[0125]

[0126] As shown in Table 1, the planned trajectory of agent A proceeds longitudinally from north to south, starting from coordinates (6,10) and the target passes through the intersection area (10,10); its grid number corresponds to 6×20+10=130 to 10×20+10=210.

[0127] Agent B plans its trajectory to move horizontally from west to east, starting from coordinates (10,6), and its target also extends to the intersection area (10,10); its grid number starts at 10×20+6=206.

[0128] The intersection region (10,10) is a potential spatiotemporal conflict point. When the perceptual domains of both parties overlap in space or time, they constitute a competing region.

[0129] Detailed execution flow of timing scheduling:

[0130] At time t0: Agent A is located at (6,10) (index 130), and Agent B is located at (10,6) (index 206). The perception domains of both agents, moving with the planning nodes, first detect spatiotemporal overlap at the intersection of grid lines, triggering the permission arbitration process. According to the preset permission rules, Agent A has a higher permission level than Agent B. The scheduling mechanism determines Agent A as the agent with the highest permission and writes / registers Agent A's permission flag in the shared state of the competing region.

[0131] At time t1: Agent A, as the highest-authority agent, continues to advance and generates the next trajectory node (7,10) (number 150). Agent B completes its existing inference step size in the current planning cycle and advances one step to coordinates (10,7) (number 207). At this time, Agent B's perceptual domain detects that the competing region has been written with the authorization flag by Agent A, and confirms that it is not the agent with the highest authority.

[0132] At time t2: Agent A continues to advance to (8,10) (sequence number 170) according to the original plan. After Agent B recognizes that it does not have the highest authority, it stops expanding new spatial location nodes to the competing grid (10,10) and enters the in-place waiting mechanism: the spatial coordinates remain unchanged at (10,7) (sequence number 207), only the time step increases, and it continues to monitor the status of the authority flag in the competing area.

[0133] At time t3: Agent A advances to (9,10) (order number 190), and its buffer zone safely passes through the predetermined area. Agent B remains in a paused state, with its spatial coordinates locked at (10,7) (order number 207), without changing its grid order, and continues to read the permission flag.

[0134] At time t4: Agent A reaches the critical intersection (10,10) (sequence number 210). The permission flag remains held by Agent A. Since Agent B is at (10,7) at time t1, the two are always spatially separated by 3 discrete steps, completely within the absorption gap of the outer perception domain, ensuring that there is no risk of physical conflict between Agent B's inner hard buffer domain and Agent A's envelope.

[0135] Subsequent release and recovery mechanism: After t4, when agent A's planned trajectory crosses (10,10) and its perceptual domain completely leaves the contested area, the permission flag is automatically reset (released). Agent B, which is in the waiting state, rereads the perceptual domain state and re-arbitrates. After determining that it has obtained the highest permission and the buffer domain security check passes, it releases the waiting state and resumes planning and progressing towards (10,8), (10,9), and (10,10).

[0136] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0137] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0138] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] It should also be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0140] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A multi-agent parallel scheduling method, characterized in that the steps are as follows: include: Construct a buffer zone that completely surrounds the target intelligent agent and moves synchronously with the planning and advancement of the target intelligent agent, as well as a perception zone outside the buffer zone; The perceptual domain is greater than the distance corresponding to at least two target agents' preset step sizes in both the positive and negative directions of each spatial dimension; During the planning and implementation process, the spatiotemporal occupancy information of the spatial area covered by the sensing domain is read; If the spatiotemporal occupancy information indicates that there is pre-occupancy, then calculate whether there is spatiotemporal overlap between the target agent's buffer domain and the buffer domains of other agents within a specific time interval. If there is spatiotemporal overlap, the corresponding spatial region is identified as the competition region. The recorded permission flags in the competition region are read, and the permission level of the target agent itself is compared with the permission level associated with the recorded permission flags to determine the agent with the highest permission. If the target agent is determined to be the agent with the highest authority, then write or update the authority flag in the competition area, and continue to plan candidate trajectory nodes towards the competition area. Otherwise, suspend the expansion of candidate trajectory nodes towards the competitive region; Once the target agent's perception domain has crossed and left the competition area along the planned trajectory as the planning progresses, the permission flags within the competition area are reset.

2. The multi-agent parallel scheduling method according to claim 1, characterized in that, When the target agent fails to read the permission flag in the competition area through its own perception domain during the current planning cycle, the target agent does not perform a write operation and continues to advance the current planning to generate a corresponding candidate trajectory node. When the target agent reads the recorded permission flag in the competitive region through its own perception domain in the next planning cycle, and determines that it is not the agent with the highest permission, it triggers the execution of the pause to expand the candidate trajectory node towards the competitive region.

3. The multi-agent parallel scheduling method according to claim 1, characterized in that, The spatiotemporal occupancy information includes the spatial location identifier of the corresponding spatial region and the specific time interval; the specific time interval is composed of the start time when the buffer domain of the other intelligent agent enters the spatial region and the end time when it leaves the spatial region; The calculation of whether the buffer domain of the target agent overlaps with the buffer domains of other agents within a specific time interval includes: When the target intelligent agent plans to advance towards the spatial region, the expected time interval for the target intelligent agent's buffer zone to traverse the spatial region is calculated. If the expected time interval is compared with the specific time interval, and there is a temporal overlap, it is determined that there is a spatiotemporal overlap.

4. The multi-agent parallel scheduling method according to claim 1, characterized in that, The construction of the buffer domain that completely surrounds the target agent includes: Obtain the preset geometric envelope size, maximum moving speed, and preset safety factor of the target intelligent agent; Based on the product of the safety factor and the maximum moving speed, and in conjunction with the preset geometric envelope size, the boundary distance of the buffer domain of the target agent is calculated and determined.

5. The multi-agent parallel scheduling method according to claim 1, characterized in that, The rules for calculating the permission levels include: The permission level of each agent is determined by weighting calculation based on at least one of the following: the preset task priority, the remaining path length cost, the estimated arrival time to the competition area, the task deadline, and the agent's unique ID. The agent with the highest permission level is determined as the agent with the highest permission; and the permission flag contains or is associated with the permission level of the agent with the highest permission.

6. The multi-agent parallel scheduling method according to claim 1, characterized in that, When the target agent is determined to be an agent without the highest authority, the expansion of candidate trajectory nodes towards the competition region is paused, including: At the current candidate trajectory node, plan an in-situ waiting node with a fixed spatial location and an increasing time step, and repeatedly read the permission flag of the competing area in subsequent planning cycles. Once the permission flag within the competition area is detected to be reset, the permission levels of each agent are re-compared to determine the agent with the highest permission within the competition area; this continues until the target agent is determined to be the agent with the highest permission.

7. The multi-agent parallel scheduling method according to claim 1, characterized in that, When the target agent is determined to be an agent without the highest authority, the expansion of candidate trajectory nodes towards the competition region is paused, including: Using the current candidate trajectory node of the target agent as the planning starting point, an alternative candidate trajectory that bypasses the competing region is replanned.

8. The multi-agent parallel scheduling method according to claim 1, characterized in that, Writing or updating permission flags in the contested region includes: Obtain the spatial center coordinates, competition region identifier, or shared spatiotemporal unit of the competing region; Write the permission flag into a shared storage structure that is bound to the spatial center coordinates, the competition area identifier, or the shared spatiotemporal unit.

9. The multi-agent parallel scheduling method according to claim 1, characterized in that, The planning of candidate trajectory nodes is executed through a preset planner; the preset planner includes at least one of Dijkstra planner, heuristic search planner, sampling-based planner, local path planner, or global path planner.

Images