Human-robot collaborative warehouse multi-agent path finding method based on intention estimation

Abstract

This invention discloses a human-machine collaborative warehousing multi-agent path planning method based on intent estimation, comprising: constructing a global topology graph of the warehousing environment; obtaining the current state information of operators and multiple agents in the warehousing environment to generate a current observation vector; continuously probabilistically inferring the operator's work target intent based on the operator's historical movement trajectory and current observation vector; inferring the nodes or edges of the global topology graph of the warehousing environment that the operator may occupy within a preset time range based on the intent estimation results; determining the set of agents affected by conflicts during the multi-agent path finding process; performing path replanning for agents affected by conflicts in a local subgraph, iteratively generating conflict-free local path schemes and merging them into the global path. This invention achieves efficient conflict handling and real-time replanning in dynamic environments, taking into account the safety, near-optimal nature, and computational efficiency of path planning, and is suitable for large-scale human-machine collaborative warehousing scenarios.

Description

Technical Field

[0001] This invention belongs to the field of intelligent warehousing and mobile robot technology, specifically involving a human-machine collaborative warehousing multi-agent path finding method based on intent estimation. It involves technologies such as human-machine collaboration, behavioral intent modeling, and multi-agent path planning, and is suitable for safe and efficient collaborative path planning and scheduling control of multiple mobile robots in warehousing operation environments where humans and machines work together. Background Technology

[0002] With the rapid development of e-commerce and intelligent logistics, warehousing systems are evolving towards high automation and intelligence. Automated warehousing systems based on mobile robots have been widely applied to picking, handling, and replenishment operations, improving overall operational efficiency and warehousing throughput through multi-robot collaborative work. However, in actual warehousing environments, mobile robots typically need to work collaboratively with human operators such as pickers and maintenance personnel in the same workspace, forming a typical human-robot co-operation scenario. Because humans possess autonomous decision-making capabilities, their movement behavior is uncertain and unpredictable, making traditional multi-agent pathfinding methods, which only address robot-to-robot paths, difficult to directly apply to human-robot collaborative warehousing environments.

[0003] In existing technologies, multi-agent pathfinding methods mainly focus on resolving path conflicts between robots and optimizing overall efficiency, lacking effective modeling of human behavioral characteristics. In human-robot co-operation scenarios, existing methods typically treat humans as dynamic obstacles and handle them through simple obstacle avoidance or safe distance constraints. This fails to reflect the true operational intentions and future movement trends of humans, easily leading to frequent replanning of path planning results, decreased efficiency, and even human-robot interaction safety risks. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems existing in the prior art.

[0005] To address this issue, the present invention provides a human-machine collaborative warehousing multi-agent path planning method based on intent estimation, in order to solve the technical problems of frequent path conflicts and low replanning efficiency in large-scale human-machine collaborative warehousing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The first aspect of this invention provides a human-machine collaborative warehousing multi-agent path finding method based on intent estimation, comprising:

[0008] The warehouse operation space is modeled as a graph structure consisting of multiple traversable nodes and connecting edges of intelligent agents, resulting in a global topology graph of the warehouse environment.

[0009] The current state information of workers and multiple intelligent agents in the warehouse environment is obtained to determine the correlation between workers and nodes in the global topology graph of the warehouse environment, and to generate a current observation vector to characterize the movement trend of workers relative to each target node. The state information of workers includes at least the posture information of workers.

[0010] Based on the historical movement trajectory and current observation vector of the operator, the operator's intention to perform the operation is continuously inferred probabilistically to obtain the operator's intention estimation result;

[0011] Based on the intent estimation results, the nodes or edges of the global topology of the warehouse environment that the operators may occupy within a preset time range are inferred, and the global path finding results are calculated online.

[0012] During the multi-agent pathfinding process, when a path conflict risk is detected between agents or between an agent and a worker, the set of agents affected by the conflict is determined.

[0013] For the set of agents affected by the conflict, path replanning is performed within a limited local subgraph, and local path schemes that do not conflict with the paths of agents not affected by the conflict are generated.

[0014] The local path scheme is updated to the global path search result to obtain a multi-agent path search result that satisfies the human-machine collaboration constraint.

[0015] In some embodiments, determining the correlation between operators and nodes in the global topology map of the warehouse environment, and generating a current observation vector characterizing the movement trend of operators relative to each target node, includes:

[0016] Calculate the correlation between the current location of the operator and each node in the global topology map of the warehouse environment. And thereby construct the association vector correlation The calculation formula is:

[0017]

[0018] in, This is a Gaussian function used to measure the proximity of the worker to node m. , This represents the total number of nodes in the global topology graph of the storage environment. Let V be the variance of the Gaussian function. For workers and nodes m The Euclidean distance between them; For directional consistency function, Let m be the angle between the operator's posture and the direction of node m;

[0019] Let the number of target nodes in the current warehouse operation space be . Then define the isolation matrix. for:

[0020]

[0021] in, For isolation matrix The element in the i-th row and j-th column is used to separate information related to the target node;

[0022] Associate vector With relative distance matrix Multiplying the target nodes and separating them yields the modulation range vector. : The relative distance matrix The dimension is , The Middle line, number Column elements For nodes and nodes The pixel distance between the optimal paths;

[0023] Calculate the candidate correlation vectors for all possible locations that the operator can reach after moving the same distance from the last observation position. And obtain the corresponding candidate modulation distance vector. : ;

[0024] modulated distance vector With all alternative modulation distance vectors The matrix formed Perform element-level comparison to generate the current observation vector. :

[0025]

[0026] in, For matrix The element in the m-th row and j-th column represents the alternative modulation distance value corresponding to the j-th target node when the operator moves the same distance from the historical observation position to the m-th alternative position.

[0027] In some embodiments, when continuously inferring the probabilistic intentions of workers regarding their work objectives, a lifetime intention estimation model based on a Hidden Markov Model is used, and a state transition matrix T is constructed, wherein...

[0028] The Hidden Markov Model (HMM) sets up g+3 hidden states, covering all possible intent scenarios of the operator, including the intent to go to the j-th target node. The intention to reach the target node and complete the task. Intentions are uncertain Irrational behavior or failure to register target nodes ; This represents the target number of nodes in the current warehouse operating space.

[0029] The constructed state transition matrix T is as follows:

[0030]

[0031] In the formula, The probability of a change in the target decision. The probability of initiating a new target. The probability of achieving the transition to the target. The probability of hesitation after achieving the goal. The probability of irrational decision-making being disrupted. This represents the probability of irrational recovery.

[0032] In some embodiments, the lifetime intention estimation model estimates long-term intentions based on the worker's historical movement trajectory and current observation vector, and infers the worker's short-term movement direction and target location through the long-term intention estimation.

[0033] In some embodiments, the preset time range is set as the forward look-ahead time distance L of the operator, and the setting process includes:

[0034] If the operator's intention is clear, the look-ahead time distance L is set to a fixed value;

[0035] If the operator's intention is unclear, then the multi-intention compatible behavior features of the operator's historical behavior are extracted, and the look-ahead time distance L is determined by the duration 'a' of the multi-intention compatible behavior. The multi-intention compatible behavior refers to the operator's action simultaneously pointing to multiple target nodes. The duration 'a' of the multi-intention compatible behavior is the time interval from the occurrence of the behavior to the operator's explicit intention to favor a certain target node. Then the look-ahead time distance L satisfies: , and These are the minimum and maximum durations of all multi-intent-compatible behaviors, respectively. This is the buffer time.

[0036] In some embodiments, determining the set of agents affected by the conflict includes:

[0037] Traverse the set of conflicts between workers and agents in the current plan. ,in, This refers to intelligent agents that conflict with the current operational personnel in the planning stage; The point in time of conflict; The vertex of the path conflict refers to the point in time of the conflict. The location of path conflicts among multiple agents will be determined by including the conflicting vertices in the path. intelligent agents Directly include the set of agents affected by the conflict ;

[0038] For those not directly included in the conflict set C For each agent, the original path length and the shortest path length around the conflict vertex are calculated. If the relative difference between the two exceeds a preset threshold, the agent is determined to be an indirectly affected agent and is included in the system. .

[0039] In some embodiments, the local subgraph is a storage environment subgraph constructed around the conflict vertices, and its nodes and edges are subsets of the global topology of the storage environment;

[0040] When replanning paths for agents affected by conflicts in the local subgraph, the path and position information of unaffected agents are kept unchanged. A rolling time window mechanism is adopted to generate a set of candidate vertices for agents affected by conflicts in each planning cycle based on the latest operator status information and intent estimation results. The optimal vertex is selected through collision filtering and cost priority strategy, and conflict-free local path schemes are generated iteratively.

[0041] In some embodiments, the planning period of the rolling time window mechanism is taken as the preset time range.

[0042] A second aspect of the present invention provides a human-machine collaborative warehousing multi-agent pathfinding device based on intent estimation, comprising:

[0043] The topology graph construction module is used to model the warehouse operation space as a graph structure composed of multiple intelligent agent traversable nodes and connecting edges, thereby obtaining a global topology graph of the warehouse environment.

[0044] The state acquisition module is used to acquire the current state information of the operators and multiple intelligent agents in the warehouse environment, so as to determine the correlation between the operators and the nodes of the global topology graph of the warehouse environment, and generate the current observation vector to characterize the movement trend of the operators relative to each target node. The state information of the operators includes at least the posture information of the operators.

[0045] The intent estimation module is used to continuously infer the operator's work target intent based on the operator's historical movement trajectory and current observation vector, and obtain the operator's intent estimation result.

[0046] The conflict detection module is used to infer, based on the intent estimation results, the nodes or edges of the global topology of the warehouse environment that the operator may occupy within a preset time range, and to calculate the global path finding results online.

[0047] The impact analysis module is used to determine the set of agents affected by path conflict when a risk of path conflict is detected between agents or between an agent and a worker during the multi-agent pathfinding process.

[0048] The local path planning module is used to perform path replanning only for the set of agents affected by the conflict within a limited local subgraph, and generate local path schemes that do not conflict with the paths of agents not affected by the conflict.

[0049] The path fusion module is used to update the local path scheme into the global path search result to obtain a multi-agent path search result that satisfies the human-machine collaboration constraint.

[0050] The third aspect of the present invention provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, are used to implement the human-machine collaborative warehousing multi-agent path finding method based on intent estimation according to any embodiment of the first aspect of the present invention.

[0051] The present invention has the following beneficial effects:

[0052] 1) This invention constructs a lifetime intention estimation framework through a Hidden Markov Model (HMM), which can accurately infer the long-term and short-term movement intentions of workers, enabling robots to avoid potential conflicts in advance and significantly improving the safety of human-robot collaborative operations.

[0053] 2) The LaCAM*-ImpD collaborative architecture is adopted. The affected agents are accurately screened through the affected agent detection algorithm, which limits the scope of replanning to local conflict areas, avoids the surge in computational overhead caused by global replanning, and improves the replanning efficiency in large-scale scenarios.

[0054] 3) The candidate vertex set design takes into account both global path reuse and local dynamic adjustment, and the cost-first strategy ensures the near-optimal nature of the path, achieving a balance between real-time performance, security and optimality.

[0055] 4) This method is applicable to large-scale warehousing scenarios with more than 100 intelligent agents. It can flexibly adapt to dynamic environmental changes and the uncertainty of worker intentions, and has broad application prospects. Attached Figure Description

[0056] Figure 1 This is an overall flowchart of a human-machine collaborative warehousing multi-agent path finding method based on intent estimation, provided in the first aspect of the present invention.

[0057] Figure 2 This is a schematic diagram of the state transition of the Hidden Markov Intent Estimation Algorithm used in the method of this embodiment of the invention;

[0058] Figure 3 This is a schematic diagram of the structure of an electronic device provided in a third aspect embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the application will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0060] Conversely, this application covers any alternatives, modifications, equivalent methods, and schemes made within the spirit and scope of this application as defined by the claims. Furthermore, to provide the public with a better understanding of this application, certain specific details are described in detail below. However, this application can be fully understood by those skilled in the art even without these detailed descriptions.

[0061] See Figure 1 The present invention provides a human-machine collaborative warehousing multi-agent path finding method based on intent estimation, comprising the following steps:

[0062] Step S1. Warehouse environment modeling: Model the warehouse operation space as a graph structure consisting of multiple traversable nodes and connecting edges of warehouse agents to obtain the global topology graph of the warehouse environment;

[0063] Step S2. Verification of worker movement: Obtain the current state information of workers and multiple warehouse agents in the global topology map of the warehouse environment, so as to judge the correlation between workers and nodes in the global topology map of the warehouse environment, and generate a current observation vector to represent the movement trend of workers relative to each target node. The state information of workers includes at least the posture information of workers.

[0064] Step S3. Worker Intent Estimation: Based on the worker's historical movement trajectory and current observation vector, continuously perform probabilistic inference on the worker's work target intent to obtain the worker's intent estimation result;

[0065] Step S4. Global Path Planning: Based on the intent estimation results obtained in Step S3, infer the nodes or edges of the global topology of the warehouse environment that the operators may occupy within the preset time range, and calculate the global path finding results online.

[0066] Step S5. Screening of affected agents: During the multi-agent pathfinding process, when a risk of path conflict is detected between warehouse agents or between warehouse agents and operators, the set of warehouse agents affected by the conflict is determined.

[0067] Step S6. Local Path Generation: For warehouse robots within the set of warehouse robots affected by the conflict mentioned above, path replanning is performed within a limited local subgraph, and a local path scheme that does not conflict with the path of the unaffected warehouse robot is generated.

[0068] Step S7. Path Fusion: Update the above local path scheme to the global path search result obtained in step S4 to obtain the multi-agent path search result that satisfies the human-machine collaboration constraint.

[0069] In some embodiments, step S1 specifically includes:

[0070] Constructing a global graph model of the warehousing environment ,in This represents the set of vertices corresponding to all traversable locations of the agent. This represents the set of edges that allow direct movement between vertices (i.e., connectivity between locations). A generalized Voronoi diagram (GVD) is constructed based on the global graph model G of the warehouse environment as the global topology graph of the warehouse environment. This divides the warehouse operation space into non-overlapping walkable areas, ensuring clear boundaries for the movement space of agents and workers. A D* dynamic path planning algorithm is employed to pre-calculate the pixel distances of the optimal paths between all vertices, storing the path length information as a relative distance matrix. This provides foundational data support for subsequent intent estimation and pathfinding, i.e. The dimension is , The Middle line, number Column elements For nodes and nodes The pixel distance of the optimal path between them. , This represents the total number of nodes in the global topology graph of the warehouse environment.

[0071] In some embodiments, step S2 specifically includes:

[0072] Step S21: Initialize the initial path from the warehouse agent to the target point according to the LaCAM* algorithm; specifically, adopt the hierarchical search architecture of the LaCAM* algorithm, construct a low-dimensional space constraint tree in the high-level decision-making through cost pruning and redundant configuration filtering, reduce the global search range, reduce computational overhead, and at the same time retain the near-optimality of the global path.

[0073] Step S22. Using AR devices and / or visual sensors and other sensing devices, collect the position coordinates and orientation information (such as direction of travel) of the workers in the global topology map of the warehouse environment in real time to form the current status information of the workers; in addition, collect the current status information of multiple warehouse intelligent agents in the global topology map of the warehouse environment in real time. Optionally, the status information of the warehouse intelligent agents includes at least the posture information of the warehouse intelligent agents themselves.

[0074] Step S23. Verify the movement of the workers based on the collected current information of the workers and the warehouse intelligent agent, specifically including:

[0075] Step S231. Calculate the correlation between the current position of the operator and each node in the GVD. And thereby construct the association vector (Its elements are the correlation between the operators and each node) ), correlation The calculation formula is:

[0076]

[0077] in, This is a Gaussian function used to measure the proximity of the worker to node m. , The variance of the Gaussian function is obtained through experimental calibration. For directional consistency function, Let m be the angle between the operator's posture and the direction of node m. The smaller the angle, the better. The larger the value, the higher the consistency between the worker's movement direction and the direction of node m; conversely, the smaller the value, the larger the angle. The smaller the value, the lower the consistency between the worker's movement direction and the direction of node m.

[0078] Step S232. Let the number of target nodes (belonging to vertices) in the current warehouse operation space be... , Then define the isolation matrix. for:

[0079]

[0080] in, For isolation matrix The element in the i-th row and j-th column is used to separate information related to the target node. , .

[0081] Step S233. Associativity vector With relative distance matrix Multiplying the target nodes and separating them yields the modulation range vector. Its vector dimension is related to the number of target nodes. Consistent. Modulated distance vector The calculation formula is:

[0082]

[0083] Step S234. Considering the continuity of the worker's movement, calculate the candidate correlation vectors corresponding to all possible locations the worker may reach after moving the same distance from the last observed position. (Its elements represent the correlation between the possible locations of the workers and each node.) ), and obtain the corresponding candidate modulation distance vector. :

[0084]

[0085] modulated distance vector With all alternative modulation distance vectors The matrix formed Perform element-level alignment and generate the current observation vector using the following formula. :

[0086]

[0087] in, For matrix The element in the m-th row and j-th column represents the alternative modulation distance value corresponding to the j-th target node when the operator moves the same distance from the historical observation position to the m-th alternative position (corresponding to the m-th node in GVD). When the operator moves from node m to target node j, the current observation vector... Corresponding element Approaching 1; when the operator moves away from node m towards the target node j, the current observation vector... Corresponding element Approaching 0.

[0088] It is understandable that by analyzing the matrix With modulation distance vector Element-level comparison and normalization are performed to generate an observation vector v. This observation vector converts the changes in modulation distance between the operator and each target node at different alternative locations into a trend quantization value between 0 and 1. The closer the value is to 1, the more significantly the distance between the worker and the target node j decreases when the worker moves from the current observation position to the m-th alternative position, and the movement trend is closer to that target node; conversely, the closer the value is to 1, the more significant the distance is to 1. The closer the result is to 0, the greater the correlation distance, and the further the movement trend moves away from the target node. This quantitative result provides key observational features for subsequent intention estimation based on Hidden Markov Models (HMMs), supporting continuous and accurate inference of the movement intentions of workers.

[0089] In some embodiments, such as Figure 2 As shown, step S3 uses a lifetime intention estimation model built based on a Hidden Markov Model (HMM) to continuously infer the movement intentions of workers, specifically including:

[0090] Step S31. Define the hidden states of the HMM: Set a total of g+3 hidden states to cover all possible intent scenarios of the operator, including: the intent to go to the j-th target node. The intention to reach the target node and complete the task. Intentions are uncertain Irrational behavior or failure to register target nodes .

[0091] Step S32. Construct the dimension as State transition matrix T:

[0092]

[0093] The parameters in the state transition matrix T were determined through experimental calibration, including the probability of change in the target decision. (This represents the probability of transitioning from a clear intention to an indefinite intention, i.e., from a state) Transition to state The probability of starting a new target), (This represents the probability of going from having no clear intention to having a clear current intention, i.e., from state...) Transition to state The probability of target completion and transfer. (This represents the probability of reaching the current target node and completing the target task, i.e., from the state) Transition to state (probability of the target being achieved) and the probability of hesitation after the target is achieved. (This represents the probability of transitioning to a state without clear intention after the goal is achieved, i.e., from state) Transition to state The probability of irrational decision-making (probability of irrationality) (This represents the probability that an agent transitions from a rational state to an irrational state, i.e., from a state of rationality to irrationality.) Transition to state The probability of irrational recovery (This represents the probability that an agent transitions from an irrational state to a rational state, i.e., from a state of rationality...) Transition to state The probability of maintaining the intention to proceed to the j-th target node), The probability that it remains unchanged is Maintain the intention to reach the target node and complete the task. The probability that it remains unchanged is The intention to maintain is uncertain. The probability is Maintaining irrational behavior or failing to register target nodes The probability is This ensures a smooth transition between states for operators during continuous work, adapting to dynamic changes in intent.

[0094] Step S33. Intent Inference: Using the current observation vector generated in step S2... The input HMM is processed using the Viterbi algorithm and dynamic programming to find the most probable sequence of hidden states. If a target node is added or removed during the agent's movement, the target node set and state transition matrix T are expanded or reduced simultaneously. The Viterbi algorithm directly adapts to the new HMM structure without retraining. This yields the probability values ​​of the worker going to each target node, serving as the worker's intent estimation. Based on this result, long-term and short-term intents are distinguished: long-term intent (LTI) represents the worker's overall operational goal, with the probability of goal decision change. Low (e.g., 0.2); Short-term Intent (STI) refers to the local objective at the current stage, with high dynamism (probability of objective decision change). (High, such as 0.5).

[0095] Understandably, the parameters in the state transition matrix T describe the inherent patterns of changes in the operators' intentions. The division between long-term intention (LTI) and short-term intention (STI) reflects both the stability of the operators' overall operational goals (low probability of goal decision change, e.g., 0.2) and the dynamism of local goals in the current stage (high probability of goal decision change, e.g., 0.5). This hierarchical representation allows the intention estimation results to possess both long-term planning and real-time adaptability, providing a more reliable basis for agent collaborative decision-making.

[0096] In some embodiments, step S4 specifically includes:

[0097] Step S41. Determine the look-ahead time distance L based on the operator's intention estimation results: If the intention is clear (the probability of the operator going to a certain target node is significantly higher than that of other target nodes), then L is set to a fixed value; if the intention is uncertain (such as the probability of the operator going to multiple target nodes is close), then extract the multi-intention compatible behavior features of the operator's historical behavior, and determine L based on the duration a of the multi-intention compatible behavior.

[0098] Furthermore, multi-intent compatible behavior refers to an operator's actions simultaneously targeting multiple target nodes (e.g., approaching two shelves at the same time at a path intersection). The duration 'a' of multi-intent compatible behavior is the time interval from the occurrence of the behavior to the operator's explicit intention to favor a particular target node. Then, the look-ahead time distance L simultaneously satisfies the following requirements:

[0099] Minimum constraint: ,in It is the minimum duration of all multi-intention compatible behaviors, ensuring that the look-ahead time distance L can completely contain the action sequence of a single multi-intention compatible behavior;

[0100] Avoid redundancy: ,in It is the maximum duration of all multi-intent-compatible behaviors. A buffer time (0.3-0.5 seconds) is provided to prevent the introduction of irrelevant motion noise due to an excessively long look-ahead time distance L.

[0101] Step S42. Trigger dynamic replanning with a look-ahead time distance L as the period to avoid frequent invalid replanning. In each planning period, the LaCAM* algorithm is used for global path planning and global conflict detection to generate a conflict set between the workers and the agent in the current planning. ,in, This refers to warehouse intelligent agents that conflict with operational personnel in the current planning stage; The conflict time point is the specific moment when multiple workers and warehouse agents are at risk of collision due to path overlap during path planning; The vertex of the path conflict refers to the point in time of the conflict. Next, identify the locations of path conflicts among multiple warehouse intelligent agents; check if the conflict set C is empty. If it is empty, proceed to step S5; otherwise, maintain the original path.

[0102] Understandably, the global path planning strategy employed in this embodiment dynamically adjusts the look-ahead time distance L based on the intent estimation results, achieving intent-driven adaptive replanning. When the operator's intent is clear (the probability of going to a certain target node is significantly higher), a fixed look-ahead distance L is used, balancing planning foresight and computational efficiency. When the intent is uncertain (the probability values ​​of multiple target nodes are close, i.e., multi-intent compatible behavior), the duration of historical multi-intent compatible behavior is used as the look-ahead distance, periodically triggering dynamic replanning, balancing the adaptability of multi-target parallel intents and system load control. The definition of multi-intent compatible behavior precisely adapts to scenarios such as intersecting warehouse operation paths and multi-target parallelism, ensuring that the dynamic adjustment of the look-ahead distance aligns with actual behavior and achieving an optimal balance between planning effectiveness and computational efficiency in the global search space.

[0103] In some embodiments, step S5 uses the Impact Detection (ImpD) method to detect conflicts in the current path configuration in real time, including vertex conflicts (two warehouse agents occupying the same vertex in the global graph model G at the same time) and edge conflicts (two warehouse agents exchanging positions at adjacent time points). The ImpD method is then used to filter the set of warehouse agents affected by conflicts. The specific screening steps include:

[0104] Step S51. Directly affects filtering: Traversing the conflict set The path will contain conflicting vertices. Warehouse intelligence Directly incorporate the set of warehouse agents affected by the conflict .

[0105] Step S52. Indirect Influence Screening: For those elements in the conflict set C that were not directly included... For the warehouse agent, calculate the original path length and the shortest path length to bypass the conflict vertex. If the relative difference between the two [(shortest bypass length - original path length) / original path length] exceeds a preset threshold... If so, the warehouse intelligence agent is determined to be an indirectly affected warehouse intelligence agent and included in the list. .

[0106] Understandably, by employing the above screening strategy, only warehouse agents that are directly related to the conflict or whose detour costs are too high are included in the scope of replanning, thus avoiding the waste of computing resources caused by global replanning.

[0107] In some embodiments, step S6 specifically includes:

[0108] Step S61. Local candidate vertex generation

[0109] Around each conflict vertex in the conflict set C Delineate local subgraphs , This represents the set of vertices corresponding to all traversable locations for the conflicting agents. This represents the set of edges between corresponding vertices that allow direct movement (i.e., connectivity between positions). The local subgraph has the same topology as the global topology of the warehouse environment, but its scope is limited, containing only conflict-related areas and surrounding passable nodes. This represents the set of warehouse agents affected by the conflict. Each warehouse agent generates a set of candidate vertices, which includes three categories:

[0110] (1) Path continuation vertex: pre-compute scattered paths (a set of multiple spatially separated but globally optimal alternative paths generated in a pre-compute manner). In the pre-compute scattered path set, find the path segment with the highest matching degree with the agent's current position, thereby maximizing the reuse of the globally near-optimal path (i.e., the globally approximate optimal path).

[0111] (2) Local neighborhood vertices: local subgraph The vertices directly connected to the current position of the warehouse agent affected by the conflict adapt to local motion rules;

[0112] (3) Staying vertex: The current location of the warehouse agent affected by the conflict, a waiting strategy used to resolve the instantaneous conflict.

[0113] Step S62. Generation of Local Conflict-Free Configurations

[0114] The path and location information of unaffected warehouse agents are fixed, and a rolling time window mechanism is used to only process the affected agents within each planning period. The warehouse agent in the system performs local path adjustments, specifically including:

[0115] Step S621. Collision Filtering: Based on the current global pathfinding results and the latest worker status information and intent estimation results, remove vertices in the candidate vertex set that are occupied by other warehouse agents affected by conflicts, and obtain a conflict-free candidate vertex set. .

[0116] Step S622. Optimal Vertex Selection: Using a LaCAM* cost-first strategy, prioritize selecting conflict-free, globally near-optimal paths; if the path continuation vertices of the pre-calculated distributed paths conflict with the intended location of the workers, then select the conflict-free candidate vertex set. Distance from the center to the target vertex The smallest vertex ensures that the local path cost is consistent with the global cost metric.

[0117] Step S63. Iterative generation: For each affected warehouse agent, sequentially execute the candidate vertex generation in step S61 and the local conflict-free configuration generation in step S62, and iteratively generate conflict-free local configurations.

[0118] It should be noted that the planning period of the rolling time window mechanism is consistent with the look-ahead time distance L used in the global path planning strategy.

[0119] It is understood that step S6 of this embodiment of the invention achieves efficient and robust conflict resolution through local candidate vertex generation and local conflict-free configuration iteration: a topologically consistent but limited local subgraph is defined around the conflicting vertex, generating three types of candidate vertex sets for the affected agent, including path continuation vertices, local neighborhood vertices, and in-place vertices. This maximizes the reuse of global near-optimal path resources by pre-computing the path continuation vertices of the dispersed paths, and adapts local motion rules and instantaneous conflict waiting strategies by using local neighborhood vertices and in-place vertices. Based on this, the positions of unaffected agents are fixed, and collision filtering and LaCAM* cost-first optimal vertex selection are performed only on the affected agents (prioritizing the reuse of dispersed paths, and secondarily selecting the vertex with the smallest distance to align with global costs), and conflict-free local configurations are generated iteratively. This avoids the computational overhead of global replanning and ensures the consistency between local adjustments and global path performance, improving the response efficiency and operational stability of the multi-agent warehousing system in dynamic conflict scenarios.

[0120] In some embodiments, step S7 specifically includes:

[0121] The locally conflict-free configurations generated in step S6 are concatenated into a complete replanning configuration, which is then integrated into the global search tree of LaCAM*. The node costs and parent pointers of the search tree are updated, and the original global replanning logic is replaced. This achieves synergy between local adjustments and global optimization, ensuring that the final path is conflict-free and has near-optimal properties.

[0122] The following are specific embodiments of the present invention:

[0123] This embodiment applies to a large-scale flexible warehousing system comprising 150 warehousing agents and 1 operator. The warehousing layout is 170m × 84m, with 4 target task nodes (g=4) and a detour cost threshold. =0.3.

[0124] The path planning method according to the embodiments of the present invention includes the following specific steps:

[0125] Step S1. Construct a global graph model A GVD graph is constructed based on the warehouse layout. The optimal path between all vertices is pre-calculated using the D* dynamic path planning algorithm to generate the relative distance matrix F.

[0126] Step S2. Collect the position and orientation information of the workers in real time through AR glasses, and calculate the correlation. and the current observation vector v;

[0127] Step S3. Set 7 hidden states for the HMM framework. The state transition matrix T uses preset parameters, where the target decision change probability is... =0.2, probability of initiating a new target =0.1, Probability of target completion and transition =0.1, probability of hesitation after the goal is achieved =0.05, probability of irrational decision-making disturbance =0.05, probability of irrational recovery =0.1; Inferring the operator's intention using the Viterbi algorithm;

[0128] Step S4. Based on the estimation results of the workers' intentions, determine the look-ahead time distance L = 5 time steps. Using the look-ahead time distance L as the period, construct a constraint tree using the LaCAM* algorithm. High-level decision-making filters effective configurations through cost pruning, and generates global path finding results and the conflict set C between workers and agents in the current planning.

[0129] Step S5. Three vertex conflicts were detected using the ImpD method, and the warehouse agents affected by the conflicts were selected.

[0130] Step S6. Construct a local subgraph around the conflict vertices in the conflict set C. For each warehouse agent affected by conflict, 3-5 candidate vertices are generated; after collision filtering and cost-first selection, a conflict-free local configuration is generated.

[0131] Step S7. Finally, merge the conflict-free local configurations into the global search tree to complete the global path update.

[0132] This embodiment demonstrates that, in large-scale scenarios, this method reduces the average response time of replanning and improves computational efficiency compared to traditional global replanning methods, achieving a balance between efficiency and near-optimality.

[0133] A second aspect of the present invention provides a human-machine collaborative warehousing multi-agent pathfinding device based on intent estimation, comprising:

[0134] The topology graph construction module is used to model the warehouse operation space as a graph structure composed of multiple intelligent agent traversable nodes and connecting edges, thereby obtaining a global topology graph of the warehouse environment.

[0135] The state acquisition module is used to acquire the current state information of the operators and multiple intelligent agents in the warehouse environment, so as to determine the correlation between the operators and the nodes of the global topology graph of the warehouse environment, and generate the current observation vector to characterize the movement trend of the operators relative to each target node. The state information of the operators includes at least the posture information of the operators.

[0136] The intent estimation module is used to continuously infer the operator's work target intent based on the operator's historical movement trajectory and current observation vector, and obtain the operator's intent estimation result.

[0137] The conflict detection module is used to infer, based on the intent estimation results, the nodes or edges of the global topology of the warehouse environment that the operator may occupy within a preset time range, and to calculate the global path finding results online.

[0138] The impact analysis module is used to determine the set of agents affected by path conflict when a risk of path conflict is detected between agents or between an agent and a worker during the multi-agent pathfinding process.

[0139] The local path planning module is used to perform path replanning only for the set of agents affected by the conflict within a limited local subgraph, and generate local path schemes that do not conflict with the paths of agents not affected by the conflict.

[0140] The path fusion module is used to update the local path scheme into the global path search result to obtain a multi-agent path search result that satisfies the human-machine collaboration constraint.

[0141] It should be noted that the foregoing explanation of the embodiment of the human-machine collaborative warehousing multi-agent path finding method based on intent estimation also applies to the human-machine collaborative warehousing multi-agent path finding device based on intent estimation in this embodiment, and will not be repeated here.

[0142] To implement the above embodiments, this disclosure also proposes a computer-readable storage medium storing a computer program that is executed by a processor to perform the human-machine collaborative warehousing multi-agent path finding method based on intent estimation described in the above embodiments.

[0143] The following is for reference. Figure 3The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of the present invention. It should be noted that the electronic device in the embodiments of the present invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs, desktop computers, and servers. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0144] like Figure 3 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 102 or a program loaded from a storage device 108 into a random access memory (RAM) 103. The RAM 103 also stores various programs and data required for the operation of the electronic device. The processing unit 101, ROM 102, and RAM 103 are interconnected via a bus 104. An input / output (I / O) interface 105 is also connected to the bus 104.

[0145] Typically, the following devices can be connected to I / O interface 105: input devices 106 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, etc.; output devices 107 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 108 including, for example, magnetic tapes, hard disks, etc.; and communication devices 109. Communication device 109 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0146] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, this embodiment includes a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device 109, or installed from a storage device 108, or installed from a ROM 102. When the computer program is executed by the processing device 101, it performs the functions defined in the methods of the embodiments of this disclosure.

[0147] It should be noted that the computer-readable medium described above in this invention can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0148] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0149] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the aforementioned human-machine collaborative warehousing multi-agent path finding method based on intent estimation.

[0150] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and Python, as well as conventional procedural programming languages ​​such as the "C-" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0151] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0152] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0153] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0154] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0155] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0156] Those skilled in the art will understand that implementing all or part of the steps of the methods in the above embodiments can be accomplished by instructing related hardware through a program. The developed program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0157] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0158] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A human-machine collaborative warehousing multi-agent pathfinding method based on intent estimation, characterized in that, include: The warehouse operation space is modeled as a graph structure consisting of multiple traversable nodes and connecting edges of intelligent agents, resulting in a global topology graph of the warehouse environment. The current state information of workers and multiple intelligent agents in the warehouse environment is obtained to determine the correlation between workers and nodes in the global topology graph of the warehouse environment, and to generate a current observation vector to characterize the movement trend of workers relative to each target node. The state information of workers includes at least the posture information of workers. Based on the historical movement trajectory and current observation vector of the operator, the operator's intention to perform the operation is continuously inferred probabilistically to obtain the operator's intention estimation result; Based on the intent estimation results, the nodes or edges of the global topology of the warehouse environment that the operators may occupy within a preset time range are inferred, and the global path finding results are calculated online. During the multi-agent pathfinding process, when a path conflict risk is detected between agents or between an agent and a worker, the set of agents affected by the conflict is determined. For the set of agents affected by the conflict, path replanning is performed within a limited local subgraph, and local path schemes that do not conflict with the paths of agents not affected by the conflict are generated. The local path scheme is updated to the global path search result to obtain a multi-agent path search result that satisfies the human-machine collaboration constraint.

2. The method according to claim 1, characterized in that, The process of determining the correlation between operators and nodes in the global topology map of the warehouse environment, and generating a current observation vector to characterize the movement trend of operators relative to each target node, includes: Calculate the correlation between the current location of the operator and each node in the global topology map of the warehouse environment. And thereby construct the association vector correlation The calculation formula is: in, This is a Gaussian function used to measure the proximity of the worker to node m. , This represents the total number of nodes in the global topology graph of the storage environment. Let V be the variance of the Gaussian function. For workers and nodes m The Euclidean distance between them; For directional consistency function, Let m be the angle between the operator's posture and the direction of node m; Let the number of target nodes in the current warehouse operation space be . Then define the isolation matrix. for: in, For isolation matrix The element in the i-th row and j-th column is used to separate information related to the target node; Associate vector With relative distance matrix Multiplying the target nodes and separating them yields the modulation range vector. : The relative distance matrix The dimension is , The Middle line, number Column elements For nodes and nodes The pixel distance between the optimal paths; Calculate the candidate correlation vectors for all possible locations that the operator can reach after moving the same distance from the last observation position. And obtain the corresponding candidate modulation distance vector. : ; modulated distance vector With all alternative modulation distance vectors The matrix formed Perform element-level comparison to generate the current observation vector. : in, For matrix The element in the m-th row and j-th column represents the alternative modulation distance value corresponding to the j-th target node when the operator moves the same distance from the historical observation position to the m-th alternative position.

3. The method according to claim 1, characterized in that, When continuously inferring the work objectives and intentions of the workers, a lifetime intention estimation model based on a hidden Markov model is used, and a state transition matrix T is constructed, where, The Hidden Markov Model (HMM) sets up g+3 hidden states, covering all possible intent scenarios of the operator, including the intent to go to the j-th target node. The intention to reach the target node and complete the task. Intentions are uncertain Irrational behavior or failure to register target nodes ; This represents the target number of nodes in the current warehouse operating space. The constructed state transition matrix T is as follows: In the formula, The probability of a change in the target decision. The probability of initiating a new target. The probability of achieving the transition to the target. The probability of hesitation after achieving the goal. The probability of irrational decision-making being disrupted. This represents the probability of irrational recovery.

4. The method according to claim 3, characterized in that, The lifetime intention estimation model estimates long-term intentions based on the worker's historical movement trajectory and current observation vector, and infers the worker's short-term movement direction and target location through the long-term intention estimation.

5. The method according to claim 1, characterized in that, The preset time range is set as the forward look-ahead time distance L for the operator. The setting process includes: If the operator's intention is clear, the look-ahead time distance L is set to a fixed value; If the operator's intention is unclear, then the multi-intention compatible behavior features of the operator's historical behavior are extracted, and the look-ahead time distance L is determined by the duration 'a' of the multi-intention compatible behavior. The multi-intention compatible behavior refers to the operator's action simultaneously pointing to multiple target nodes. The duration 'a' of the multi-intention compatible behavior is the time interval from the occurrence of the behavior to the operator's explicit intention to favor a certain target node. Then the look-ahead time distance L satisfies: , and These are the minimum and maximum durations of all multi-intent-compatible behaviors, respectively. This is the buffer time.

6. The method according to claim 1, characterized in that, The determination of the set of agents affected by the conflict includes: Traverse the set of conflicts between workers and agents in the current plan. ,in, This refers to intelligent agents that conflict with the current operational personnel in the planning stage; The point in time of conflict; The vertex of the path conflict refers to the point in time of the conflict. The location of path conflicts among multiple agents will be determined by including the conflicting vertices in the path. intelligent agents Directly include the set of agents affected by the conflict ; For those not directly included in the conflict set C For each agent, the original path length and the shortest path length around the conflict vertex are calculated. If the relative difference between the two exceeds a preset threshold, the agent is determined to be an indirectly affected agent and is included in the system. .

7. The method according to claim 1, characterized in that, The local subgraph is a storage environment subgraph constructed around the conflict vertices, and its nodes and edges are all subsets of the global topology of the storage environment; When replanning paths for agents affected by conflicts in the local subgraph, the path and position information of unaffected agents are kept unchanged. A rolling time window mechanism is adopted to generate a set of candidate vertices for agents affected by conflicts in each planning cycle based on the latest operator status information and intent estimation results. The optimal vertex is selected through collision filtering and cost priority strategy, and conflict-free local path schemes are generated iteratively.

8. The method according to claim 7, characterized in that, The planning period for the rolling time window mechanism is the preset time range.

9. A human-machine collaborative warehousing multi-agent pathfinding device based on intent estimation, characterized in that, include: The topology graph construction module is used to model the warehouse operation space as a graph structure composed of multiple intelligent agent traversable nodes and connecting edges, thereby obtaining a global topology graph of the warehouse environment. The state acquisition module is used to acquire the current state information of the operators and multiple intelligent agents in the warehouse environment, so as to determine the correlation between the operators and the nodes of the global topology graph of the warehouse environment, and generate the current observation vector to characterize the movement trend of the operators relative to each target node. The state information of the operators includes at least the posture information of the operators. The intent estimation module is used to continuously infer the operator's work target intent based on the operator's historical movement trajectory and current observation vector, and obtain the operator's intent estimation result. The conflict detection module is used to infer, based on the intent estimation results, the nodes or edges of the global topology of the warehouse environment that the operator may occupy within a preset time range, and to calculate the global path finding results online. The impact analysis module is used to determine the set of agents affected by path conflict when a risk of path conflict is detected between agents or between an agent and a worker during the multi-agent pathfinding process. The local path planning module is used to perform path replanning only for the set of agents affected by the conflict within a limited local subgraph, and generate local path schemes that do not conflict with the paths of agents not affected by the conflict. The path fusion module is used to update the local path scheme into the global path search result to obtain a multi-agent path search result that satisfies the human-machine collaboration constraint.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when executed by a processor, are used to implement the human-machine collaborative warehousing multi-agent path finding method based on intent estimation as described in any one of claims 1 to 8.

Images