Multi-agv space-time conflict prediction and cooperative elimination method and system based on infrared light limited communication

Abstract

The application provides a multi-AGV space-time conflict prediction and cooperative elimination method and system based on infrared light limited communication. An AGV continuously collects its own motion state characteristics and calculates the relative relationship characteristics between itself and surrounding AGVs, uses a locally deployed LSTM model to predict conflicts in multiple time steps in the future, and in the case of determining that the conflict prediction is correct, exchanges priority level information with surrounding AGVs through an infrared communication module, according to the conflict type and the exchanged priority level information, queries a locally stored conflict elimination strategy matrix, generates corresponding speed adjustment instructions, and finally executes the speed adjustment instructions to realize autonomous elimination of the conflict, and continuously monitors in the next detection period until no conflict occurs. The method realizes adaptive prediction and autonomous elimination of multi-AGV conflicts under the condition that only sparse nodes can communicate, and greatly improves the running efficiency and running robustness of the AGV system in an offline environment.

Description

Technical Field

[0001] The embodiments of this application relate to the field of automatic control technology, and in particular to a method and system for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared light limited communication. Background Technology

[0002] Automated Guided Vehicles (AGVs) are key equipment for achieving automated production in logistics warehousing, smart manufacturing, and other fields. Their existence successfully eliminates data barriers between the production site and management systems. After being integrated with systems such as Manufacturing Execution Systems (MES) and Warehouse Management Systems (WMS), AGVs can achieve full-process traceability of materials, providing solid data support for the digital transformation of factories.

[0003] The collaborative scheduling performance of AGVs directly affects the operational efficiency and cost control of the entire production line, MES, and WMS. Furthermore, its scheduling performance is closely related to the communication environment. Based on decision-making logic and operating modes, scheduling methods are typically divided into online scheduling and offline scheduling. Online scheduling offers advantages such as real-time response, dynamic adjustment, and single-point fault tolerance, while offline scheduling excels in globally optimal deployment, resource allocation during operation, predictability, and verifiability. Compared to online scheduling, offline scheduling strategies receive less attention, yet they are unavoidable issues. In scenarios with poor or limited communication, most AGVs, after losing communication with the scheduling center, can only perform distance-based emergency braking and other operations. In certain situations where AGVs operate offline or with weak network connectivity, such as in confidential production scenarios or temporary post-disaster operations, online scheduling becomes ineffective. In these cases, the AGVs must rely entirely on their onboard sensors and locally deployed scheduling systems to predict and resolve temporary, localized conflicts to ensure the smooth operation of the entire WMS (Work Management System). This process is highly prone to conflicts, which often require manual intervention, significantly impacting AGV efficiency. Therefore, in real-world industrial settings, most companies combine offline and online scheduling: first, offline scheduling is used for overall baseline planning, and then online scheduling is used to handle unexpected situations during operation. For routine production scenarios, this combination ensures overall efficiency while accommodating dynamic changes.

[0004] Nevertheless, most studies focus on "offline scheduling optimization schemes under conditions where task batches are known, planning is strong, layout is fixed, bottlenecks are clear, and unexpected obstacles are not likely to occur." In this case, due to the lack of operational data support and optimization strategies, offline scheduling is powerless in the face of environments with random task arrivals and strong dynamics, especially in the case of job conflicts arising from multiple AGVs, ultimately resulting in a significant decrease in operational efficiency.

[0005] In recent years, an increasing number of artificial intelligence technologies have been applied to AGV conflict resolution. Spatiotemporal evolution-based prediction methods can mine the temporal or spatial characteristics of AGV motion data, enabling early prediction of potential conflicts. Meanwhile, purely physical communication technologies have made short-distance information exchange possible in offline scenarios.

[0006] The inventors of this application, through analysis of existing similar patents and papers, have found that current methods for conflict prediction and resolution in AGV scheduling systems have the following shortcomings.

[0007] First, existing pre-planning methods are not well-suited for offline scenarios. While conflict avoidance methods based on pre-planning perform well in online scheduling scenarios, they rely heavily on the global information aggregation and real-time computing capabilities of the scheduling center. In situations where communication is limited or completely offline, the information link between the scheduling center and the AGV is interrupted, and the pre-planned scheme cannot be updated in a timely manner, resulting in a lack of effective response mechanisms when facing sudden conflicts.

[0008] Second, existing Petri net modeling methods suffer from high computational complexity. While Petri nets can accurately depict system state transitions, the network structure expands exponentially with the increase in the number of AGVs and the scale of the road network, leading to a severe state space explosion problem that makes it difficult to meet real-time requirements. Furthermore, Petri net models are primarily designed for online scheduling scenarios, and their deployment and computational resource consumption in offline environments are significant.

[0009] Third, existing dynamic adjustment strategies lack localized autonomous decision-making capabilities. Most existing dynamic adjustment methods rely on centralized decision-making by a scheduling center, and individual AGVs lack independent conflict prediction and resolution capabilities. In offline scenarios, the single point of failure risk of this centralized architecture is amplified; once communication is interrupted, the entire system's conflict handling capability will be completely lost.

[0010] Fourth, traditional offline scheduling schemes have poor adaptability. Traditional offline pre-planning schemes cannot cope with emergencies. Any event that deviates from the preset plan will cause the original scheduling scheme to fail, which may eventually lead to the entire system remaining shut down until manual intervention.

[0011] Fifth, traditional localized solutions lack predictive capabilities. Purely local rule-based obstacle avoidance solutions can only detect instantaneous collisions and cannot predict potential conflicts in the next few seconds. Their approach to conflict resolution is "passive," resulting in untimely conflict resolution and prolonged conflict duration. Summary of the Invention

[0012] To address the aforementioned technical issues, embodiments of this application propose a method and system for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication. This aims to utilize spatiotemporal evolution prediction, infrared communication, and local autonomous decision-making to solve the problem of AGV systems being unable to autonomously handle conflicts due to the failure of the scheduling center in offline or weak network scenarios. It achieves adaptive prediction and autonomous resolution of conflicts among multiple AGVs under the condition of "only sparse nodes can communicate," significantly improving the operating efficiency and robustness of AGV systems in offline environments.

[0013] To achieve the above objectives, embodiments of this application propose a method for spatiotemporal conflict prediction and collaborative resolution of multiple AGVs based on infrared light limited communication, applied to AGVs. The method includes the following steps: continuously collecting its own motion state features according to a preset step size, calculating the relative relationship features between itself and surrounding AGVs based on its own motion state features, and storing them in a preset sliding window buffer; when the sliding window buffer is full, inputting the temporal feature sequence in the sliding window buffer into a locally deployed LSTM model, which, after multi-layer temporal feature extraction and fully connected layer feature transformation, outputs conflict flag prediction values ​​and conflict type prediction values ​​for multiple future time steps; and determining the conflict... If the confidence level of the predicted conflict flag value exceeds a preset confidence threshold, the conflict prediction is considered valid; otherwise, it is considered invalid. After a conflict prediction is considered valid, the device ID and priority level information are sent to surrounding AGVs via the infrared communication module, and the device ID and priority level information returned by surrounding AGVs are received. Based on the conflict type prediction value and the exchanged priority level information, the conflict resolution strategy matrix stored locally is queried to generate the corresponding speed adjustment command. The speed adjustment command is executed to achieve autonomous conflict resolution, and data is collected again in the next detection cycle to input into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

[0014] To achieve the above objectives, embodiments of this application also propose a multi-AGV spatiotemporal conflict prediction and collaborative resolution system based on infrared light limited communication, used to implement the multi-AGV spatiotemporal conflict prediction and collaborative resolution method based on infrared light limited communication as described above. The system includes: a data acquisition and feature calculation caching module, a conflict prediction module, a conflict determination module, a priority level information exchange module, an instruction generation module, an instruction execution module, and a continuous monitoring module; the data acquisition and feature calculation caching module is used to continuously acquire its own motion state features according to a preset step size, and calculate the relative relationship features between itself and surrounding AGVs based on its own motion state features, storing them in a preset sliding window buffer; the conflict prediction module is used to input the temporal feature sequence in the sliding window buffer into a locally deployed LSTM model when the sliding window buffer is full, and the LSTM model extracts features through multiple layers of temporal features and fully connected layer features. After conversion, the system outputs predicted conflict flag values ​​and conflict type values ​​for multiple future time steps. A conflict determination module checks if the confidence level of the predicted conflict flag value exceeds a preset confidence threshold; if it does, the conflict prediction is considered valid; otherwise, it is considered invalid. A priority level information exchange module, after determining that a conflict prediction is valid, sends the local device ID and priority level information to surrounding AGVs via an infrared communication module and receives the device ID and priority level information returned by surrounding AGVs. An instruction generation module, based on the predicted conflict type value and the exchanged priority level information, queries the locally stored conflict resolution strategy matrix to generate corresponding speed adjustment instructions. An instruction execution module executes the speed adjustment instructions to achieve autonomous conflict resolution. A continuous monitoring module continues to collect data in the next detection cycle and inputs it into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

[0015] To achieve the above objectives, embodiments of this application also propose an electronic device, including a processor and a memory, wherein the memory stores instructions executable by the processor, and the processor is configured to execute the instructions such that the electronic device can implement the above-described method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared light limited communication.

[0016] To achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a computer program that, when executed by a processor, enables a method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared light limited communication, as described above.

[0017] Optionally, each of the multiple AGVs continuously collects its own motion state features according to a preset step size, including two-dimensional coordinates, heading angle, speed, speed change rate, current road segment number, and distance to the end of the road segment. Based on its own motion state features, it calculates the relative relationship features between itself and surrounding AGVs, including relative distance, relative azimuth angle, heading angle difference, distance change rate, and node preemption risk flag. Each AGV maintains a preset sliding window buffer. After completing the data collection for the current time step, it stores its own motion state features and the relative relationship features between itself and surrounding AGVs in the preset sliding window buffer to form a time-series feature sequence.

[0018] Optionally, the LSTM model uses a multi-layer stacked LSTM structure to extract features from the input temporal feature sequence. The first layer LSTM unit, which contains several hidden state nodes, captures short-term dependencies in the temporal feature sequence. The high-level features are gradually extracted through the multi-layer stacked LSTM structure, and finally a set of deep feature representations that can characterize the motion evolution of the AGV is obtained. The fully connected layer of the LSTM model maps the deep feature representation to the output space. The output of the fully connected layer is divided into two branches. The first branch outputs the predicted values ​​of the conflict flags for multiple future time steps. The predicted values ​​of the conflict flags are 0 or 1, where 0 indicates no conflict and 1 indicates conflict. The second branch outputs the predicted values ​​of the conflict types for multiple future time steps. The conflict types include meeting when moving towards each other, node preemption, node staying, and no conflict.

[0019] Optionally, the LSTM model is trained based on a conflict prediction dataset. If the original data related to the movement of the AGV has been collected and stored on site, the conflict information of the original data is labeled through a rule-based conflict judgment logic to obtain a conflict prediction dataset. If there is no available original data, a conflict prediction dataset is generated by modeling and simulation. The LSTM model is trained using a custom weighted cross-entropy loss function to balance different conflict types and combinations. After training, the model is solidified and packaged in an open neural network exchange format to form an independently callable inference module and embedded in the AGV to achieve real-time inference on the edge.

[0020] Optionally, after determining that a conflict prediction is successful, the AGV sends its own device ID and priority level information to surrounding AGVs via the infrared communication module, and receives the device ID and priority level information returned by the surrounding AGVs. This includes: after determining that a conflict prediction is successful, the current AGV's main control chip sends an encapsulated data frame containing its own device ID and priority level information to the infrared communication module via serial communication. The infrared communication module encodes the data frame into an infrared pulse sequence in NEC format. If there is no obstruction between the current AGV and surrounding AGVs, a direct beam mode is used to communicate point-to-point with the surrounding AGVs to exchange device IDs and priority level information. If there is obstruction between the current AGV and surrounding AGVs, a node relay mode is used. The AGV transmits a signal to the node relay. After receiving the signal from the current AGV, the node relay broadcasts it. The infrared communication module of the surrounding AGV captures and decodes the broadcast signal to parse the device ID and priority level information of the current AGV. Similarly, the infrared communication module of the surrounding AGV also transmits a signal to the node relay, which is then captured by the current AGV after being broadcast. Thus, the exchange of device IDs and priority level information between the two parties is completed.

[0021] Optionally, based on the predicted conflict type and the priority level information of the exchange, the conflict resolution strategy matrix stored locally is queried to generate corresponding speed adjustment instructions. This includes: each AGV queries the conflict resolution strategy matrix stored locally based on the predicted conflict type and the priority level information of the exchange; high-priority AGVs maintain their original plan and pass normally at their original speed, while low-priority AGVs generate corresponding speed adjustment instructions to slow down and give way; the speed adjustment instructions are executed to achieve autonomous conflict resolution, and data continues to be collected in the next detection cycle to be input into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state. This includes: in the next detection cycle, if the high-priority AGV has passed, i.e., the conflict flag has returned to a conflict-free state, then the low-priority AGV generates corresponding speed adjustment instructions to restore its original speed and execute its original plan.

[0022] Optionally, each AGV establishes a three-level local decision-making framework for conflict prediction, priority determination, and strategy matching. The first level outputs the conflict flag prediction value, conflict type prediction value, and confidence level by the LSTM model. The second level obtains the priority level information of the conflicting parties through the infrared communication module. The third level makes a joint decision based on the conflict type prediction value and priority level information to match differentiated resolution strategies.

[0023] The embodiments of this application propose a method for spatiotemporal conflict prediction and collaborative resolution of multiple AGVs based on infrared light limited communication, which achieves the following improvements compared with traditional conflict prediction and resolution methods.

[0024] First, it reduces the subjectivity of conflict prediction. This application achieves objective quantitative prediction of conflicts through an LSTM model, replacing the traditional offline approach that relies on human experience to judge conflicts. This eliminates the subjectivity and inconsistency of human judgment, making the conflict detection standard uniform and reproducible.

[0025] Second, it achieves offline autonomous operation of conflict prediction and collaborative resolution. This application does not rely on a centralized scheduling center and a global wireless network, solving the problem of AGV system paralysis in communication-constrained scenarios. At the same time, the matching of resolution strategies enables automatic conflict handling without manual intervention, improving the efficiency of the entire AGV system.

[0026] Third, it boasts low deployment costs and good compatibility. The hardware in this application uses low-cost, general-purpose components, and the LSTM model supports cross-platform deployment using open neural network exchange formats. It can be directly embedded into existing AGV onboard systems without requiring large-scale modifications.

[0027] Fourth, it effectively improves computational efficiency and real-time performance. Petri nets, which characterize system states and facilitate online scheduling, experience an exponential increase in computational load as the number of AGVs increases or the road network expands. This, coupled with the fact that all instructions must pass through the scheduling center for successful dispatch, often fails to meet real-time requirements. The LSTM model used in this application maintains linear computational complexity, has a shorter single inference time, better real-time performance, higher inference efficiency, and is suitable for embedded platforms. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies of this application will be briefly introduced below. Obviously, the following drawings are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The drawings described herein are only used to explain this application and are not intended to limit this application.

[0029] Figure 1 This is a flowchart of a method for predicting and collaboratively resolving spatiotemporal conflicts of multiple AGVs based on infrared light limited communication, provided in one embodiment of this application; Figure 2 This is a structural diagram of an LSTM model provided in one embodiment of this application; Figure 3 This is a schematic diagram of the topological scene used when training the model provided in one embodiment of this application; Figure 4 This is a visualization of the trajectory information predicted by the AGV during a single task, provided in one embodiment of this application. Figure 5This is a schematic diagram of various indicators under ten consecutive rounds of training provided in one embodiment of this application; Figure 6 This is a schematic diagram of the indicators of each AGV combination in the test set provided in one embodiment of this application; Figure 7 This is a distribution diagram showing the difference between the time when a potential conflict is first predicted and the time when conflict information is actually first marked in the test set provided in one embodiment of this application; Figure 8 This is a schematic diagram of two infrared communication modes provided in one embodiment of this application; Figure 9 This is a comparison chart of the duration of conflict before and after the introduction of this solution in one embodiment of this application; Figure 10 This is a schematic diagram of a multi-AGV spatiotemporal conflict prediction and collaborative resolution system based on infrared light limited communication provided in another embodiment of this application; Figure 11 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. Those skilled in the art will understand that many technical details have been provided in the embodiments of this application to facilitate better understanding. However, the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments. The division of the following embodiments is for ease of description and should not constitute any limitation on the specific implementation of this application. The following embodiments can be combined with and referenced by each other without contradiction.

[0031] In recent years, an increasing number of artificial intelligence technologies have been applied to the field of AGV conflict resolution technology. Spatiotemporal evolution prediction methods can mine the temporal or spatial characteristics of AGV motion data, enabling early prediction of potential conflicts. At the same time, communication technologies using purely physical media such as infrared and ultrasound have made short-range information exchange possible in offline scenarios.

[0032] This application designs a method and system for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared light limited communication. It supports the ability to predict conflicts in advance and autonomously match corresponding resolution strategies when each AGV encounters a conflict under the special offline working condition where "only sparse nodes can communicate".

[0033] Most multi-AGV scheduling projects fall under online operating conditions. For example, Chinese invention patent application number 202511365283.4 (hereinafter referred to as Patent 1) tends to avoid conflicts by using collision warning based on AGV parameters and adjusting strategies such as path replanning through various scheduling algorithms. The data analyzed is mostly concentrated at the efficiency-related task level, and the way to resolve conflicts is more inclined to make pre-planning to eliminate the possibility of conflict, rather than taking measures to prevent the conflict that may have been caused from actually happening.

[0034] Although Chinese invention patent application number 202610038552.4 (hereinafter referred to as Patent 2) uses LSTM as the core technology for conflict prediction and matches a resolution strategy after prediction, it has a fundamentally different technical approach compared with this application. Patent 2 is a centralized online scheduling scheme for a specific scenario of single-channel dense warehouses, where all decisions rely on the global data aggregation and real-time calculation of the scheduling center. In contrast, this application is a distributed autonomous scheduling scheme for general offline industrial scenarios, which constructs a localized decision-making closed loop through edge model deployment and physical layer communication technology. The specific differences are as follows: First, the applicable scenarios and corresponding road networks are different. Patent 2 is only designed for the single specific scenario of single-channel dense warehouses. The algorithm and strategy are highly adapted to the inbound and outbound operations of linear channels and cannot be extended to complex cross-road networks. This application is for general offline industrial scenarios where "only sparse nodes can communicate," and the topology scenario setting is based on this actual working condition (please refer to...). Figure 3 The potential application scenarios include various communication-restricted environments such as confidential production workshops, temporary post-disaster operations, and areas with electromagnetic interference, making its scope of application far greater than that of Patent 2. Secondly, the degree of communication dependence differs. Patent 2 adopts a completely centralized architecture, requiring all AGVs' position, speed, and task status data to be uploaded to the scheduling center in real time via a global sensor network and wireless communication. A communication interruption immediately paralyzes the system. This application, considering the situation where all AGVs have consistent global speeds in actual production, combines [the following is a separate point:] Based on [the following is a separate point:] , the application scenarios include confidential production workshops, temporary post-disaster operations, and areas with electromagnetic interference, making its scope of application far greater than that of Patent 2. The algorithm's path planning implements the generation of trajectory information for each AGV during each task (please refer to...). Figure 4The AGV only needs to synchronize trajectory information at sparse nodes. When completely offline, this asynchronous information sharing mechanism can meet the data requirements for model training, which to some extent solves the core pain point that the AGV system cannot operate normally in communication-limited scenarios. Third, the system architecture and reliability are different. Patent 2 is a centralized decision-making architecture, in which all decisions are made uniformly by the scheduling center, and the AGV only acts as a passive execution terminal. A single point of failure in the scheduling center will cause the entire system to shut down. This application is a distributed local decision-making architecture. After the model is successfully deployed, each AGV can independently complete conflict prediction, priority judgment and strategy execution. The failure of a single AGV will not affect the system. The system's reliability is significantly improved, which does not affect the normal operation of other equipment. Fourth, the decision response speed is different. In Patent 2, conflict detection, priority calculation, and strategy generation are all completed in the scheduling center. The AGV needs to wait for instructions from the center to execute, resulting in unavoidable communication and decision delays. In this application, all decisions are completed on the AGV's local controller without waiting for instructions from the center, resulting in a relatively faster response speed. Fifth, the deployment cost and upgrade difficulty are different. The closed-loop operation of Patent 2 requires a complete sensor network, scheduling server, and wireless communication gateway to achieve, which requires a large investment and a long cycle, limiting its suitability for rapid upgrades of existing AGV systems.

[0035] By analyzing existing similar patents and papers, the inventors of this application have found that current conflict prediction and resolution methods for AGV scheduling systems have the following shortcomings.

[0036] First, existing pre-planning methods are not well-suited for offline scenarios. While conflict avoidance methods based on pre-planning perform well in online scheduling scenarios, they rely heavily on the global information aggregation and real-time computing capabilities of the scheduling center. In situations where communication is limited or completely offline, the information link between the scheduling center and the AGV is interrupted, and the pre-planned scheme cannot be updated in a timely manner, resulting in a lack of effective response mechanisms when facing sudden conflicts.

[0037] Second, existing Petri net modeling methods suffer from high computational complexity. While Petri nets can accurately depict system state transitions, the network structure expands exponentially with the increase in the number of AGVs and the scale of the road network, leading to a severe state space explosion problem that makes it difficult to meet real-time requirements. Furthermore, Petri net models are primarily designed for online scheduling scenarios, and their deployment and computational resource consumption in offline environments are significant.

[0038] Third, existing dynamic adjustment strategies lack localized autonomous decision-making capabilities. Current dynamic adjustment methods largely rely on centralized decision-making by a scheduling center, leaving individual AGVs without independent conflict prediction and resolution capabilities. In offline scenarios, the single point of failure risk of this centralized architecture is amplified; once communication is interrupted, the entire system's conflict handling capability will be completely lost.

[0039] Fourth, traditional offline scheduling schemes have poor adaptability. Traditional offline pre-planning schemes cannot cope with emergencies. Any event that deviates from the preset plan will cause the original scheduling scheme to fail, which may eventually lead to the entire system remaining shut down until manual intervention.

[0040] Fifth, traditional localized solutions lack predictive capabilities. Purely local rule-based obstacle avoidance solutions can only detect instantaneous collisions and cannot predict potential conflicts in the next few seconds. Their approach to conflict resolution is "passive," resulting in untimely conflict resolution and prolonged conflict duration.

[0041] To address the aforementioned challenges and meet the needs of intelligent manufacturing and warehousing logistics for autonomous and collaborative operation of AGV systems without global communication, this application proposes a multi-AGV conflict proactive resolution method and system based on spatiotemporal evolution prediction. Utilizing the ability of spatiotemporal evolution prediction to capture long-term dependencies or spatial characteristics of time-series data, it achieves accurate prediction of the type and timing of potential conflicts. By employing dual-mode AGV infrared communication technology combining direct beam and node relay, it solves the problem of reliable exchange of device information and task priority information between AGVs in offline scenarios. A three-level local autonomous decision-making framework of "conflict prediction - priority discrimination - strategy matching" is constructed, fusing model prediction results with infrared priority information, and matching differentiated adaptive resolution strategies for different conflict types. This effectively improves the operating efficiency and robustness of the AGV system in offline environments, providing a feasible solution for automated operations in special industrial scenarios such as confidential production workshops and post-disaster temporary operations.

[0042] One embodiment of this application proposes a method for predicting and collaboratively resolving spatiotemporal conflicts in multiple AGVs based on infrared light limited communication, which is applied to AGVs. The implementation details of the method for predicting and collaboratively resolving spatiotemporal conflicts in multiple AGVs based on infrared light limited communication proposed in this embodiment are described in detail below. The following implementation details are provided for ease of understanding and are not necessary for implementing this solution.

[0043] The specific process of the multi-AGV spatiotemporal conflict prediction and collaborative resolution method based on infrared limited communication proposed in this embodiment can be described as follows: Figure 1 As shown, it includes: Step 11: Continuously collect its own motion state features according to the preset step size, and calculate the relative relationship features between itself and the surrounding AGVs based on its own motion state features, and store them in the preset sliding window buffer.

[0044] Specifically, in the current scenario, multiple AGVs are working, and it is uncertain whether there is a risk of node preemption conflict. At this time, each AGV needs to continuously collect its own motion state features according to the preset step size, and calculate the relative relationship features between itself and the surrounding AGVs based on its own motion state features, and store the two in the preset sliding window buffer for caching.

[0045] In one example, multiple AGVs continuously collect their own motion state features according to a preset step size, including two-dimensional coordinates, heading angle, speed, rate of change of speed, current segment number, and distance to the end of the segment. Then, based on their own motion state features, they calculate the relative relationship features between themselves and surrounding AGVs, including relative distance, relative azimuth angle, heading angle difference, distance change rate, and node preemption risk flags. Each AGV maintains a preset sliding window buffer. After completing data collection for the current time step, it stores its own motion state features and its relative relationship features with surrounding AGVs into the preset sliding window buffer to form a temporal feature sequence.

[0046] In one example, AGV1 and AGV2 travel towards the same node from different directions. Both AGV1 and AGV2's planned paths include this node, and their estimated arrival time difference is less than a safety threshold, posing a risk of node preemption conflict. AGV1 currently executes a high-priority task, while AGV2 executes a regular task. AGV1 and AGV2 continuously collect their own motion state features in 0.1-second steps and calculate their relative relationship features with each other based on these features. Both AGV1 and AGV2 maintain a sliding window buffer with a length of 30 time steps. After completing data collection and feature calculation for each time step, both AGV1 and AGV2 store their own motion state features and their relative relationship features with surrounding AGVs in the sliding window buffer.

[0047] Step 12: When the sliding window buffer is full, the temporal feature sequence in the sliding window buffer is input into the locally deployed LSTM model. After multi-layer temporal feature extraction and fully connected layer feature transformation, the LSTM model outputs the predicted values ​​of conflict flags and conflict types for multiple future time steps.

[0048] Specifically, once the sliding window buffer is full, the AGV needs to input the temporal feature sequence in the sliding window buffer into the locally deployed LSTM model. After multi-layer temporal feature extraction and fully connected layer feature transformation, the LSTM model outputs the predicted values ​​of the conflict flag and the conflict type for multiple future time steps.

[0049] In one example, the structure of the LSTM model is as follows: Figure 2As shown, the LSTM model uses a multi-layer stacked LSTM structure to extract features from the input temporal feature sequence. The first layer LSTM unit, containing several hidden state nodes, captures short-term dependencies in the temporal feature sequence. High-level features are gradually extracted through the multi-layer stacked LSTM structure, ultimately yielding a set of deep feature representations that characterize the motion evolution of the AGV. The fully connected layer of the LSTM model maps the deep feature representation to the output space. The output of the fully connected layer is divided into two branches. The first branch outputs the predicted conflict flag values ​​for multiple future time steps, with values ​​of 0 or 1 (0 indicating no conflict, 1 indicating conflict). The second branch outputs the predicted conflict type values ​​for multiple future time steps, including but not limited to encounters between moving objects, node preemption, node dwell, and no conflict.

[0050] In one example, the LSTM model is trained based on a conflict prediction dataset. If raw data related to the AGV's movement has already been collected and stored on-site, conflict information is labeled into the raw data using rule-based conflict judgment logic to obtain the conflict prediction dataset. If no raw data is available, a conflict prediction dataset is generated through modeling and simulation. It's important to note that the data acquisition step size can be adjusted according to actual working conditions. To ensure the accuracy and consistency of data labeling, program logic automatically labels the conflict type and occurrence time during data acquisition.

[0051] In one example, the LSTM model is trained using a custom weighted cross-entropy loss function to balance different conflict types and combinations. After training, it is packaged and solidified in an open neural network exchange format to form an independently callable inference module, which is then embedded into the AGV to achieve real-time inference on the edge. The performance of the LSTM model in ten consecutive training rounds is as follows: Figure 5 As shown, the performance on the test set is as follows: Figure 6 , Figure 7 As shown ( Figure 6 The metrics for each AGV combination are displayed. Figure 7 This shows the distribution of the difference between the time when potential conflicts were first predicted and the time when conflict information was actually first marked in the test set.

[0052] In one example, a spatiotemporal conflict prediction algorithm can be used to directly output the conflict prediction result. However, the core model for conflict prediction is not limited to this; various other architectures, such as gated recurrent unit networks and temporal convolutional networks, can also achieve temporal prediction. Furthermore, conflict prediction can also be achieved through a two-stage indirect approach: "predicting motion data and then determining conflict." The first stage uses a temporal prediction model to generate future trajectory and velocity motion data based on historical motion states. The second stage, based on the predicted motion data, combines conflict judgment rules or a simple classifier to determine whether a conflict has occurred and its type. This alternative approach decouples motion prediction from conflict judgment and can serve as a substitute for the direct prediction scheme in situations where conflict judgment rules are clear and the system requires flexible configuration. Moreover, different choices of parameters and architectural levels do not affect the exclusivity of this embodiment.

[0053] Step 13: Determine whether the confidence level of the conflict flag prediction value exceeds the preset confidence level threshold. If it does, the conflict prediction is determined to be valid; otherwise, the conflict prediction is determined to be invalid.

[0054] Specifically, the AGV determines whether the confidence level of the predicted value of the conflict flag exceeds a preset confidence threshold. If it does, the conflict prediction is considered valid; otherwise, the conflict prediction is considered invalid.

[0055] In one example, the preset confidence threshold is set to 0.85, and the confidence level of the predicted conflict flag is 0.92. Therefore, the conflict prediction is considered valid, and the subsequent resolution process begins. Considering that rule-based conflict identification often succeeds when a relevant trend emerges, and by adding the lead time of the prediction results, the predicted conflict occurrence time can generally leave a few seconds of margin from the current time, providing a sufficiently long time window for local policy intervention.

[0056] Step 14: After determining that the conflict prediction is successful, send the local device ID and priority level information to the surrounding AGVs through the infrared communication module, and receive the device ID and priority level information returned by the surrounding AGVs.

[0057] Specifically, once the AGV determines that a conflict prediction has been made, it can send its own device ID and priority level information to the surrounding AGVs through the infrared communication module, and continuously receive the device ID and priority level information returned by the surrounding AGVs to complete the identity verification and priority level information exchange.

[0058] In one example, after a conflict prediction is confirmed, the current AGV's main control chip sends an encapsulated data frame to the infrared communication module via serial communication. This data frame contains the device ID and priority level information. The infrared communication module encodes the data frame into an NEC format infrared pulse sequence. If there is no obstruction between the current AGV and surrounding AGVs, a direct-beam mode is used for point-to-point communication, exchanging device IDs and priority level information. If there is obstruction between the current AGV and surrounding AGVs, a node relay mode is used. The current AGV transmits a signal to the node relay. After receiving the signal from the current AGV, the node relay broadcasts it. The infrared communication modules of the surrounding AGVs capture and decode the broadcast signal, resolving the current AGV's device ID and priority level information. Similarly, the infrared communication modules of the surrounding AGVs also transmit signals to the node relays, which are broadcast and captured by the current AGV. Thus, the exchange of priority level information between the two AGVs is completed.

[0059] This embodiment aims to design a low-cost, highly reliable physical communication solution suitable for industrial offline scenarios, solving the problem of AGVs being unable to exchange information via network. Comparing the communication distance, anti-interference capability, power consumption, and cost of various physical communication media such as infrared, ultrasonic, laser, magnetic induction, and visual recognition, it is clear that infrared light is the better choice as the core communication medium for the actual working conditions under which this embodiment is based. Its typical communication distance of 5m to 10m is sufficient to meet the requirements. Combined with a specific light-shielding cover, it achieves significant advantages in anti-interference while also maintaining good latency performance, excellent cost control, and extremely low power consumption. The AGV hardware adopts a strategy combining an MCU main control chip and an integrated infrared transceiver module, supporting NEC format encoding and decoding. Deployment can be completed without significant modifications to the original AGV hardware structure, offering advantages in scalability, cost control, and power consumption.

[0060] This embodiment aims to design infrared communication modes adaptable to different conflict scenarios, enabling reliable exchange of device IDs and priority information between AGVs in offline states. Addressing the inherent limitations of line-of-sight infrared communication, two complementary communication modes—direct beam and node relay—are designed (please refer to...). Figure 8 The former is suitable for point-to-point information exchange when two AGVs travel towards each other on a road segment, while the latter is suitable for broadcast information exchange when multiple AGVs approach the same node from different directions and there are obstructions. Reliable transmission of priority information is achieved using infrared information repeaters deployed at the node. The communication protocol uses a 3-byte NEC frame format: the first byte is the device ID, the second byte is the priority level information, and the third byte can be set as a checksum or other necessary AGV status information as needed. This design addresses, to some extent, the pain point of priority differentiation between AGVs under limited online communication conditions, providing a basis for subsequent matching and resolution strategies.

[0061] Step 15: Based on the predicted conflict type and the priority level information of the exchange, query the locally stored conflict resolution strategy matrix and generate the corresponding speed adjustment command.

[0062] Step 16: Execute the speed adjustment command to achieve autonomous conflict resolution, and continue to collect data in the next detection cycle to input into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

[0063] Specifically, after the priority level information is exchanged, the AGV will query the locally stored conflict resolution strategy matrix based on the conflict type prediction value and the exchanged priority level information, generate the corresponding speed adjustment command, and then execute the speed adjustment command to achieve autonomous conflict resolution. In the next detection cycle, it will continue to collect data to input into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

[0064] In one example, each AGV queries its locally stored conflict resolution strategy matrix based on the predicted conflict type and the exchanged priority level information. High-priority AGVs maintain their original plan and speed, while low-priority AGVs generate corresponding speed adjustment commands to slow down and yield. In the next detection cycle, if the high-priority AGV has passed (i.e., the conflict flag returns to a conflict-free state), the low-priority AGV will generate corresponding speed adjustment commands to restore its original speed and execute its original plan.

[0065] In one example, AGV1 and AGC2 query their locally stored conflict resolution strategy matrices and independently determine their respective action instructions based on the current conflict type (node ​​preemption conflict) and their own priority status. As a high-priority vehicle, AGV1 is instructed by the strategy matrix to maintain its original plan and speed to pass the node normally. AGV2, as a low-priority vehicle, is instructed to slow down and yield, reducing its speed to 1.0 m / s. AGV2 will begin a smooth deceleration at the next time step to delay its arrival time at the node, allowing AGV1 to pass first. Both vehicles continue to collect time-series data in 0.1-second increments and input it into the model for continuous monitoring. When AGV1 passes the node, the model prediction shows that the conflict flag has returned to a conflict-free state. AGV2 receives a recovery instruction and restores its speed to the globally fixed speed, continuing to execute its original planned path. The entire process from conflict prediction to resolution requires no manual intervention or participation from the scheduling center, achieving fully autonomous local decision-making. Ultimately, the efficiency improvement is reflected in reducing the duration of various conflicts, as detailed in [link to relevant documentation]. Figure 9 .

[0066] Understandably, the conflict resolution strategy matrix is ​​constructed based on conflict type and priority information. For node preemption conflicts, high-priority AGVs maintain their planned speed, while low-priority AGVs slow down to give way. Once the high-priority vehicle has passed the node, they resume normal speed. For node dwell conflicts, AGVs already occupying the node maintain their current state, while those not occupying the node wait in place, following a first-come, first-served principle. A joint threshold mechanism sets differentiated prediction thresholds for different AGV combinations and conflict types. Resolution strategies are only triggered when the prediction confidence exceeds the corresponding threshold, balancing safety and operational efficiency.

[0067] Ultimately, each AGV establishes a three-level local decision-making framework for conflict prediction, priority determination, and strategy matching. The first level outputs the conflict flag prediction value, conflict type prediction value, and confidence level by the LSTM model. The second level obtains the priority level information of the conflicting parties through the infrared communication module. The third level makes a joint decision based on the conflict type prediction value and priority level information, and matches differentiated resolution strategies.

[0068] In one example, a conflict resolution strategy engine based on fuzzy logic reasoning can be designed. This engine fuzzifies numerical inputs such as conflict prediction confidence, priority difference, and task urgency into semantic variables of high, medium, and low. It then infers smooth speed control commands through a predefined fuzzy rule base (e.g., "if the risk is high and the priority difference is large, then immediately apply emergency braking"), replacing binary rigid deceleration gears and improving smoothness in handling boundary conditions. As an alternative to the complete algorithm framework, this conflict resolution problem can be modeled as a partially observable Markov decision process, and an adaptive conflict resolution strategy network can be trained using a deep deterministic policy gradient algorithm.

[0069] This embodiment proposes a method for spatiotemporal conflict prediction and collaborative resolution of multiple AGVs based on infrared light limited communication, which achieves the following improvements compared with traditional conflict prediction and resolution methods.

[0070] First, it reduces the subjectivity of conflict prediction. This embodiment achieves objective quantitative prediction of conflicts through an LSTM model, replacing the traditional offline approach that relies on human experience to judge conflicts. This eliminates the subjectivity and inconsistency of human judgment, making the conflict detection standard uniform and reproducible.

[0071] Secondly, it achieves offline autonomous operation of conflict prediction and collaborative resolution. This embodiment does not rely on a centralized scheduling center and a global wireless network, solving the problem of AGV system paralysis in communication-limited scenarios. At the same time, the matching of resolution strategies enables automatic conflict handling without manual intervention, improving the efficiency of the entire AGV system.

[0072] Third, it boasts low deployment costs and high compatibility. The hardware in this embodiment uses low-cost, general-purpose components, and the LSTM model supports cross-platform deployment using open neural network exchange formats. It can be directly embedded into existing AGV onboard systems without requiring large-scale modifications.

[0073] Fourth, it effectively improves computational efficiency and real-time performance. Petri nets, which characterize system states and facilitate online scheduling, experience an exponential increase in computational load as the number of AGVs increases or the road network expands. This, coupled with the fact that all instructions must pass through the scheduling center for successful dispatch, often fails to meet real-time requirements. The LSTM model used in this embodiment maintains linear computational complexity, has shorter single inference time, better real-time performance, higher inference efficiency, and is suitable for embedded platforms.

[0074] The steps described above are merely for clarity in describing the technical solution. In actual implementation, they can be combined into one step, or certain steps can be broken down into multiple steps, as long as they involve the same logical relationship, they are all within the scope of protection of this application. Any insignificant modifications or designs added to the algorithm or process, as long as they do not change the core of the algorithm or process, are also within the scope of protection of this application.

[0075] Another embodiment of this application proposes a multi-AGV spatiotemporal conflict prediction and collaborative resolution system based on infrared light limited communication. The details of the multi-AGV spatiotemporal conflict prediction and collaborative resolution system based on infrared light limited communication proposed in this embodiment are described in detail below. The following content is only for the convenience of understanding and is not necessary for implementing this solution. Figure 10 This is a schematic diagram of the structure of a multi-AGV spatiotemporal conflict prediction and collaborative resolution system based on infrared light limited communication proposed in this embodiment, including: a data acquisition and feature calculation cache module 21, a conflict prediction module 22, a conflict determination module 23, a priority level information exchange module 24, an instruction generation module 25, an instruction execution module 26, and a continuous monitoring module 27.

[0076] The data acquisition and feature calculation cache module 21 is used to continuously acquire its own motion state features according to a preset step size, calculate the relative relationship features between itself and the surrounding AGVs based on its own motion state features, and store them in a preset sliding window buffer.

[0077] The conflict prediction module 22 is used to input the temporal feature sequence in the sliding window buffer into the locally deployed LSTM model after the sliding window buffer is full. After the LSTM model performs multi-layer temporal feature extraction and fully connected layer feature transformation, it outputs the conflict flag prediction value and conflict type prediction value for multiple future time steps.

[0078] The conflict determination module 23 is used to determine whether the confidence level of the conflict flag prediction value exceeds a preset confidence level threshold. If it exceeds the threshold, the conflict prediction is determined to be valid; otherwise, the conflict prediction is determined to be invalid.

[0079] The priority level information exchange module 24 is used to send the local device ID and priority level information to the surrounding AGVs through the infrared communication module after determining that the conflict prediction is successful, and to receive the device ID and priority level information returned by the surrounding AGVs.

[0080] The instruction generation module 25 is used to query the locally stored conflict resolution strategy matrix based on the conflict type prediction value and the priority level information of the exchange, and generate the corresponding speed adjustment instruction.

[0081] The instruction execution module 26 is used to execute speed adjustment instructions to achieve autonomous conflict resolution.

[0082] The continuous monitoring module 27 is used to continue collecting data in the next detection cycle and input it into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

[0083] It is worth noting that all modules involved in this embodiment are logical modules. In practical applications, a logical module can be a physical module, a part of a physical module, or an organic combination of multiple physical modules. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce modules that are not closely related to solving the technical problems proposed in this application. However, this does not mean that other modules are absent from this embodiment.

[0084] It is not difficult to see that this embodiment is a system embodiment corresponding to the above method embodiments, and this embodiment can be implemented in conjunction with the above method embodiments. The relevant technical details and technical effects mentioned in the above method embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above method embodiments.

[0085] Another embodiment of this application provides an electronic device, such as Figure 11 As shown, it includes a processor 31 and a memory 32. The memory 32 stores instructions that the processor 31 can execute. When the processor 31 is configured to execute the instructions, the electronic device can realize a multi-AGV spatiotemporal conflict prediction and collaborative resolution method based on infrared light limited communication as described in the above method embodiment.

[0086] The memory and processor are connected via a bus, which includes any number of interconnecting buses and bridges. The bus can connect various circuits of one or more processors and memories, as well as other circuits such as peripherals, voltage regulators, and power management circuits—all well-known in the art and therefore not described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which also receives and transmits data to the processor.

[0087] The processor manages the bus and handles general processing, providing various functions, including but not limited to timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory, on the other hand, is used to store data used by the processor during operation.

[0088] Another embodiment of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, can implement a method for predicting and collaboratively resolving spatiotemporal conflicts of multiple AGVs based on infrared light limited communication as described in the above method embodiments.

[0089] That is, those skilled in the art will understand that all or part of the steps in the above method embodiments can be implemented by a program instructing related hardware. The program is stored in a storage medium and includes several instructions to cause a device (such as a microcontroller, chip, etc.) or processor to execute all or part of the steps of the method described in the method embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0090] It will be understood by those skilled in the art that the above embodiments are specific implementations of this application, and various changes in form and detail can be made in practical applications without departing from the spirit and scope of this application. For those skilled in the art, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication, applied to AGVs, characterized in that, The method includes: The system continuously collects its own motion state features according to a preset step size, calculates the relative relationship features between itself and surrounding AGVs based on its own motion state features, and stores them in a preset sliding window buffer. Once the sliding window buffer is full, the temporal feature sequence in the sliding window buffer is input into the locally deployed LSTM model. After multi-layer temporal feature extraction and fully connected layer feature transformation, the LSTM model outputs the predicted values ​​of conflict flags and conflict types for multiple future time steps. Determine whether the confidence level of the predicted value of the conflict flag exceeds the preset confidence level threshold. If it does, the conflict prediction is determined to be valid; otherwise, the conflict prediction is determined to be invalid. After determining that the conflict prediction is successful, the device ID and priority level information are sent to the surrounding AGVs through the infrared communication module, and the device ID and priority level information are received back from the surrounding AGVs. Based on the predicted conflict type and the priority level information of the exchange, query the conflict resolution strategy matrix stored locally and generate the corresponding speed adjustment command. The system executes speed adjustment commands to achieve autonomous conflict resolution, and continues to collect data in the next detection cycle to input into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

2. The method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication according to claim 1, characterized in that, Multiple AGVs continuously collect their own motion status characteristics according to a preset step size, including two-dimensional coordinates, heading angle, speed, speed change rate, current road segment number, and distance to the end of the road segment. Based on their own motion status characteristics, they calculate the relative relationship characteristics between themselves and surrounding AGVs, including relative distance, relative azimuth angle, heading angle difference, distance change rate, and node preemption risk indicator. Each AGV maintains its own preset sliding window buffer. After completing the data acquisition for the current time step, it stores its own motion state characteristics and the relative relationship characteristics between itself and surrounding AGVs into the preset sliding window buffer to form a time-series feature sequence.

3. The method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication according to claim 2, characterized in that, The LSTM model uses a multi-layer stacked LSTM structure to extract features from the input temporal feature sequence. The first layer LSTM unit, which contains several hidden state nodes, captures the short-term dependencies in the temporal feature sequence. The high-level features are gradually extracted through the multi-layer stacked LSTM structure, and finally a set of deep feature representations that can characterize the motion evolution of AGV are obtained. The fully connected layer of the LSTM model maps the deep feature representation to the output space. The output of the fully connected layer is divided into two branches. The first branch outputs the predicted values ​​of the conflict flags for multiple future time steps. The predicted values ​​of the conflict flags are 0 or 1, where 0 indicates no conflict and 1 indicates conflict. The second branch outputs the predicted values ​​of the conflict types for multiple future time steps. The conflict types include meeting when moving towards each other, node preemption, node staying, and no conflict.

4. The method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication according to claim 3, characterized in that, The training of the LSTM model is based on the conflict prediction dataset. If the original data related to the movement of the AGV has been collected and stored on site, the conflict information of the original data is labeled through the regular conflict judgment logic to obtain the conflict prediction dataset. If there is no available original data, the conflict prediction dataset is generated by modeling and simulation. The LSTM model is trained using a custom weighted cross-entropy loss function to balance different conflict types and combinations. After training, it is packaged and solidified in an open neural network exchange format to form an independently callable inference module and embedded in the AGV to achieve real-time inference on the edge.

5. The method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication according to claim 3, characterized in that, After determining that a conflict prediction has been made, the device ID and priority level information are sent to surrounding AGVs via the infrared communication module, and the device ID and priority level information returned by surrounding AGVs are received, including: After determining that the conflict prediction is successful, the current AGV's main control chip sends a packaged data frame to the infrared communication module via serial communication. The data frame contains the local device ID and priority level information. The infrared communication module encodes the data frame into an infrared pulse sequence in NEC format. If there is no obstruction between the current AGV and the surrounding AGVs, the direct-fire mode is adopted to communicate directly with the surrounding AGVs point-to-point and exchange device ID and priority level information. If there is an obstruction between the current AGV and surrounding AGVs, a node relay mode is adopted. The AGV sends a signal to the node relay, and the node relay receives the signal from the current AGV and broadcasts it. The infrared communication module of the surrounding AGVs captures and decodes the broadcast signal, parsing out the device ID and priority level information of the current AGV. Similarly, the infrared communication module of the surrounding AGVs also sends a signal to the node relay, which is then broadcast and captured by the current AGV. At this point, the two parties complete the exchange of device ID and priority level information.

6. The method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication according to claim 5, characterized in that, Based on the predicted conflict type and the priority level information of the exchange, the conflict resolution strategy matrix stored locally is queried to generate the corresponding speed adjustment instructions, including: Each AGV queries the locally stored conflict resolution strategy matrix based on the predicted conflict type and the priority level information of the exchange. High-priority AGVs maintain their original plan and pass normally at the original speed, while low-priority AGVs generate corresponding speed adjustment instructions to slow down and give way. Execute speed adjustment commands to achieve autonomous conflict resolution, and continue collecting data in the next detection cycle to input into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state, including: In the next inspection cycle, if the high-priority AGV has passed, that is, the conflict flag returns to a conflict-free state, the low-priority AGV generates the corresponding speed adjustment instruction to restore the original speed and execute the original plan.

7. The method for predicting and collaboratively resolving spatiotemporal conflicts among multiple AGVs based on infrared limited communication according to claim 6, characterized in that, Each AGV establishes a three-level local decision-making framework for conflict prediction, priority determination, and strategy matching. The first level outputs the conflict flag prediction value, conflict type prediction value, and confidence level by the LSTM model. The second level obtains the priority level information of the conflicting parties through the infrared communication module. The third level makes a joint decision based on the conflict type prediction value and priority level information and matches differentiated resolution strategies.

8. A multi-AGV spatiotemporal conflict prediction and collaborative resolution system based on infrared optical limited communication, used to implement the multi-AGV spatiotemporal conflict prediction and collaborative resolution method based on infrared optical limited communication as described in any one of claims 1 to 7, characterized in that, The system includes: The data acquisition and feature calculation caching module is used to continuously acquire its own motion state features according to a preset step size, and calculate the relative relationship features between itself and the surrounding AGVs based on its own motion state features, and store them in a preset sliding window buffer. The conflict prediction module is used to input the temporal feature sequence in the sliding window buffer into the locally deployed LSTM model after the sliding window buffer is full. After the LSTM model performs multi-layer temporal feature extraction and fully connected layer feature transformation, it outputs the conflict flag prediction value and conflict type prediction value for multiple future time steps. The conflict determination module is used to determine whether the confidence level of the conflict flag prediction value exceeds the preset confidence level threshold. If it exceeds the threshold, the conflict prediction is determined to be valid; otherwise, the conflict prediction is determined to be invalid. The priority level information exchange module is used to send the local device ID and priority level information to the surrounding AGVs through the infrared communication module after determining that the conflict prediction is successful, and to receive the device ID and priority level information returned by the surrounding AGVs. The instruction generation module is used to query the locally stored conflict resolution strategy matrix based on the conflict type prediction value and the priority level information of the exchange, and generate the corresponding speed adjustment instruction. The instruction execution module is used to execute speed adjustment instructions to achieve autonomous conflict resolution; The continuous monitoring module is used to continue collecting data in the next detection cycle and input it into the LSTM model for continuous monitoring until the conflict flag returns to a conflict-free state.

9. An electronic device, characterized in that, include: The processor and memory, wherein the memory stores instructions that the processor can execute, and the processor is configured to, when executing the instructions, enable the electronic device to implement a method for predicting and collaboratively resolving spatiotemporal conflicts of multiple AGVs based on infrared light limited communication as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it can realize a method for predicting and collaboratively resolving spatiotemporal conflicts of multiple AGVs based on infrared light limited communication as described in any one of claims 1 to 7.

Images