An automatic guided vehicle (AGV) control method and device, an AGV, and a storage medium

Abstract

Embodiments of the present application provide an AGV control method and device, an AGV and a storage medium, and relate to the technical field of intelligent logistics. The method comprises: in the moving process of the AGV, in response to detecting a target dynamic obstacle, calling an AI model to perform matching analysis on the target dynamic obstacle and each dynamic obstacle recorded in a predetermined obstacle library to obtain a target type of the target dynamic obstacle; in response to satisfying a fault analysis condition, analyzing a fault probability value of the AGV based on at least one auxiliary feature data of the AGV at present; in response to satisfying a risk evaluation condition, performing motion risk evaluation on the AGV based on a motion route of the target dynamic obstacle currently existing, the fault probability value at present and a current motion route of the AGV; and executing a coping strategy corresponding to the obtained risk evaluation value for the AGV. It can be seen that the present application can achieve the purpose of ensuring the safe operation of the AGV.

Description

Technical Field

[0001] This application relates to the field of intelligent logistics technology, and in particular to an automated guided vehicle (AGV) control method, device, AGV, and storage medium. Background Technology

[0002] Currently, AGV (Automated Guided Vehicle) technology has developed rapidly and has been widely used in the field of intelligent logistics. For example, the control terminal (also known as the control center) issues task instructions on the transportation of goods to the AGV so that the AGV can perform the corresponding operations.

[0003] Ensuring the safe operation of AGVs during their movement is undoubtedly a crucial prerequisite for their stable task completion. Therefore, how to guarantee the safe operation of AGVs has always been a concern. Summary of the Invention

[0004] The purpose of this application is to provide an automated guided vehicle (AGV) control method, device, AGV, and storage medium to ensure the safe operation of the AGV. The specific technical solution is as follows:

[0005] In a first aspect, embodiments of this application provide an automated guided vehicle (AGV) control method, applied to AGVs, the method comprising:

[0006] During the movement of the AGV, in response to the detection of a target dynamic obstacle, at least one first feature data of the target dynamic obstacle is determined; based on the determined first feature data, an AI model is invoked to match and analyze the target dynamic obstacle with each dynamic obstacle recorded in a predetermined obstacle library to obtain the target type of the target dynamic obstacle; and based on the target type and the determined first feature data, the movement path of the target dynamic obstacle is predicted; the at least one first feature data is: feature data obtained by analyzing the data about the target dynamic obstacle collected by each target sensor of the AGV respectively;

[0007] In response to meeting the fault analysis conditions, at least one auxiliary feature data of the AGV is determined, and the fault probability value of the AGV is analyzed based on the determined auxiliary feature data; the at least one auxiliary feature data is: feature data obtained by analyzing the data collected by each auxiliary sensor of the AGV for the AGV or the current scene respectively;

[0008] as well as,

[0009] In response to meeting the risk assessment conditions, a motion risk assessment is performed on the AGV based on the movement path of the currently existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV, to obtain a risk assessment value;

[0010] The AGV will be subject to a response strategy corresponding to the obtained risk assessment value.

[0011] Secondly, embodiments of this application provide an automated guided vehicle (AGV) control device, applied to AGVs, the device comprising:

[0012] A first analysis module is configured to, during the movement of the AGV, in response to the detection of a target dynamic obstacle, determine at least one first feature data of the target dynamic obstacle; based on the determined first feature data, perform matching analysis between the target dynamic obstacle and each dynamic obstacle recorded in a predetermined obstacle library by calling an AI model to obtain the target type of the target dynamic obstacle; and, based on the target type and the determined first feature data, predict the movement path of the target dynamic obstacle; wherein the at least one first feature data is: feature data obtained by analyzing the data about the target dynamic obstacle collected by each target sensor of the AGV respectively;

[0013] The second analysis module is used to determine at least one auxiliary feature data of the AGV in response to meeting the fault analysis conditions, and to analyze the fault probability value of the AGV based on the determined auxiliary feature data; the at least one auxiliary feature data is: feature data obtained by analyzing the data collected by each auxiliary sensor of the AGV for the AGV or the current scene; and,

[0014] The third analysis module is used to assess the motion risk of the AGV based on the motion path of the existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV in response to the fulfillment of the risk assessment conditions, and to obtain the risk assessment value.

[0015] The response module is used to execute response strategies for the AGV that correspond to the obtained risk assessment values.

[0016] Thirdly, embodiments of this application provide an AGV, including:

[0017] Memory, used to store computer programs;

[0018] The processor, when executing the program stored in the memory, implements the above-mentioned automated guided vehicle (AGV) control method.

[0019] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned automated guided vehicle (AGV) control method.

[0020] Fifthly, embodiments of this application also provide a computer program product containing instructions that, when run on a computer, cause the computer to execute any of the above-described automated guided vehicle (AGV) control methods.

[0021] Beneficial effects of the embodiments in this application:

[0022] In the AGV control method provided in this application embodiment, if a target dynamic obstacle is detected during the AGV's movement, the movement path of the target dynamic obstacle is predicted by combining the analysis results of the data collected by various target sensors (i.e., the first feature data) and the AI ​​model. This provides a data basis for analyzing the operational risks of the AGV from the perspective of dynamic obstacle analysis. Furthermore, when the fault analysis conditions are met, the AGV's fault probability value is analyzed using the auxiliary feature data obtained by analyzing the data collected by various auxiliary sensors. This provides a data basis for analyzing the operational risks of the AGV from the perspective of whether the AGV is faulty. Thus, in response to meeting the risk assessment conditions, a movement risk assessment is performed on the AGV based on the current movement path of the target dynamic obstacle, the current fault probability value, and the current movement path of the AGV. Based on the obtained risk assessment value of the AGV, a corresponding response strategy is executed for the AGV, thereby controlling the operational risks during the AGV's movement. Therefore, this solution can achieve the goal of ensuring the safe operation of the AGV.

[0023] Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0025] Figure 1 A flowchart illustrating an automated guided vehicle (AGV) control method provided in this application embodiment;

[0026] Figure 2 Another flowchart of an automated guided vehicle (AGV) control method provided in an embodiment of this application;

[0027] Figure 3(a) is a schematic diagram of the structure of an AGV provided in an embodiment of this application from one perspective;

[0028] Figure 3(b) is a structural schematic diagram of an AGV provided in an embodiment of this application from another perspective.

[0029] Figure 4 This is a schematic diagram of the structure of an automated guided vehicle (AGV) control device provided in an embodiment of this application;

[0030] Figure 5 This is a schematic diagram of the structure of an AGV provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art based on this application are within the scope of protection of this application. Furthermore, in the technical solutions of this application, the operations of obtaining, storing, using, processing, transmitting, providing, and disclosing user personal information are all performed with the user's authorization.

[0032] To ensure the safe operation of AGVs, this application provides an AGV control method. This AGV control method can be applied to AGVs, specifically, it can utilize the AGV's main control unit.

[0033] Furthermore, the AGV may be equipped with at least one target sensor to detect obstacles during its movement. For example, each target sensor may include at least one depth camera and at least one laser sensor, which may include one or more of the following: a 3D laser sensor (also known as a 3D lidar or 3D laser), a navigation laser sensor (also known as a navigation lidar or navigation laser), and an obstacle avoidance laser sensor (also known as an obstacle avoidance lidar or obstacle avoidance laser).

[0034] In addition, the AGV may be equipped with various auxiliary sensors to detect equipment malfunctions; for example, the auxiliary sensors may include sensors for monitoring the AGV itself and / or sensors for monitoring the environment in which the AGV is located. In specific applications, the sensors for monitoring the AGV itself may include, but are not limited to, at least one of vibration detection sensors and noise detection sensors; while the sensors for monitoring the environment in which the AGV is located may include, but are not limited to, smoke detectors, etc.

[0035] Furthermore, sensors of the same type among the target sensors can be deployed on different sides of the AGV to ensure comprehensive sensor data acquisition. Also, for different types of target sensors, they can be deployed in different areas of the same side according to their acquisition range to ensure effective sensor data acquisition. Additionally, the deployment position of any auxiliary sensor on the AGV can be set based on the functions it can perform, combined with human experience, to ensure that the auxiliary sensor can effectively acquire information.

[0036] In this embodiment, the AGV is equipped with at least each target sensor and each auxiliary sensor. Therefore, compared with the problem of the single environmental perception dimension of AGVs in related technologies (AGVs in related technologies rely on ground codes, texture images, lasers or magnetic strips for navigation, and obstacle detection relies on photoelectric / laser), the AGV in this embodiment can enrich the environmental perception dimension, that is, improve the diversity of environmental perception dimensions, and provide a basis for the safe operation and management of AGVs.

[0037] Furthermore, the solution provided in this embodiment can be applied to security management and control of intelligent logistics, and is particularly suitable for high-risk material transportation, unmanned factories, and security linkage in smart parks in industrial scenarios, etc.

[0038] One method for controlling an automated guided vehicle (AGV) may include the following steps:

[0039] During the movement of the AGV, in response to the detection of a target dynamic obstacle, at least one first feature data of the target dynamic obstacle is determined; based on the determined first feature data, the target dynamic obstacle is matched and analyzed with each dynamic obstacle recorded in a predetermined obstacle library by calling an AI model to obtain the target type of the target dynamic obstacle; and based on the target type and the determined first feature data, the movement path of the target dynamic obstacle is predicted; the at least one first feature data is: feature data obtained by analyzing the data about the target dynamic obstacle collected by each target sensor of the AGV respectively;

[0040] In response to meeting the fault analysis conditions, at least one auxiliary feature data of the AGV is determined, and based on the determined auxiliary feature data, the fault probability value of the AGV is analyzed; the at least one auxiliary feature data is: feature data obtained by analyzing the data collected by each auxiliary sensor of the AGV for the AGV or the current scene; and,

[0041] In response to meeting the risk assessment conditions, a motion risk assessment is performed on the AGV based on the movement path of the existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV, and a risk assessment value is obtained.

[0042] The AGV will be subject to a response strategy corresponding to the obtained risk assessment value.

[0043] In the AGV control method provided in this application embodiment, if a target dynamic obstacle is detected during the AGV's movement, the movement path of the target dynamic obstacle is predicted by combining the analysis results of the data collected by various target sensors (i.e., the first feature data) and the AI ​​model. This provides a data basis for analyzing the operational risks of the AGV from the perspective of dynamic obstacle analysis. Furthermore, when the fault analysis conditions are met, the AGV's fault probability value is analyzed using the auxiliary feature data obtained by analyzing the data collected by various auxiliary sensors. This provides a data basis for analyzing the operational risks of the AGV from the perspective of whether the AGV is faulty. Thus, in response to meeting the risk assessment conditions, a movement risk assessment is performed on the AGV based on the current movement path of the target dynamic obstacle, the current fault probability value, and the current movement path of the AGV. Based on the obtained risk assessment value of the AGV, a corresponding response strategy is executed for the AGV, thereby controlling the operational risks during the AGV's movement. Therefore, this solution can achieve the goal of ensuring the safe operation of the AGV.

[0044] The following is a detailed description of an automated guided vehicle (AGV) control method provided in the embodiments of this application, with reference to the accompanying drawings.

[0045] like Figure 1 As shown, an automated guided vehicle (AGV) control method, applied to AGVs, may include the following steps:

[0046] S101, during the movement of the AGV, in response to detecting a target dynamic obstacle, at least one first feature data of the target dynamic obstacle is determined;

[0047] The at least one first feature data is: feature data obtained by analyzing the data on the target dynamic obstacles collected by each target sensor of the AGV.

[0048] S102, Based on the determined first feature data, the target dynamic obstacle and each dynamic obstacle recorded in the predetermined obstacle library are matched and analyzed by calling the AI ​​model to obtain the target type of the target dynamic obstacle;

[0049] S103, based on the target type and the determined first feature data, predict the movement path of the target dynamic obstacle;

[0050] S101-103, and S104-S106 below, are steps performed during the movement of the AGV.

[0051] For this AGV, during the movement of the AGV in response to any task, a control operation (or prevention and control operation) for safe operation can be triggered. For example, any task can be: an inbound task or an outbound task performed after the goods are picked up, or a movement task in response to the operator's voice command, or a charging task in response to the detection that the charging conditions are met, etc.

[0052] Furthermore, in the current scenario where the AGV is located (such as a warehousing scenario), both the AGV and any dynamic obstacle are moving objects. Since there is a risk of collision between two moving objects during their respective movements, any dynamic obstacle may affect the safe operation of the AGV. Therefore, in this embodiment, the dynamic obstacle dimension is used as an analysis dimension for the safe operation of the AGV. During the movement of the AGV, the various target sensors installed on the AGV can periodically collect data, thereby providing a data foundation for the analysis of the dynamic obstacle dimension.

[0053] Specifically, during the movement of the AGV, the presence of target dynamic obstacles can be continuously or periodically detected. In response to the detection of a target dynamic obstacle—that is, when a trigger signal for the presence of a target dynamic obstacle is received—a corresponding analysis process can be executed: determining at least one first feature data of the target dynamic obstacle; based on this first feature data, matching the target dynamic obstacle with various dynamic obstacles recorded in a predetermined obstacle library using an AI model to obtain the target type of the target dynamic obstacle; and predicting the movement path of the target dynamic obstacle based on the target type and the at least one first feature data. It can be understood that the at least one first feature data is information that typically characterizes the graphical structural form of an obstacle. Therefore, the matching analysis process using an AI model is essentially a matching of graphical structural forms, which can be considered a process of AI graph recognition.

[0054] The target dynamic obstacle can be any dynamic obstacle existing within the obstacle detection range of the AGV. The obstacle detection range of the AGV is determined by the acquisition range (i.e., sensing range) of each target sensor. Accordingly, for the target dynamic obstacle, some or all of the sensors in each target sensor can collect data about the target dynamic obstacle (also called sensor data). Furthermore, there can be multiple methods for detecting the existence of the target dynamic obstacle, and this application does not limit this. For example, in one implementation, for the sensor data collected by at least one target sensor, a predetermined dynamic obstacle recognition algorithm or dynamic target detection method is invoked to analyze whether the target dynamic obstacle exists. For example, taking a 3D laser sensor as an example, the spatial position, shape contour, and motion state of the obstacle are obtained through the acquisition of three-dimensional point clouds. The algorithm layer identifies whether it is a dynamic obstacle through point cloud segmentation and target tracking (such as Kalman filtering). Of course, it is also reasonable to use other sensors besides at least one target sensor, such as a camera, to analyze whether the target dynamic obstacle exists.

[0055] Understandably, when a target dynamic obstacle is detected, it is possible to determine which target sensors' sensor data contains data about the same target dynamic obstacle and the location information of the target dynamic obstacle by analyzing whether the acquisition range of each target sensor overlaps with the acquisition range of the sensor used to detect the target dynamic obstacle, and if so, the specific overlap relationship.

[0056] Furthermore, different types of target sensors have different types of first feature data, and can acquire first feature data from multiple consecutive frames (i.e., obtain multiple first feature data from multiple consecutive frames of sensor data) for analysis. For example, for a depth camera (i.e., a depth camera as a target sensor), the sensor data it acquires is a depth image. If a target dynamic obstacle is within its acquisition range, feature data of the target dynamic obstacle containing the target dynamic obstacle can be acquired, such as one or more of spatial geometric features (e.g., contours), motion features, and relative position features (i.e., relative to the target sensor itself or the AGV itself belonging to the depth camera), etc. The analyzed feature data is used as the first feature data of the target dynamic obstacle corresponding to the depth camera. For example, for a laser sensor (i.e., a laser sensor as a target sensor), the sensor data it collects is point cloud data. If a target dynamic obstacle is within its collection range, point cloud data containing the target dynamic obstacle can be collected. Accordingly, based on the collected point cloud data containing the target dynamic obstacle, feature data of the target dynamic obstacle, such as contour features, can be analyzed, and the analyzed feature data is used as the first feature data of the target dynamic obstacle corresponding to the laser sensor. The process of performing feature data analysis based on the sensor data collected by each target sensor is not specifically limited; any method capable of feature data analysis can be applied to the embodiments of this application. Furthermore, the type of feature data is not limited as long as the analyzed feature data can characterize the spatial position and motion characteristics of the target dynamic obstacle.

[0057] Furthermore, after acquiring the first feature data, considering that the main factor affecting the safe operation of the AGV's movement process is the movement path of the target dynamic obstacle, a path prediction analysis can be performed on the target dynamic obstacle based on the acquired first feature data to obtain the movement path, i.e., the movement path of the target dynamic obstacle is estimated for use in the analysis process for safe operation. Specifically, for S102, if there are multiple first feature data, these multiple first feature data can be fused to obtain feature data to be used; after obtaining the feature data to be used, an AI model can be called to match and analyze the feature data to be used with the obstacle data of each dynamic obstacle in the predetermined obstacle library, and determine the obstacle type of the dynamic obstacle to which the matching obstacle data belongs, as the target type of the target dynamic obstacle. Then, after obtaining the target type of the target dynamic obstacle, S103 can be triggered by combining the at least one first feature data.

[0058] In cases where multiple first feature data points are acquired, these first feature data points may include data obtained from sensor data analysis of different types of target sensors, and / or data obtained from sensor data analysis of the same type of target sensors. In such cases, the acquired first feature data can be fused to obtain fused data, which can then be used for subsequent analysis. For example, the fusion process may include: weighted fusion of feature data representing the same attribute (such as contour attribute); and after weighted fusion, concatenating the feature data of different attributes to obtain the feature data to be used. This is not limited to this step. Furthermore, considering that the obstacle type of the target dynamic obstacle affects its movement path, an analysis of at least the obstacle type can be performed on the target dynamic obstacle based on the feature data to be used.

[0059] It should be noted that the AI ​​model can be configured with or associated with a predetermined obstacle library. This library can store obstacle data for different types of dynamic obstacles, i.e., it can store feature data templates for different types of dynamic obstacles. Furthermore, different types of dynamic obstacles can be associated with descriptive information containing at least the obstacle type. Therefore, after obtaining the feature data to be used for the target dynamic obstacle, the AI ​​model can be invoked to match the feature data with the obstacle data in the predetermined obstacle library. This includes, for example, matching analysis to calculate data similarity, and determining the obstacle type of the dynamic obstacle to which the matching obstacle data belongs. The descriptive information associated with different types of dynamic obstacles can also include other information related to route prediction analysis, such as motion pattern information. Thus, in addition to using the target type and at least one first feature data, motion pattern information can also be used to analyze the movement path of the target obstacle.

[0060] Furthermore, the way to invoke the AI ​​model may include, but is not limited to, inputting the feature data to be utilized and the prompt word information into the AI ​​model. The content of the prompt word information is used to instruct the AI ​​model to perform the following process: matching the feature data to be utilized with the obstacle data of each dynamic obstacle in the predetermined obstacle library, and determining the type of dynamic obstacle to which the matching obstacle data belongs, and identifying it as the target type.

[0061] Furthermore, there are multiple ways to perform route prediction analysis on the target dynamic obstacle based on the target type and the acquired first feature data to obtain the movement route. Any method that can perform route prediction analysis on the target dynamic obstacle based on the target type and the acquired first feature data can be applied to this application. For example, using the target type and the acquired first feature data as input and calling an AI model, or calling a pre-trained neural network model for route planning, or calling a route prediction algorithm for the target type, etc., are all reasonable.

[0062] S104, in response to meeting the fault analysis conditions, determine at least one auxiliary feature data of the AGV at present, and analyze the fault probability value of the AGV based on the determined auxiliary feature data;

[0063] Among them, the at least one auxiliary feature data is: feature data obtained by analyzing the data collected by each auxiliary sensor of the AGV.

[0064] In the current scenario where the AGV is located, the safety of the environment in which the AGV operates and whether the AGV itself is faulty both affect its safe operation during movement. Therefore, safety analysis can be conducted from the perspective of whether the AGV is faulty. Since the environment in which the AGV operates can also affect its ability to malfunction, the analysis of whether the AGV is faulty includes not only the AGV itself but also the analysis of its environment. Therefore, to better ensure safe operation, various auxiliary sensors can be installed on the AGV to monitor its own sensors or the environment in which the AGV operates.

[0065] Each auxiliary sensor can periodically collect data, providing a data foundation for analyzing the failure probability of the AGV. Furthermore, for any auxiliary sensor, feature extraction can be performed on the sensor data collected to obtain corresponding auxiliary feature data, which can be used to analyze whether the AGV is faulty. For example, smoke signal features (such as smoke concentration, smoke scattered light intensity / absorbed light intensity) can be extracted from the signal collected by a smoke detector to obtain corresponding auxiliary feature data for the smoke detector; for vibration detection sensors, vibration signal features in characteristic frequency bands can be extracted from the collected sensor data to obtain corresponding auxiliary feature data for the vibration sensor; and for noise detection sensors, audio signal features in characteristic decibel bands can be extracted from the collected sensor data to obtain corresponding auxiliary feature data for the noise detection sensor.

[0066] Furthermore, pre-set fault analysis conditions allow for analysis of whether these conditions are met. Upon fulfillment of these conditions, the system determines the current auxiliary feature data for the AGV and analyzes its fault probability based on this data. Multiple fault analysis conditions can exist. For example, fault analysis can be performed periodically, with conditions including entering a cycle. Alternatively, the fault analysis condition can be determined when sensor data collected by an auxiliary sensor meets conditions indicating data anomalies, which is also reasonable.

[0067] Furthermore, there are various ways to analyze the failure probability value of the AGV based on the acquired auxiliary feature data, and this application does not impose any specific limitations. For example, in one implementation, the acquired auxiliary feature data is input into a pre-trained neural network model to obtain the failure probability value of the AGV; or, in another implementation, an AI model can be used to analyze the failure probability value of the AGV based on the acquired auxiliary feature data, and so on.

[0068] S105, in response to meeting the risk assessment conditions, based on the movement path of the currently existing target dynamic obstacle, the current failure probability value and the current movement path of the AGV, a motion risk assessment is performed on the AGV to obtain a risk assessment value;

[0069] Combining the analysis dimensions of dynamic obstacles and AGV malfunction, a motion risk assessment can be performed on the AGV when the risk assessment conditions are met. These risk assessment conditions can be varied. For example, the motion risk analysis of the AGV can be performed periodically. The risk assessment conditions could include: entering a cycle; or, detecting a target dynamic obstacle; or, detecting a target obstacle and determining that the risk assessment conditions are met when the most recent failure probability value is not lower than a predetermined threshold. All of these are reasonable.

[0070] Since the target dynamic obstacle, the failure probability value, and the AGV's movement route are all changing during the AGV's movement, the movement route of the target dynamic obstacle, the current failure probability value, and the current movement route of the AGV all refer to the most recently obtained data when the risk assessment conditions are met. The most recently obtained movement route of the target obstacle, the most recently obtained failure probability value, and the AGV's movement route when the risk assessment conditions are met are respectively used as the movement route of the target dynamic obstacle, the current failure probability value, and the current movement route of the AGV.

[0071] Furthermore, there are multiple ways to assess the motion risk of an AGV based on the movement path of the existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV. For example, in one implementation, an AI model can be invoked to analyze the risk assessment value using a pre-defined risk assessment value setting strategy. This strategy involves setting the risk assessment value based on inputs including the dynamic obstacle's movement path, the failure probability value, and the AGV's movement path. For example, in another implementation, a pre-trained neural network model can be invoked to predict the risk assessment value using inputs including the dynamic obstacle's movement path, the failure probability value, and the AGV's movement path. For example, in another implementation, the current movement path of the existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV are used as input. The input, along with predefined prompts, are fed into the AI ​​model. The AI ​​model then analyzes the risk assessment value according to the prompts. For instance, the prompts can indicate: based on the current movement path of the target dynamic obstacle and the current movement path of the AGV, analyze the probability of a collision between the AGV and the target dynamic obstacle. Then, through weighted and other comprehensive analysis methods, perform a comprehensive analysis of the obtained collision probability and the current failure probability value. Based on the results of this comprehensive analysis... The system uses a predefined method for assigning risk assessment values ​​to perform motion risk assessments on the AGV to predict the risk assessment value; alternatively, the prompt message contains multiple historical reference data and instructions for performing risk assessments on the current input content based on these historical reference data. Each historical reference data includes: the predicted motion path for a historical dynamic obstacle (a dynamic obstacle detected by the sample AGV during its movement), the historical failure probability value of the sample AGV (the failure probability determined by the sample AGV during the aforementioned movement), the historical movement path of the sample AGV (the movement path of the sample AGV during the aforementioned movement), and the risk assessment value of the sample AGV.

[0072] Furthermore, by matching the movement path of the existing target dynamic obstacle with the current movement path of the AGV, the focus can be on reflecting the collision risk. On the other hand, by reflecting the current failure probability value, the focus can be on reflecting the equipment risk of the AGV itself. Therefore, by conducting a motion risk assessment on the AGV, regardless of the specific assessment method used, the level of the risk assessment value obtained can represent the number of risk types and the probability of each risk type. Ultimately, the higher the risk assessment value, the greater the impact on operational safety.

[0073] S106, Implement response strategies for AGVs that correspond to the obtained risk assessment values.

[0074] After obtaining the risk assessment value of the AGV, risk control can be implemented. This involves executing corresponding response strategies based on the AGV's risk assessment value. These strategies are tailored to the AGV's risk assessment value, ensuring its safe operation during movement. If the AGV's risk assessment value is below a predetermined threshold, the response strategy can be empty, meaning no action is required. This predetermined threshold distinguishes between risks requiring intervention. Furthermore, different response strategies can be set for different risk assessment values, including but not limited to: initiating a local obstacle avoidance process and uploading local obstacle avoidance events; triggering early warning components to issue warning signals; constructing electronic fences for restricted / dangerous areas to trigger deceleration or prohibition to prevent collisions; and coordinating with other equipment for safety management. In this application, the risk assessment value obtained through analysis can not only represent the level of risk, but also implicitly represent the type of risk currently existing in the AGV. Thus, the corresponding response strategy can be located through the risk assessment value obtained through analysis. Furthermore, the correspondence between different risk assessment values ​​and their respective response strategies can be set based on experience and can be iteratively updated periodically or non-periodically.

[0075] For example, in one implementation, step S105 above may include steps A01-A03:

[0076] Step A01: Analyze the risk level to which the obtained risk assessment value belongs to obtain the target risk level; where different risk levels correspond to different risk assessment value ranges;

[0077] Step A02: Determine the response strategy corresponding to the target risk level according to the pre-set correspondence between each risk level and response strategy.

[0078] Step A03: Implement the response strategy corresponding to the target risk level for the AGV.

[0079] In this implementation, different risk levels can be set, and corresponding response strategies can be defined for each risk level. That is, different risk levels have different response strategies. After obtaining the risk assessment value of the AGV, it is possible to analyze which risk level the AGV falls into (different risk assessment values ​​within the same risk assessment value range can implicitly represent the same risk type and correspond to the same response strategy), thereby obtaining the target risk level of the AGV. Then, the response strategy corresponding to the target risk level is executed for the AGV. This method allows for rapid and accurate determination of response strategies. Furthermore, the correspondence between each risk level and response strategy can be set based on experience and can be iteratively updated periodically or non-periodically.

[0080] In one alternative implementation, after determining the target risk level, the method provided in this application embodiment may further include:

[0081] The target risk level is reported to the control terminal of the AGV so that the control terminal receives the target risk level reported by the AGV. In response to receiving the target risk level reported by at least a specified number of AGVs in the current scenario, the route reconstruction instruction is issued to the AGVs in the current scenario.

[0082] In response to receiving a route reconfiguration command from the control terminal, route reconfiguration processing is performed.

[0083] In this approach, when multiple AGVs in the current scenario all have high risk levels, it indicates that the overall operation of the AGVs in the current scenario is risky. If intervention is not timely, the consequences for the current scenario will be quite serious. Therefore, this control terminal can be used to achieve global control over the operational safety of the AGVs in the current scenario, further improving operational safety.

[0084] In the AGV control method provided in this application embodiment, if a target dynamic obstacle is detected during the AGV's movement, the movement path of the target dynamic obstacle is predicted by combining the analysis results of the data collected by various target sensors (i.e., the first feature data) and the AI ​​model. This provides a data basis for analyzing the operational risks of the AGV from the perspective of dynamic obstacle analysis. Furthermore, when the fault analysis conditions are met, the AGV's fault probability value is analyzed using the auxiliary feature data obtained by analyzing the data collected by various auxiliary sensors. This provides a data basis for analyzing the operational risks of the AGV from the perspective of whether the AGV is faulty. Thus, in response to meeting the risk assessment conditions, a movement risk assessment is performed on the AGV based on the current movement path of the target dynamic obstacle, the current fault probability value, and the current movement path of the AGV. Based on the obtained risk assessment value of the AGV, a corresponding response strategy is executed for the AGV, thereby controlling the operational risks during the AGV's movement. Therefore, this solution can achieve the goal of ensuring the safe operation of the AGV. Furthermore, compared to the low efficiency of multi-sensor data fusion and the independent operation of each sensor in related technologies, this embodiment can improve fusion efficiency and effectively combine the various sensors. In addition, compared to the large response delay of AGV anomaly detection in related technologies, which relies on the central server (i.e., the control terminal) for decision-making, the AGV in this embodiment does not need to rely on the central server and can achieve autonomous safety control, thus improving response efficiency. Moreover, in this embodiment, safe operation control can be effectively performed during the movement of the AGV, thereby solving the problem of the separation of security and transportation functions in related technologies.

[0085] Alternatively, in another embodiment of this application, such as Figure 2 As shown in the embodiments of this application, an automated guided vehicle (AGV) control method may include the following steps:

[0086] S201, during the movement of the AGV, in response to detecting a target dynamic obstacle, at least one first feature data of the target dynamic obstacle is determined;

[0087] The at least one first feature data is: feature data obtained by analyzing the data on the target dynamic obstacles collected by each target sensor of the AGV.

[0088] S202, Based on the determined first feature data, the target dynamic obstacle and each dynamic obstacle recorded in the predetermined obstacle library are matched and analyzed by calling the AI ​​model to obtain the target type of the target dynamic obstacle;

[0089] S203, based on the target type and the determined first feature data, predict the movement path of the target dynamic obstacle;

[0090] In this embodiment, the movement process of the AGV is a movement process performed in response to a target task, which is an inbound or outbound task executed after the goods are picked up.

[0091] Steps S201-S203 are the same as S101-S103 in the above embodiments, and will not be described again here.

[0092] S204, in response to meeting the fault analysis conditions, determine at least one auxiliary characteristic data of the AGV at present, and analyze the fault probability value of the AGV based on the determined auxiliary characteristic data;

[0093] Step S204 is the same as S104 in the above embodiment, and will not be described again here.

[0094] S205, acquire the first type of data collected by the target depth camera set on the AGV for the current goods, and acquire the second type of data collected by the barcode reader camera set on the AGV for the current goods; and, based on the first type of data and the second type of data, determine the goods matching degree between the current goods and the target goods;

[0095] The target cargo refers to the cargo that the AGV is set to pick up.

[0096] After the AGV completes its pickup, the matching degree between the current cargo and the target cargo has a critical impact on the AGV's safe operation. For example, if the load weight is mismatched, it may trigger the safety threshold failure of the AGV's power and braking systems; if the dimensions / volume are mismatched, it may cause scraping and collision risks during spatial passage; if the center of gravity is mismatched, it may trigger the AGV to tip over or the cargo to fall, posing a significant safety risk. In other words, abnormal cargo matching (low matching / mismatch) will break the AGV's safe operating threshold, causing safety risks such as driving deviation, collisions, and tipping over, and will also indirectly affect the effectiveness of the AGV's navigation and control logic. Therefore, during the AGV's movement in response to the target task, the cargo matching degree between the current cargo and the target cargo can be analyzed, and the cargo matching degree can also be applied to the motion risk assessment process.

[0097] It should be noted that, in order to incorporate cargo matching into the motion risk assessment process, the AGV can be equipped with target depth cameras and barcode readers, thus providing a basis for cargo matching from different dimensions. The target depth camera can be deployed at a location capable of capturing the outline of the current cargo, and similarly, the barcode reader can be deployed at a location capable of reading the tag information of the current cargo. In actual use, the specific positions of the target depth camera and barcode reader can be relative to the actuators of the AGV, but are not limited to these.

[0098] For example, in one implementation, determining the matching degree between the current goods and the target goods based on the first type of data and the second type of data may include the following steps:

[0099] An AI model is invoked to perform matching analysis between the cargo outline in the first type of data and the cargo outline template corresponding to the target cargo, to obtain a first analysis result. Then, the second type of data is matched with the identification information of the target cargo to obtain a second analysis result. Based on the first and second analysis results, the cargo matching degree between the current cargo and the target cargo is determined.

[0100] In this implementation, an AI model is used to analyze the matching degree of goods from two dimensions to improve the accuracy of the analysis. Specifically, the matching analysis of the goods outlines in the first type of data with the corresponding goods outline template of the target goods can be considered part of the AI ​​image recognition process, i.e., obtaining the first analysis result by analyzing the outline similarity of the goods. The matching analysis of the second type of data with the identification information of the target goods can be considered part of the text similarity analysis, i.e., obtaining the second analysis result by analyzing text similarity.

[0101] The method for determining the matching degree between the current goods and the target goods based on the first analysis result and the second analysis result may include: selecting the first analysis result with the lowest matching value from the first analysis result and the matching degree that matches the first analysis result as the matching degree between the current goods and the target goods; or, taking a weighted average of the matching value represented by the first analysis result and the matching value represented by the second analysis result to obtain the matching degree between the current goods and the target goods.

[0102] The above-described method for determining the matching degree between the current goods and the target goods is merely an example and should not be construed as limiting the embodiments of this application. Any method capable of analyzing the matching degree based on the first analysis result and the second analysis result can be applied to this application.

[0103] S206, in response to meeting the risk assessment conditions, based on the movement path of the currently existing target dynamic obstacle, the current failure probability value, the current movement path of the AGV, and the cargo matching degree, a motion risk assessment is performed on the AGV to obtain a risk assessment value.

[0104] In this embodiment, since the cargo matching degree is obtained from the analysis, the cargo matching degree, together with the movement route of the currently existing target dynamic obstacle, the current failure probability value, and the current movement route of the AGV, can be applied to the motion risk assessment to conduct motion risk assessment from more analytical dimensions.

[0105] S206 is similar to the motion risk assessment method described in S105. The difference between step S206 and S105 is that step S206 adds data on the dimension of cargo matching degree.

[0106] S207, Implement response strategies for AGVs that correspond to the obtained risk assessment values.

[0107] Step S207 is the same as S106 in the above embodiment, and will not be described again here.

[0108] In this embodiment, for the AGV's movement process, the analysis provides a data foundation for AGV operational risk assessment from the perspectives of dynamic obstacle analysis, AGV malfunction analysis, and cargo matching degree analysis. Thus, in response to meeting risk assessment conditions, a movement risk assessment is performed on the AGV based on the movement path of the existing target dynamic obstacles, the current failure probability value, the AGV's current movement path, and the cargo matching degree. Based on the obtained risk assessment value, corresponding response strategies are implemented for the AGV, thereby enabling more comprehensive control over operational risks during AGV movement. Therefore, this solution can further ensure the safe operation of AGVs.

[0109] Optionally, in another embodiment of this application, the automated guided vehicle (AGV) control method provided by the present invention may further include the following steps B01-B02:

[0110] Step B01: During the movement of the AGV or when the AGV is stationary, in response to the activation of the sentry mode, based on the data periodically collected by each designated sensor, it is determined whether any of the predetermined security trigger conditions are met.

[0111] The designated sensors include one or more of the target sensors and the auxiliary sensors.

[0112] The AGV may be provided with a sentry mode. No matter during the moving process of the AGV or during the stationary state of the AGV, if the sentry mode is enabled, a corresponding monitoring process is executed, that is, the monitoring process is performed by means of data collected by each designated sensor to realize the security function.

[0113] Different types of designated sensors collect different types of data, so different types of designated sensors may correspond to different security trigger conditions. For example, if a smoke detector is used as a designated sensor, the security trigger condition may include the presence of smoke; while for a noise detection sensor as a designated sensor, the security trigger condition may include: whether the noise exceeds a predetermined threshold; if a depth camera is used as a designated sensor, the security trigger condition may include: presence of object intrusion, for example, presence of a moving object entering the current scene or a person on the blacklist entering the current scene, etc.

[0114] Step B02: if satisfied, execute the security response operation corresponding to the satisfied security trigger condition;

[0115] Wherein, the security response operation includes a control operation for the AGV itself or an operation for a preset linkable security device.

[0116] Different security trigger conditions may correspond to different security response operations. When the condition is satisfied, the security response operation corresponding to the satisfied security trigger condition may be executed to perform effective security protection. The operation for the preset linkable security device may include: a linkage operation for a predetermined monitoring device in the current scene, etc.

[0117] For example, if the satisfied security trigger condition is any of the following conditions, the corresponding security response operation may include the AGV locking its position and starting 360° panoramic shooting:

[0118] A condition indicating smoke alarm trigger, a condition indicating abnormal vibration or noise, a condition indicating that the obstacle avoidance sensor is triggered, and a condition indicating that the image collected by any depth camera matches the anomaly library in the AI model.

[0119] If the satisfied security trigger condition is any of the following conditions, the corresponding security response operation may include the AGV performing tracking and synchronizing alarm information to the control center (i.e., the aforementioned control terminal):

[0120] A moving object enters the current scene or a person on the blacklist enters the current scene.

[0121] It should be noted that after the sentry mode is activated, if the smoke detector is not triggered, there is no abnormal vibration or noise, the obstacle avoidance sensor is not triggered, and the image captured by the depth camera does not match the anomaly database in the AI ​​model, it indicates that there is no anomaly in the current scene. At this time, the AGV can perform operations such as cruising along the preset route or normally executing the work tasks issued by the platform, without triggering any security response operations.

[0122] In this embodiment, by activating the sentry mode, further security operations can be performed during the movement of the AGV or when the AGV is stationary, thereby further ensuring the safe operation of the AGV and the safety of the current scene.

[0123] Optionally, in another embodiment of this application, the automated guided vehicle (AGV) control method provided by the present invention may further include the following steps:

[0124] In response to the fulfillment of the mapping triggering condition, the system acquires second feature data about the obstacle obtained by processing the data periodically collected from each target sensor, and performs three-dimensional mapping based on the obtained second feature data for use during the movement of the AGV.

[0125] And / or,

[0126] In response to the route update conditions during the movement of the AGV, the third feature data about the obstacle is obtained by processing the data collected by each target sensor, and the route prediction of the AGV is performed based on the obtained third feature data to obtain the updated route on which the AGV movement is based.

[0127] In this embodiment, the mapping triggering conditions may include: initial mapping conditions (such as the conditions for triggering initial mapping that are met when the AGV is deployed to the current scene) or real-time mapping conditions triggered during movement. The mapping triggering conditions can be the same as those in related technologies, and will not be elaborated upon here. Additionally, route update conditions may include, but are not limited to: entering a route update cycle, or detecting an obstacle avoidance command or update command, etc.

[0128] The second and third characteristic data about obstacles can include: characteristic data about static obstacles and characteristic data about dynamic obstacles.

[0129] Since the target sensors can include sensors of the same type or different types, the same obstacle can have feature data from the same type of sensor or feature data from different types of sensors. During the 3D mapping process based on the obtained second feature data, the feature data of the same obstacle in the obtained second feature data can be fused. Based on the fused feature data, a preset mapping algorithm is called to perform 3D mapping. It can be understood that for each obstacle, because feature data from multiple target sensors are fused, the feature data of that obstacle can represent more feature points of that obstacle, thus making the mapping of that obstacle more accurate.

[0130] Similarly, based on the obtained third feature data, during the route prediction process for the AGV, the feature data of the same obstacle in the obtained third feature data can be fused. Based on the fused feature data, a preset route update algorithm is called to update the route. It is understood that for each obstacle, since feature data from multiple target sensors are fused, the feature data of that obstacle can characterize more feature points, thereby improving the accuracy of the route update.

[0131] In this embodiment, by combining various target sensors, an accurate 3D map can be constructed for application during the movement of the AGV, and the accuracy of route prediction can be improved, thereby further enhancing the operational safety during the movement process.

[0132] Alternatively, in another embodiment of this application, each of the target sensors and auxiliary sensors and a sensor control unit are divided into a packaged unit. The components within each packaged unit communicate with each other via CAN (Controller Area Network) or Ethernet. Each packaged unit communicates with other components outside the packaged unit via optical fiber.

[0133] Each encapsulation unit contains a sensor control unit for performing specific analyses on data collected by sensors within the same encapsulation unit. These analyses include, but are not limited to, denoising and feature extraction of the data collected by sensors within the same encapsulation unit. Correspondingly, the AGV's main control unit can directly utilize the feature data processed by the sensor control unit for control purposes.

[0134] It is understandable that, in the architecture design of this AGV, sensors that are close in layout (physically close) and their corresponding sensor control units can be encapsulated in one package unit. Data is transmitted between each package unit through optical fiber. The main communication line uses optical fiber communication, and CAN or Ethernet communication is used within the package unit (depending on the communication protocol of its internal components). The internal wiring harness uses integrated wiring harness and is designed to resist interference (wrapped with a shielding layer and wired separately from the power line, and the outer shell is enclosed with a sheet metal box for anti-interference protection).

[0135] In this embodiment, the communication within the AGV system adopts a communication gateway that integrates Ethernet, CAN communication and fiber optic communication. By classifying and encapsulating data for each part of the vehicle, the vehicle's electromagnetic interference resistance is improved, wiring harnesses are made lighter, and the manufacturability and maintainability of the vehicle are enhanced.

[0136] The following describes the AGV to which the AGV control method provided in this application is applicable, using a specific embodiment. In this embodiment, the AGV achieves autonomous navigation, real-time risk warning, and proactive safety protection through multi-source sensor fusion and edge AI algorithms. It should be noted that edge AI algorithms refer to artificial intelligence algorithms that run directly on terminal devices (such as cameras, sensors, industrial controllers, etc., which in this application can refer to AGVs). These algorithms can process data without relying on cloud servers. Their core characteristic is localized computing, enabling real-time response, reduced latency, and reduced network transmission burden. In this application, the AGV is equipped with several sensors. As a terminal device, the AGV's deployed artificial intelligence algorithms, used to process the sensor data in real time to achieve control functions such as autonomous positioning, path planning, dynamic obstacle recognition, and risk prediction, belong to the category of edge AI algorithms.

[0137] For the AGVs to which it applies:

[0138] The hardware architecture consists of several parts: the perception layer, the computing layer, and the execution layer.

[0139] The perception layer consists of a depth camera, 3D laser, navigation laser, obstacle avoidance laser, barcode reader camera, ultrasonic sensor, PIO (Programmed Input / Output Optical Transmitter), vibration sensor, noise sensor, and smoke detector. The functions of each sensor are briefly described below:

[0140] Depth camera: Real-time 3D scene reconstruction, dynamic obstacle recognition;

[0141] 3D laser sensor: SLAM (Simultaneous Localization and Mapping) mapping and global path planning;

[0142] Navigation laser sensors and obstacle avoidance laser sensors: SLAM mapping and near-field emergency braking protection;

[0143] Barcode reader and PIO: coordinate information recognition, material identification verification and inventory management; in this embodiment, PIO can be used for traceability (e.g., during picking up goods, it interfaces with the PIO on the cargo handling machine to transmit data and obtain the attribute information of the goods on the cargo handling machine, such as quantity, location, placement time, etc.), information interaction (e.g., by interfacing with the PIO on the cargo handling machine, it verifies whether the position to which the AGV moves is the same as the position to be picked up), positioning (e.g., after the PIO on the AGV is aligned with the PIO on the cargo handling machine, goods can be picked up and placed, ensuring the stability of picking up and placing goods), etc.

[0144] Ultrasonic sensor: supplementary detection of low-lying obstacles;

[0145] Vibration detection sensors, noise detection sensors, and smoke detectors: for predicting equipment malfunctions and providing early warnings of environmental anomalies such as fires.

[0146] The computing layer consists of a main control unit, a security control unit, a sensor control unit, and a communication module. The functions of each part are as follows:

[0147] Main control unit: integrates CPU (Central Processing Unit), NPU (Neural Processing Unit), GPU (Graphics Processing Unit), etc., and can quickly process data streams such as images and point clouds acquired by the perception layer;

[0148] Safety control unit: Equipped with dual MCUs (Microcontroller Units), it receives, encrypts, and verifies critical safety data such as braking and emergency stopping, detects equipment faults in real time, records fault codes, and uploads them to the main control unit and host computer.

[0149] Sensor control unit: Receives data from PIO, photoelectric, ultrasonic, and other sensors;

[0150] Communication module: Gigabit industrial Ethernet, supporting dual-mode communication such as 5G / WiFi6, and supporting MESH networking; for example, the communication module may include an optical fiber communication module.

[0151] The execution layer consists of a drive system, a storage and retrieval mechanism, and a human-machine interaction system. The functions of each part are as follows:

[0152] Walking drive system: The vehicle's walking drive uses a one-to-two drive unit to drive the omnidirectional steering wheel assembly;

[0153] Storage and retrieval execution system: storage and retrieval execution mechanism drive, for example, can be a plate chain telescopic mechanism;

[0154] Human-computer interaction system: Three-color LED light strips are arranged on all four sides (360° visible), sound and light alarm, and communication interfaces such as WIFI, RS485 and RJ45 network ports are reserved, which are compatible with access control gates, fire sprinklers and other security equipment linkage.

[0155] To facilitate understanding of the AGV provided in the embodiments of this application, Figures 3(a) and 3(b) exemplarily show the structural schematic diagrams of an AGV provided in this application from different perspectives. The sensors installed in the AGV may include: a smoke detector 301, a 3D laser 302, a noise detection sensor 303, a first vibration detection sensor 304, a navigation laser 305, an obstacle avoidance laser 306, a first depth camera 307, a second depth camera 308, a first barcode reader 309, a driving recorder 310, a voice intercom module 311, a second vibration detection sensor 312, a second barcode reader 313, a third depth camera 314, and a fourth depth camera 315.

[0156] Furthermore, in this embodiment, the AGV system signal adopts a communication gateway that integrates Ethernet, CAN communication, and fiber optic communication. By classifying and encapsulating data into units for each part of the vehicle, the vehicle's electromagnetic interference resistance is improved, wiring harnesses are made lighter, and the manufacturability and maintainability of the vehicle are enhanced. The data communication relationships between the various encapsulation units are described in the relevant descriptions in the above embodiments and will not be repeated here.

[0157] The following is an example from the perspective of AGV processing logic:

[0158] The entire process of AGV operation encompasses several parts: system startup, multi-source data acquisition, data fusion and decision-making, execution-level drive control, and data closed-loop. The process details for each part are as follows:

[0159] The system startup process includes:

[0160] a. Hardware self-test:

[0161] (1) Iterate through all sensors to perform status verification;

[0162] (2) Confirm the computing resources of the computing layer;

[0163] b. Environmental modeling:

[0164] (1) Use 3D laser to perform environmental scanning of the current scene to construct an initial point cloud map (for navigation and positioning).

[0165] (2) Load platform task information (i.e., task information issued by the control terminal) and preset navigation path and electronic fence data;

[0166] The multi-source data acquisition process includes:

[0167] (1) Synchronous collection of environmental information:

[0168] Depth camera: Acquires depth information of the current scene environment and target objects, such as high-resolution color depth information;

[0169] Laser arrays (such as arrays of navigation laser sensors, arrays of 3D laser sensors, arrays of obstacle avoidance laser sensors, etc.): generate 360° point cloud data.

[0170] Environmental sensors (i.e., vibration detection sensors, noise detection sensors, smoke alarms, etc.): collect vibration signals, noise signals, and smoke alarm signals in real time;

[0171] (2) Data preprocessing:

[0172] The depth information and point cloud data acquired by the depth camera are denoised, and coordinate system transformation is performed (such as transforming the depth information and point cloud data to the same coordinate system).

[0173] Vibration and noise signals are decomposed and their features are extracted.

[0174] Extraction of frequency band signal data (such as extraction of frequency and amplitude).

[0175] The data fusion decision-making process includes:

[0176] (1) Feature extraction:

[0177] Visual channel: Extracting obstacle contour features from depth information acquired by the depth camera;

[0178] Laser Channel: For obstacle avoidance laser data, extract obstacle point cloud data; for navigation laser data, analyze motion trajectory.

[0179] Environmental channel: For the signals collected by the vibration detection sensor, characteristic frequency band signals are extracted; for the signals collected by the noise detection sensor, characteristic decibel audio signals are extracted; and smoke detection signals are detected by the smoke detector.

[0180] Code reader camera and PIO: Acquire AGV coordinate information and material-related information (code, outline, etc.).

[0181] (2) Data fusion: The data acquired by each sensor are dynamically fused and the fault probability value P (0~1) and risk assessment parameter Ф are generated by the AI ​​model. The fault probability value P is determined by the signals of the vibration detection sensor, the noise detection sensor and the smoke alarm. The risk assessment parameter Ф is determined by the state of the laser and the depth camera and the dynamic obstacle triggering signal detected by them, the material matching degree λ (0~1) and the fault probability value.

[0182] (3) Determination of risk and failure level:

[0183] For example, the risk level is divided into three levels (such as Level I, II, and III), determined by risk assessment parameters (the parameter values ​​of the risk assessment parameters refer to the risk assessment values ​​mentioned above). The trigger threshold can be preset and dynamically adjusted weekly based on actual on-site operational data using an AI model. The trigger thresholds for Levels I, II, and III can be set as α, β, and γ, respectively. For example, the situations corresponding to the abnormal levels are as follows:

[0184] When α < Ф ≤ β, the risk level is Level I: Local environmental anomalies, such as obstacles, material mismatches, or other situations that are not equipment malfunctions or will not cause harm to personnel or equipment (α < Ф ≤ β).

[0185] When β<Ф≤γ, the risk level is Level II: Fire / Intrusion / Equipment Anomaly Detection, such as vehicle fire, equipment collision, damage to main components, etc., where the equipment itself is abnormal and may cause harm to personnel or the equipment itself.

[0186] When Ф>γ, the risk level is Level III: group anomaly, multiple devices trigger Level II anomaly alarm (Ф>γ).

[0187] When Ф≤α, the characterization is normal.

[0188] Among them, the execution layer drives the control process:

[0189] The response strategy is determined based on the platform tasks and risk assessment parameters, as exemplified below:

[0190] a. When there are no abnormalities, the platform issues a trolley to store and retrieve goods, and the trolley operates normally;

[0191] b. Level I response: Initiate local obstacle avoidance and upload event logs;

[0192] c. Level II response: Activate the vehicle's audible and visual alarm and establish an electronic fence; if the vehicle catches fire, it will trigger the nearest fire hydrant to spray water for fire extinguishing;

[0193] d. Level III response triggers group path reconstruction, links sentry mode and workshop fire protection system, and activates the whole-site alarm.

[0194] The data closed-loop process includes:

[0195] (1) Data storage:

[0196] Main control MCU side: The storage part caches the raw data within the last 10 minutes (loop storage);

[0197] Cloud platform (such as the control terminal or the corresponding cloud service server mentioned above): Permanently stores abnormal event data;

[0198] (2) Model optimization:

[0199] The visual detection model is updated weekly through learning.

[0200] Optimize the strategy for generating comprehensive risk assessment parameters monthly;

[0201] (3) Fault tolerance mechanism:

[0202] a. Automatic degradation handling in case of single sensor failure: If one or more sensors fail, the obstacle avoidance mode switches to rule-based obstacle avoidance. In other words, when all sensors are functioning normally, the obstacle avoidance mode can be based on multi-sensor fusion data combined with obstacle avoidance algorithms to ensure more accurate and flexible obstacle avoidance, adapting to complex scenarios. However, when one or more sensors fail or malfunction, the obstacle avoidance mode can be rule-based. Rule-based obstacle avoidance can be understood as using pre-set, fixed obstacle avoidance logic / instructions, without relying on real-time complex sensor data, but only on basic post-fault information (or general safety rules).

[0203] b. Enable local decision caching when communication is interrupted. That is, if the communication link fails, the local cached data will be stored in SSD (Solid State Disk).

[0204] In addition, in this embodiment, the AGV is also equipped with a sentry security mode. The implementation of the sentry security mode involves mode switching logic and evidence chain construction process:

[0205] (1) Mode switching logic, for example, can be shown in Table 1 below:

[0206] Table 1

[0207] (2) Construction of the chain of evidence:

[0208] The AGV's dashcam video is overlaid with vibration / noise data timestamps. For example, the processing logic is as follows:

[0209] When an alarm event occurs, the dashcam extracts video frames and overlays timestamps, and simultaneously associates them with sensor data (such as vibration signal spectrum, noise decibel curve, depth camera obstacle trajectory, etc.) to generate an encrypted evidence package and upload it to the cloud platform.

[0210] In this embodiment, AI image recognition (hereinafter referred to as AI image recognition) and multi-source perception big model are involved: constructing a database of on-site working conditions, obstacles and materials, so that the AGV body can identify and locate goods and obstacles. Based on the fusion of multi-source information from multiple sensors, a map can be built and located in real time in an unknown environment, realizing global path planning and autonomous navigation, and achieving efficient, safe and intelligent autonomous operation.

[0211] This involves energy-saving sensing and scheduling: dynamically shutting down unnecessary sensors based on on-site operating conditions to reduce system power consumption;

[0212] It involves cross-device linkage control and sentry mode: the sentry security mode and multi-device linkage control, after the AGV recognizes the fire signal, it automatically unlocks the coordinates of the nearest fire hydrant to spray fire extinguishing;

[0213] An AGV anti-interference electrical system involving the integration of multiple communication modules: By classifying and encapsulating data into units for each part of the vehicle, the vehicle's electromagnetic interference resistance is improved, wiring harnesses are made lighter, and the manufacturability and maintainability of the vehicle are enhanced.

[0214] Corresponding to the above method embodiments, such as Figure 4 As shown in the figure, this application embodiment also provides an automated guided vehicle (AGV) control device, applied to AGVs, the device comprising:

[0215] The first analysis module 410 is configured to, during the movement of the AGV, in response to detecting a target dynamic obstacle, determine at least one first feature data of the target dynamic obstacle; based on the determined first feature data, perform matching analysis between the target dynamic obstacle and each dynamic obstacle recorded in a predetermined obstacle library by calling an AI model to obtain the target type of the target dynamic obstacle; and, based on the target type and the determined first feature data, predict the movement path of the target dynamic obstacle; wherein the at least one first feature data is: feature data obtained by analyzing the data about the target dynamic obstacle collected by each target sensor of the AGV respectively;

[0216] The second analysis module 420 is configured to, in response to meeting fault analysis conditions, determine at least one auxiliary feature data of the AGV, and analyze the fault probability value of the AGV based on the determined auxiliary feature data; the at least one auxiliary feature data is: feature data obtained by analyzing data collected by each auxiliary sensor of the AGV for the AGV or the current scene; and,

[0217] The third analysis module 430 is used to perform a motion risk assessment on the AGV based on the motion path of the currently existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV in response to the fulfillment of the risk assessment conditions, and to obtain a risk assessment value.

[0218] The response module 440 is used to execute a response strategy for the AGV that corresponds to the obtained risk assessment value.

[0219] Optionally, the movement process of the AGV is a movement process performed in response to a target task, wherein the target task is an inbound or outbound task executed after the goods are picked up.

[0220] The device further includes:

[0221] The fourth analysis module is used to acquire the first type of data collected by the target depth camera set on the AGV for the current goods, and to acquire the second type of data collected by the barcode reader camera set on the AGV for the current goods; and, based on the first type of data and the second type of data, to determine the goods matching degree between the current goods and the target goods; wherein, the target goods are the goods to be picked up set by the AGV.

[0222] The third analysis module 430 is specifically used to perform a motion risk assessment on the AGV in response to the fulfillment of risk assessment conditions, based on the movement path of the currently existing target dynamic obstacle, the current failure probability value, the current movement path of the AGV, and the cargo matching degree, and obtain a risk assessment value.

[0223] Optionally, the fourth analysis module determines the matching degree between the current goods and the target goods based on the first type of data and the second type of data, including:

[0224] An AI model is invoked to perform matching analysis between the cargo outline in the first type of data and the cargo outline template corresponding to the target cargo, to obtain a first analysis result. Then, the second type of data is matched with the identification information of the target cargo to obtain a second analysis result. Based on the first and second analysis results, the cargo matching degree between the current cargo and the target cargo is determined.

[0225] Optionally, the response module includes:

[0226] The risk level determination unit is used to analyze the risk level to which the obtained risk assessment value belongs, and to obtain the target risk level; different risk levels correspond to different risk assessment value ranges;

[0227] The strategy determination unit is used to determine the response strategy corresponding to the target risk level according to a pre-set correspondence between each risk level and the response strategy.

[0228] The strategy execution unit is used to execute a response strategy corresponding to the target risk level for the AGV.

[0229] Optionally, the device further includes:

[0230] The reporting module is used to report the target risk level to the control terminal of the AGV, so that the control terminal receives the target risk level reported by the AGV, and in response to receiving feedback from at least a specified number of AGVs in the current scenario that the target risk level is not lower than a specified level, issues a route reconstruction instruction to the AGVs in the current scenario.

[0231] The reconfiguration module is used to perform route reconfiguration processing in response to route reconfiguration commands received from the control terminal.

[0232] Optionally, the device further includes:

[0233] The identification module is used to identify whether any of the predetermined security trigger conditions are met during the movement of the AGV or when the AGV is stationary, in response to the activation of the sentry mode; wherein the designated sensors include one or more of the target sensors and the auxiliary sensors.

[0234] An execution module is used to execute a security response operation corresponding to the security triggering condition if the condition is met; wherein the security response operation includes control operation of the AGV itself or operation of preset linked security devices.

[0235] Optionally, the device further includes:

[0236] The first processing module is configured to, in response to the fulfillment of a mapping trigger condition, acquire second feature data about obstacles obtained by processing data collected from various target sensors, and perform three-dimensional mapping based on the acquired second feature data for use during the movement of the AGV; and / or,

[0237] The second processing module is used to respond to the route update conditions during the movement of the AGV, obtain the third feature data about the obstacles after processing the data collected by each target sensor, and perform route prediction for the AGV based on the obtained third feature data to obtain the updated route on which the AGV movement is based.

[0238] Optionally, each of the target sensors and auxiliary sensors and a sensor control unit are divided into a packaged unit. The components within each packaged unit communicate with each other via a controller area network (CAN) or Ethernet. Each packaged unit communicates with other components outside the packaged unit via fiber optic communication.

[0239] The sensor control unit within each packaged unit is used to perform specified analysis on the data collected by the sensors within the same packaged unit.

[0240] Based on the above method embodiments, this application also provides an AGV, such as... Figure 5 As shown, it includes:

[0241] Memory 501 is used to store computer programs;

[0242] The processor 502, when executing the program stored in the memory 501, implements any of the automated guided vehicles (AGV) control methods provided in the embodiments of this application.

[0243] Furthermore, the aforementioned AGV may also include a communication bus and / or a communication interface, with the processor 502, communication interface, and memory 501 communicating with each other via the communication bus.

[0244] The communication bus mentioned in the AGV above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0245] The communication interface is used for communication between the AGV and other devices.

[0246] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0247] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0248] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements any of the above-described automated guided vehicle (AGV) control methods.

[0249] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the automated guided vehicle (AGV) control methods described in the above embodiments.

[0250] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a solid-state drive (SSD), etc.

[0251] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0252] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, embodiments such as devices, storage media, and computer programs are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0253] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A control method for an automated guided vehicle (AGV), characterized in that, Applied to AGVs, the method includes: During the movement of the AGV, in response to the detection of a target dynamic obstacle, at least one first feature data of the target dynamic obstacle is determined; based on the determined first feature data, an AI model is invoked to perform matching analysis between the target dynamic obstacle and each dynamic obstacle recorded in a predetermined obstacle library to obtain the target type of the target dynamic obstacle; and based on the target type and the determined first feature data, the movement path of the target dynamic obstacle is predicted; the at least one first feature data is: feature data obtained by analyzing the data about the target dynamic obstacle collected by each target sensor of the AGV respectively; In response to meeting the fault analysis conditions, at least one auxiliary feature data of the AGV is determined, and the fault probability value of the AGV is analyzed based on the determined auxiliary feature data; the at least one auxiliary feature data is: feature data obtained by analyzing the data collected by each auxiliary sensor of the AGV for the AGV or the current scene respectively; as well as, In response to meeting the risk assessment conditions, a motion risk assessment is performed on the AGV based on the movement path of the currently existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV, to obtain a risk assessment value; The AGV will be subject to a response strategy corresponding to the obtained risk assessment value.

2. The method according to claim 1, characterized in that, The movement process of the AGV is a movement process performed in response to a target task, which is an inbound or outbound task executed after the goods are picked up. The method further includes: Acquire the first type of data collected by the target depth camera set on the AGV for the current goods, and acquire the second type of data collected by the barcode reader camera set on the AGV for the current goods; Based on the first type of data and the second type of data, the matching degree between the current goods and the target goods is determined; wherein, the target goods are the goods to be picked up by the AGV. The AGV's motion risk is assessed based on the current movement path of the target dynamic obstacle, the current failure probability value, and the AGV's current movement path, resulting in a risk assessment value, including: Based on the movement path of the existing target dynamic obstacle, the current failure probability value, the current movement path of the AGV, and the matching degree of the cargo, the movement risk of the AGV is assessed to obtain a risk assessment value.

3. The method according to claim 2, characterized in that, The step of determining the matching degree between the current goods and the target goods based on the first type of data and the second type of data includes: An AI model is invoked to perform matching analysis between the cargo outline in the first type of data and the cargo outline template corresponding to the target cargo, to obtain a first analysis result. Then, the second type of data is matched with the identification information of the target cargo to obtain a second analysis result. Based on the first and second analysis results, the cargo matching degree between the current cargo and the target cargo is determined.

4. The method according to claim 1 or 2, characterized in that, The response strategy implemented for the AGV corresponding to the obtained risk assessment value includes: The target risk level is obtained by analyzing the risk assessment values. Different risk levels correspond to different ranges of risk assessment values. Based on the pre-defined correspondence between various risk levels and response strategies, determine the response strategy corresponding to the target risk level; The AGV will be subject to a response strategy corresponding to the target risk level.

5. The method according to claim 4, characterized in that, The method further includes: The target risk level is reported to the control terminal of the AGV so that the control terminal receives the target risk level reported by the AGV, and in response to receiving at least a specified number of AGVs in the current scenario whose target risk level is not lower than a specified level, a route reconstruction instruction is issued to the AGVs in the current scenario. In response to receiving a route reconfiguration command from the control terminal, route reconfiguration processing is performed.

6. The method according to claim 2, characterized in that, Also includes: During the movement of the AGV or when the AGV is stationary, in response to the activation of the sentry mode, based on the data periodically collected by each designated sensor, it is identified whether any of the predetermined security trigger conditions are met; wherein, each designated sensor includes one or more of the target sensor and the auxiliary sensor; If the conditions are met, execute the security response operation corresponding to the security triggering condition; wherein, the security response operation includes control operation of the AGV itself or operation of preset linked security devices.

7. The method according to claim 1 or 2, characterized in that, The method further includes: In response to the fulfillment of the mapping trigger condition, second feature data about the obstacle is obtained by processing the data collected from each target sensor, and a three-dimensional map is constructed based on the obtained second feature data for use during the movement of the AGV; and / or, In response to the route update conditions during the movement of the AGV, the third feature data about the obstacles is obtained by processing the data collected from each target sensor. Based on the obtained third feature data, the AGV's route is predicted to obtain the updated route on which the AGV's movement is based.

8. The method according to claim 2, characterized in that, Each of the target sensors and auxiliary sensors, along with a sensor control unit, is divided into a packaged unit. The components within each packaged unit communicate with each other via a controller area network (CAN) or Ethernet. Each packaged unit communicates with other components outside the packaged unit via fiber optic communication. The sensor control unit within each packaged unit is used to perform specified analysis on the data collected by the sensors within the same packaged unit.

9. An automated guided vehicle (AGV) control device, characterized in that, Applied to AGVs, the device includes: A first analysis module is configured to, during the movement of the AGV, in response to the detection of a target dynamic obstacle, determine at least one first feature data of the target dynamic obstacle; based on the determined first feature data, perform matching analysis between the target dynamic obstacle and each dynamic obstacle recorded in a predetermined obstacle library by calling an AI model to obtain the target type of the target dynamic obstacle; and, based on the target type and the determined first feature data, predict the movement path of the target dynamic obstacle; wherein the at least one first feature data is: feature data obtained by analyzing the data about the target dynamic obstacle collected by each target sensor of the AGV respectively; The second analysis module is used to determine at least one auxiliary feature data of the AGV in response to meeting the fault analysis conditions, and to analyze the fault probability value of the AGV based on the determined auxiliary feature data; the at least one auxiliary feature data is: feature data obtained by analyzing the data collected by each auxiliary sensor of the AGV for the AGV or the current scene; and, The third analysis module is used to assess the motion risk of the AGV based on the motion path of the existing target dynamic obstacle, the current failure probability value, and the current movement path of the AGV in response to the fulfillment of the risk assessment conditions, and to obtain the risk assessment value. The response module is used to execute response strategies for the AGV that correspond to the obtained risk assessment values.

10. An AGV, characterized in that, include: Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-8.

Images