Cooperative control method for multiple AGVs (Automatic Guided Vehicles) for cooperatively carrying heavy objects

Abstract

The invention discloses a cooperative control method for multiple AGVs for cooperatively carrying heavy objects, and belongs to the technical field of automatic logistics, and the method comprises the steps: taking a high-dimensional environment description vector as a model condition, inputting the model condition into an improved conditional diffusion model which is introduced into a linkage constraint feature channel in a reverse denoising stage of a conventional conditional diffusion model, and obtaining a preliminary scheme; and then correcting the preliminary scheme by adopting a popular theory and a high-dimensional differential geometric constraint, and then ensuring cooperative control of multiple AGVs by adopting a dynamic environment perception and linkage state prediction mechanism, an online feedback and local path adjustment mechanism and a multi-heavy-load synchronous control strategy in control operation according to the corrected scheme. According to the method, the initial path is obtained based on the improved conditional diffusion model, the feasibility and the collaboration of a generation result are improved, the vehicle speed, the acceleration and the steering angle are controlled and optimized by applying a high-dimensional differential geometry dynamic control method, and the dynamic consistency of synchronous movement of multiple vehicles is ensured.

Description

Technical Field

[0001] The present invention relates to the field of automated logistics technology, and in particular to a collaborative control method for multiple AGVs that collaboratively carry heavy objects. Background Art

[0002] The traditional single-vehicle handling method relies on an automated guided vehicle (AGV) to carry materials along a fixed path. The path and control of the material handling by a single vehicle are usually simple, and it is only suitable for carrying small-sized and light-weight materials. In modern manufacturing and logistics scenarios, the handling of large-sized and heavy-weight materials places higher demands on the automation and flexibility of the production line than in existing general scenarios. Traditional handling methods usually rely on single-vehicle handling or manual operation along a fixed path, which is inefficient and difficult to adapt to complex and changing production environments. How to efficiently and safely complete the transportation of large-sized and heavy-weight materials in a limited space has become a problem that needs to be solved in the process of intelligent upgrading of production lines.

[0003] Therefore, in some heavy-load scenarios in the prior art, a multi-vehicle collaborative handling method is adopted to improve the handling efficiency and stability by sharing the weight and optimizing the path, so that multiple automatic guided vehicles work together to jointly undertake the handling task. For multi-vehicle collaborative systems, the existing path planning technology usually calculates the moving path of each vehicle based on a static algorithm to ensure that there is no conflict between vehicles and to carry out collaborative handling. In order to cope with dynamic environmental changes (such as obstacles), dynamic path adjustment and obstacle avoidance technologies are introduced in the prior art. These technologies can change the path in real time to avoid dynamic obstacles and ensure the smooth progress of the handling task. In multi-vehicle collaborative handling, the existing technology adopts a synchronous control strategy to ensure the consistency of the speed, position and direction of movement of each vehicle. The synchronous control is usually achieved through centralized control or distributed control strategies to maintain the stability of the handling. Error detection and correction mechanisms are also introduced in some existing systems, using sensors and feedback systems to monitor vehicle deviations and perform error correction to maintain the accuracy and stability of the handling.

[0004] Heavy-duty automated guided vehicles are gradually being introduced into the handling of large-sized and heavy-weight materials due to their high flexibility and automation characteristics. Compared with traditional single-vehicle handling, multi-vehicle collaborative operation has certain advantages in heavy-duty scenarios. It can improve handling efficiency and system stability by sharing weight, optimizing paths and coordinating movements. However, due to the particularity of large-sized materials, multi-vehicle coordinated handling still faces the following challenges in actual applications: Complexity of path planning: In the handling of large-sized materials, due to the large size and heavy weight of the materials, multiple vehicles need to carry and transport the materials in a collaborative manner. This method makes it impossible for a single vehicle to complete path planning independently, and traditional static path planning algorithms are difficult to adapt to the path planning between multiple vehicles in highly dynamic and highly complex handling scenarios that require comprehensive consideration of material size constraints, handling space limitations, and potential obstacles in dynamic environments.

[0005] Synchronicity and precision of multi-vehicle linkage: When transporting large-sized materials, synchronization and coordination between vehicles is more important. If there is a deviation in the speed, position or direction of the vehicle, it may cause instability in material handling and even cause a safety accident. In addition, the handling of large-sized materials requires vehicles to achieve high-precision synchronous movement in space, which places certain requirements on the control and coordination between vehicles.

[0006] Real-time adjustment capability in dynamic environments: Large-sized and heavy materials usually need to be flexibly transported between production lines, and dynamic changes in the production environment (such as obstacles, road congestion, temporary scheduling changes, etc.) require real-time response during the transportation process. If the multi-vehicle system cannot quickly adjust the path and movement strategy according to the real-time environment, it will lead to the failure of the transportation task or a significant decrease in transportation efficiency.

[0007] Error accumulation and feedback reliability: In the process of multi-vehicle collaboration, there will be deviations in position or path planning between vehicles due to sensor errors, communication delays, etc. Especially for large-sized materials, slight errors may be magnified, which will have a significant impact on the stability of transportation. Therefore, how to build a reliable error correction and feedback mechanism during the transportation process is the key to ensuring transportation stability and accuracy.

[0008] In summary, designing a path planning result that can generate reliability and coordination during the joint handling of multiple load automatic guided vehicles, and the processing problem of a collaborative optimization control strategy that ensures the synchronization of vehicles and the dynamic balance of load distribution during movement has become an urgent problem to be solved in the production process of enterprises. Summary of the invention

[0009] In view of this, the purpose of the present invention is to provide a collaborative control method for multiple AGVs that collaboratively transport heavy objects, so as to solve the problem that existing static methods based on optimization or planning are difficult to handle the linkage constraints of path planning between multiple vehicles in highly dynamic and highly complex transport scenarios under actual dynamic conditions.

[0010] To achieve the above objectives, the method of the present invention proposes an improved conditional diffusion model, introduces a linkage constraint feature channel, embeds geometric constraints, load restrictions and channel width restrictions, to improve the feasibility and coordination of the generated results. Specifically, the method includes the following steps: 1) The kinematic characteristic information of multiple AGVs to be handled, the real-time feasible domain information of the multiple AGVs and the materials they jointly handle, and the high-dimensional environment description vector of the real-time environment information collected by sensors and the real-time load information of multiple AGVs are input as model conditions into the improved conditional diffusion model to obtain a preliminary plan including the initial path and action sequence of the multiple AGVs to be jointly handled; The improved conditional diffusion model is to add geometric collision penalty, overload penalty and factory aisle restriction penalty items for materials to the reverse prediction equation of the reverse denoising stage of the traditional conditional diffusion model; 2) Using the popular theory to describe the initial path and action sequence as a whole, and correcting the result of the whole continuous description, the correction value of the speed, acceleration and steering angle of each AGV when the constraint conditions are met in the manifold space is obtained, and the motion command of each AGV is determined according to the correction value to perform collaborative motion control on multiple AGVs; The constraint conditions are that the constraint function of the coupled motion of the force balance between the multiple AGVs and the materials is established, and the high-dimensional differential geometry constraint mapping constraint including multiple constraints on the AGV speed balance, load distribution balance and material posture stability is established; 3) In the collaborative motion control of multiple AGVs, the future state is predicted in the short term based on the real-time detection results of the environmental sensors, and the safety risks in the operation of multiple AGVs are evaluated based on the short-term prediction results. Different adjustment strategies are implemented according to the different safety risk assessment results to ensure the safety of multiple AGV operations; The multi-load synchronization control law is designed based on the coupled dynamic equations, and the dynamic balance of load distribution is achieved by allocating vehicle outputs in real time.

[0011] The method of the present invention has the following advantages: the high-dimensional environment description vector of the method of the present invention not only includes real-time environmental information but also integrates vehicle dynamics characteristics, ensuring the comprehensive integration of dynamic sensor information, and the vector can characterize the collaborative handling state of multiple-carrier automatic guided vehicles and large-sized materials, external environmental interference and geometric feasible domain constraints, and can dynamically reflect the collaborative handling state of multiple vehicles and large-sized materials. In addition, in the present invention, by introducing a linkage constraint feature channel in the reverse denoising stage of the traditional conditional diffusion model, embedding geometric constraints, load restrictions and channel width restrictions, the feasibility and coordination of the generated results are significantly improved. The preliminary scheme obtained based on this improved conditional diffusion model has basic feasibility and global consistency, which is intended to lay the foundation for the subsequent refined control and correction of multi-vehicle collaborative handling. After obtaining the preliminary scheme, the method of the present invention optimizes the vehicle speed, acceleration and steering angle based on popular theory and high-dimensional differential geometry constraint mapping constraints to ensure the dynamic consistency of multi-vehicle synchronous motion. After obtaining the corrected control motion instructions, multiple vehicles are controlled based on this motion instruction, and the vehicle's motion status and environmental conditions are detected in real time during the control. Different strategies are formulated based on different detection results to ensure the safety of vehicle operation while ensuring the vehicle's operating efficiency.

[0012] Based on the above, in step 1), the real-time feasible domain information includes load limit, material size and the overall feasible area of ​​multiple AGVs and materials. When the load of multiple AGVs and the material size are within the feasible domain, the materials can be transported in a linked manner under the current geometry and load constraints.

[0013] In the present invention, geometric constraints are used to ensure that the subsequent construction of the environment description model can accurately describe the feasibility of synchronous transportation between large-sized materials and multiple vehicles.

[0014] Based on the above, in step 1), the reverse prediction equation of the reverse denoising stage of the improved conditional diffusion model is: ; ; in, represents the function that makes predictions based on the current state and conditional inputs in the traditional diffusion model, represents the penalty function for large-size material features, is a hyperparameter that balances the basic denoising objective of the model with the suppression strength of the large-size material linkage constraint feature channel. are the weight coefficients of the three penalty terms respectively, They correspond to the collision detection, load verification and channel width limit verification constraint functions respectively.

[0015] The method of the present invention adds a penalty function for large-size material characteristics in the reverse prediction equation of the reverse denoising stage of the conditional diffusion model. The penalty function includes penalty items of collision detection, load verification and channel width limit verification constraint functions, thereby reducing random disturbances that do not meet the requirements of large-size collaborative handling, ensuring the feasibility and coordination of the generated results of the preliminary plan, and the obtained preliminary plan has basic feasibility and global consistency.

[0016] Based on the above, in step 2), the process of correcting the obtained overall continuous description result includes: defining a constraint function describing the coupled motion on the manifold, constructing a high-dimensional differential geometry constraint mapping for differential geometry control, dynamically correcting the preliminary plan based on the high-dimensional differential geometry method, and solving the motion correction value of each vehicle when the constraint conditions are met on the manifold space.

[0017] Based on the above, in step 2), the preliminary plan is dynamically corrected based on the high-dimensional differential geometry method: a projection correction is made along the negative gradient direction of the constraint function within the manifold to offset the components that violate synchronization and coordination.

[0018] Based on the above, in step 2), the constraint function of the coupled motion is: ,in is a time-varying curve of a generalized coordinate vector that describes the initial path and action sequence in an overall continuous manner, represents the time derivative of generalized coordinates; High-dimensional differential geometry constraint mapping for , In order to balance the AGV speed, load distribution and material posture, multiple constraints are satisfied in the manifold space. When , the coupled motion of multiple vehicles and materials is in an ideal synchronous equilibrium state; The dynamic correction formula for the preliminary scheme based on high-dimensional differential geometry method is:

[0019] in, , , They are the corrected speed, acceleration and steering angular velocity of each AGV. represents the velocity vector calculated according to the preliminary scheme, represents the gradient of the constraint function; represents the corresponding Jacobian matrix The generalized inverse matrix or projection matrix of represents the step size factor, and It is an adjustable coefficient used to distribute the correction amplitude of acceleration and steering angle.

[0020] The method of the present invention utilizes high-dimensional differential geometry dynamic control to perform multi-vehicle synchronous motion correction. In a multi-vehicle and material handling scenario, a generalized coordinate vector is constructed based on manifold theory to continuously describe the force and motion state of the multi-vehicle linkage. Coupled motion constraints and dynamic mapping are used to optimize vehicle speed, acceleration and steering angle to ensure the dynamic consistency of the multi-vehicle synchronous motion.

[0021] Based on the above, in step 3), the environmental sensor is a multi-source environmental sensor, and the real-time detection result of the environmental sensor is generated by fusing the results of the multi-source environmental sensors through time alignment and weighted fusion to generate globally consistent dynamic environmental information.

[0022] The method of the present invention integrates multi-source sensor data, processes the differences in timing and data accuracy of different vehicles through timing alignment and weighted fusion, and obtains globally consistent real-time environmental information to ensure the accuracy of the obtained real-time environmental information, providing accurate real-time environmental basis for subsequent state prediction and interference prejudgment.

[0023] Based on the above, in step 3), short-term prediction of future status includes: Combine the dynamic environment information with the motion instructions obtained in step 2) to estimate the interference sources of obstacle movement, channel blockage and uneven load; Combining the dynamic environment information with the motion instructions obtained in step 2), the deviation between the actual state of each AGV at the same time and the ideal reference state is measured.

[0024] Based on the above, in step 3), the process of evaluating the safety risks in the operation of multiple AGVs according to the short-term prediction results includes: When the estimated result of the interference source is within the safe range and the measurement result of the deviation does not exceed the set threshold, the security risk level is 0; When the estimated result of the interference source reaches the warning level or the measurement result of the deviation exceeds the set threshold, the security risk level is 1; When the estimated result of the interference source reaches the emergency level of severe obstacles or emergencies, the security risk level is 2.

[0025] Based on the above, in step 3), different adjustment strategies are implemented according to different security risk assessment results: When the security risk level is 0, only regular online monitoring needs to be maintained; When the safety risk level is 1, a moderate correction of the speed or steering angle adjustment is triggered; When the safety risk level is 2, a more stringent response strategy of local route replanning or forced stopping is immediately implemented.

[0026] The present invention uses real-time detection during the operation of multiple vehicles and adds an interference assessment process to the prediction model to quantify interference sources such as obstacles and load distribution imbalance, and output risk indicators to detect in real time whether there are safety hazards in the operation of vehicles. According to the level of the risk situation, no intervention is made when there is no risk, simple intervention and adjustment is made when there is a small risk, and path re-planning or controlled stop strategies are implemented when there is a major safety risk, so as to ensure the safe operation of the vehicle while ensuring the operation efficiency of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of the collaborative control method of multiple AGVs for collaboratively transporting heavy objects according to the present invention.

[0028] Figure 2 It is a flow chart of the path and action sequence generation process of the present invention.

[0029] Figure 3 The present invention is a flow chart of using high-dimensional differential geometry dynamic control to perform multi-vehicle synchronous motion correction. DETAILED DESCRIPTION

[0030] The technical scheme of the present invention will be clearly and completely described below in conjunction with specific embodiments, but those skilled in the art should understand that the embodiments described below are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Embodiment of a collaborative control method for multiple AGVs that collaboratively carry heavy objects This embodiment proposes a collaborative optimization method for multiple automated guided vehicles for large-size and heavy-weight material handling based on a conditional diffusion model and high-dimensional differential geometry dynamic control to optimize the handling of large-size and heavy-weight materials. Through multi-vehicle collaborative optimization, efficient and safe path planning, synchronous control, real-time dynamic adjustment and error feedback mechanism are achieved to meet the handling needs in complex dynamic environments.

[0032] like Figure 1 As shown, the collaborative control method of this embodiment includes the following steps: Step S1: construct a high-dimensional environment description model for a multi-load automated guided vehicle collaborative handling scenario.

[0033] S11. Define the basic coordinates and entity set of the multi-load collaborative handling environment.

[0034] This step is mainly to determine the basic reference system and key entities for describing high-dimensional environments, providing a coordinate basis and entity recognition basis for subsequent model construction.

[0035] A global coordinate system is established within the factory operation area to identify and record information on all automated guided vehicles and materials to be transported.

[0036] 1. Set a fixed global coordinate system ,in is the coordinate origin, and Represents the reference axes of the global coordinate system in the plane and vertical directions respectively.

[0037] 2. Order Indicates The coordinate transformation matrix corresponding to the AGV , and order Indicates the main geometric center of large-size materials and their posture reference coordinates .

[0038] 3. Collect the maximum load capacity of each automated guided vehicle , the overall size of the material And pre-designed plant layout parameters , store it in the initial environment entity collection middle.

[0039] By using the global coordinate system and entity set, the relative positions and postures between different vehicles and materials can be calculated uniformly in subsequent steps to avoid interference between multiple coordinate definitions.

[0040] S12. Establish geometric constraint relationships between large-sized materials and multi-vehicle linkage.

[0041] This step mainly considers the size boundaries of large-sized materials, load restrictions, and geometric linkage constraints of multiple workshops on the premise of the above-mentioned basic coordinate system and entity set, and builds a geometric level foundation for high-dimensional environment description.

[0042] A unified geometric description is made for the moving boundary of each automated guided vehicle and the outer shape boundary of the material to form a linkage constraint function, and on this basis, the geometric representation of the feasible space of the environment is defined.

[0043] 1. Order Indicates Automated guided vehicles at the time The geometric area occupied by Indicates that large-sized materials are The geometric extent of .

[0044] 2. The overall feasible area of ​​multiple vehicles and materials is recorded as :

[0045] in, Indicates AGV relative to the global coordinate system Coordinate transformation operation, Represents the coordinate transformation operation of the material relative to the global coordinate system.

[0046] 3. Limit the load Material size Incorporate linkage constraints and define constraint functions :

[0047] when When , it means that large-sized materials can be moved in a coordinated manner under the current geometry and load constraints; when When , it means that large-sized materials can be moved in a coordinated manner under the current geometry and load constraints; when It means that the demand for coordinated handling cannot be met.

[0048] Through geometric constraint relationships, it is ensured that the subsequent construction of the environmental description model can accurately depict the feasibility of synchronous transportation between large-sized materials and multiple vehicles.

[0049] S13. Fusion of vehicle kinematics features and online sensor data.

[0050] Based on the vehicle's kinematic characteristics and information collected by sensors, dynamics and real-time environmental perception are incorporated into the environmental description to form a high-dimensional scene representation that can reflect the movement status of the vehicle and materials at the current moment.

[0051] The speed, acceleration and steering angle kinematic characteristics of each AGV are combined with the outputs of environmental sensors and load sensors to generate high-dimensional data that can be updated in real time: 1. Order and Respectively represent The velocity and acceleration vector of the AGV at time t, let Indicates the steering angle.

[0052] 2. Multi-sensor data is recorded as , including the laser radar ranging value and the load distribution value collected by the force sensor, which is directly expressed in matrix form as follows:

[0053] in Represents the environmental distance data vector output by the lidar, Represents the load distribution data vector output by the force sensor, and is the corresponding data dimension.

[0054] 3. Combine vehicle kinematic parameters with sensor information to define the moment The comprehensive state vector :

[0055] Established Order A comprehensive vector representing the real-time sensor data, position, and posture information corresponding to the material.

[0056] 4. Stack and concatenate the comprehensive state vectors of all automated guided vehicles and materials to obtain a high-dimensional state matrix :

[0057] The matrix contains the kinematic and sensor information of all vehicles and materials at the current moment.

[0058] S14. Generate a high-dimensional environment description vector.

[0059] Based on geometric constraints, vehicle-material kinematic characteristics and real-time sensor data, a high-dimensional environmental description vector of the overall collaborative handling scenario is output, providing a better and more comprehensive feature representation for the next stage of strategy generation and control.

[0060] That is, the feasible area obtained by integrating S12 in the global coordinate system The state matrix obtained with S13 , defines the environment description function and outputs the final vector that can be used for learning or planning.

[0061] 1. Define the environment description function , whose output is of dimension A vector of:

[0062] 2. In the function Internal, feasible area The geometric properties and state matrix of The vehicle and material multi-source information in is weighted and combined to form a mapping:

[0063] in, represents the geometric parameter extraction map, represents the kinematic and sensor fusion feature map, Indicates The weighting coefficients on the dimensions, Indicates the bias associated with a particular dimension feature.

[0064] 3. Dimension Output Stacking to obtain high-dimensional environment description vector This vector can characterize the multi-load AGV and large-size materials at a certain time. The collaborative transportation state, external environmental interference and geometric feasible domain constraints.

[0065] That is, through the above steps, the geometric constraints, vehicle kinematic characteristics and sensor data of large-size and heavy-weight materials in the multi-vehicle linkage scenario are comprehensively expressed to form a high-dimensional environment description vector.

[0066] Step S2: Generate a multi-vehicle collaborative initial transport plan based on the improved conditional diffusion model, such as Figure 2 shown.

[0067] S21. Construct the conditional input and initial random distribution of collaborative transportation.

[0068] This sub-step is mainly to clarify the conditional input and initial random distribution required by the improved conditional diffusion model, and to provide a feasible search space for the diffusion process.

[0069] Based on the high-dimensional environment description vector generated in step S1 As the model condition input, it sets the initial random noise distribution in the path and action sequence space.

[0070] 1. Order Indicates at time The output is a high-dimensional environment description vector that integrates the real-time feasible domain, kinematic characteristics and sensor information of multiple-carrier AGVs and large-size materials.

[0071] 2. Define the multi-vehicle collaborative transport path and action sequence to be generated as a high-dimensional variable ,in The total dimension of all positions, velocities, postures, and parameters required to perform actions within a discrete period of time or step.

[0072] 3. Let the initial random noise distribution be , using Gaussian distribution probability distribution form as the starting point of the diffusion process:

[0073] 4. As conditional information written into the conditional channel of the improved diffusion model, The prior constraints are:

[0074] S22. Design linkage constraint characteristic channels for large-size materials and establish improved diffusion equations.

[0075] In response to the special needs of large-size material handling, this application proposes to introduce a linkage constraint channel in the traditional diffusion model denoising process so that the generated path and action sequence meet the requirements of material size, load distribution and factory channel width.

[0076] In the process of forward diffusion and reverse denoising, a large-size material linkage constraint feature channel is embedded. By introducing penalty functions and constraint functions, random disturbances that do not meet the requirements of large-size collaborative handling are reduced.

[0077] 1. Let the forward stochastic evolution equation of the standard diffusion process be:

[0078] in Indicated in The retained weight of the state at the previous moment at the iteration, represents a noise vector sampled from a standard Gaussian distribution, is the total number of forward diffusion steps.

[0079] 2. In the reverse denoising stage, an improved linkage constraint feature channel is introduced .

[0080] The reverse prediction equation is:

[0081] in: represents the function that makes predictions based on current state and conditional inputs in traditional diffusion models; It represents the penalty function for the characteristics of large-sized materials, reflecting the constraints of material size, load distribution and factory channel width on the path and action sequence; is a hyperparameter that balances the basic denoising objective of the model and the suppression strength of the large-size material linkage constraint feature channel.

[0082] 3. In In the , set multiple sub-items for geometric collision penalty, overload penalty and factory aisle restriction penalty for large-sized materials:

[0083] in are the weight coefficients of the three penalty terms respectively, They correspond to the collision detection, load verification and channel width limit verification constraint functions respectively.

[0084] 4. By using the reverse denoising process The gradient is back-propagated to suppress random disturbances that do not conform to large-scale collaborative handling, thereby gradually approaching the generated results to a real and feasible solution that meets the needs of material linkage.

[0085] S23, perform diffusion denoising iteration and output the initial transport path and action sequence.

[0086] After introducing the large-size material linkage constraint feature channel, the denoising iteration of the improved diffusion model is completed to generate an initial multi-vehicle collaboration plan that meets the constraints.

[0087] The forward diffusion and reverse denoising steps are performed sequentially, and in the reverse denoising stage, the geometry and load penalty terms are incorporated into the gradient calculation of the prediction function to obtain the final path and action sequence.

[0088] 1. Using the initial distribution in S21 and conditional input , gradually diffuse forward and obtain the intermediate state .

[0089] 2. To start, perform the reverse denoising iteration:

[0090] 3. After After iterations, the final transport path and action sequence are obtained. , and rearrange or decode them into path coordinates and action instructions for multiple vehicles at discrete times or critical state moments:

[0091] in represents the reference trajectory point sequence of each automated guided vehicle during the entire handling process, Indicates the corresponding execution action sequence, function Used to map high-dimensional vectors into executable control instruction formats.

[0092] S24. Complete the output of the initial collaborative plan and prepare for subsequent dynamic control.

[0093] The preliminary scheme, which has been corrected by the linkage constraints of large-scale materials, is output and integrated to form the input basis for subsequent high-dimensional differential geometry dynamic control.

[0094] That is, the path and action sequence obtained in the previous step are structured and encapsulated and output to the next stage control algorithm.

[0095] 1. The multi-vehicle position, posture, speed, acceleration and steering information contained in are indexed and recorded in the multi-vehicle collaborative handling solution set:

[0096] in and Respectively represent Automated guided vehicles in the handling cycle The path points and action instruction sequence within.

[0097] 2. Output the preliminary plan for multi-vehicle collaboration It serves as input for subsequent high-dimensional differential geometry dynamic control and subsequent dynamic environment perception and re-planning.

[0098] 3. The preliminary plan has basic feasibility and global consistency under the linkage constraints of large-size materials, which lays the foundation for the subsequent refined control and correction of multi-vehicle collaborative handling.

[0099] Step S3: Use high-dimensional differential geometry dynamic control to perform multi-vehicle synchronous motion correction, such as Figure 3 shown.

[0100] S31. Establish a manifold space model of multiple vehicles and materials.

[0101] This step is mainly intended to express the coupled motion characteristics of multiple vehicles carrying large-sized materials in high-dimensional space, forming a manifold structure that can be used for differential geometry control.

[0102] The preliminary collaborative solution output from step S2 ( ) is regarded as a discretized approximate solution to the system state. On this basis, the generalized coordinates of the system are defined, and the manifold theory is used to continuously describe the overall feasible state of multiple vehicles and materials.

[0103] 1. Let the generalized coordinate vector of the system be:

[0104] in Indicates Automated guided vehicles at the time The pose (position and orientation) of Indicates key states such as the material's central posture and force distribution.

[0105] 2. The feasible domain is defined as the manifold ,in:

[0106] is the number of automated guided vehicles, is the bicycle state dimension, It is the material status dimension.

[0107] 3. The preliminary plan generated in step S2 Based on the path and action sequence in The preliminary discretization map of , as the input for the next step of coupled dynamics correction:

[0108] function It is used to decode the discrete instructions of multi-vehicle posture, velocity and acceleration into time-varying curves of generalized coordinate vectors.

[0109] S32. Construct coupled motion constraint function and introduce high-dimensional differential geometry constraint mapping.

[0110] Constrain the force and motion coupling relationship when multiple vehicles are linked to transport large-sized materials to ensure the coordination and consistency between vehicles and between vehicles and materials.

[0111] In the manifold The constraint functions describing the coupled motion are defined above, and the dynamic mapping for differential geometry control is constructed to achieve the correction of velocity, acceleration and steering angle.

[0112] 1. In the manifold In , define the coupled motion constraint function:

[0113] Reflects the force balance, geometric linkage and motion consistency between vehicles and materials. Represents the time derivative of generalized coordinates.

[0114] 2. Unify the kinematic equations of each AGV with the material dynamics equations:

[0115] in Indicates Control input for AGVs, It refers to external adjustment of the material's posture or mechanical state (such as simultaneous lifting of multiple contact points, force distribution during steering, etc.).

[0116] 3. Combine all subsystem equations on the manifold to obtain the overall evolution equation of the system:

[0117] in:

[0118] It is the overall dynamics mapping of multiple vehicles and materials coupled.

[0119] 4. Define high-dimensional differential geometry constraint mapping , including multiple constraints on vehicle speed balance, load distribution balance and material posture stability:

[0120] So that in the manifold At this time, the coupled motion of multiple vehicles and materials is in an ideal synchronous balance state.

[0121] S33, perform differential geometry dynamic correction and output the corrected speed, acceleration and steering angle.

[0122] The preliminary plan is dynamically corrected based on high-dimensional differential geometry methods to solve the motion correction value of each vehicle when the constraint conditions are met in the manifold space, so as to ensure the motion consistency of multi-vehicle synchronous transportation.

[0123] The motion constraint projection in the manifold is proposed and adopted to correct the speed, acceleration and steering angle of each vehicle, and the output satisfies corrected solution.

[0124] For the coupling constraint equation in step S32 and differential geometry constraints , at the current moment Correct the generalized coordinates and velocities under the preliminary scheme:

[0125] in: According to the preliminary plan The inferred velocity vector; represents the gradient of the constraint function; represents the corresponding Jacobian matrix The generalized inverse matrix or projection matrix of ; Represents the step size factor.

[0126] That is to say, it is equivalent to making a projection correction along the negative gradient direction of the constraint function within the manifold to offset the component that violates synchronization and coordination.

[0127] The acceleration and steering angle of each vehicle are corrected in the same way:

[0128] in and It is an adjustable coefficient used to distribute the correction amplitude of acceleration and steering angle.

[0129] After the above dynamic correction, The speed, acceleration and steering angle correction of multiple vehicles satisfying high-dimensional differential geometry coupling constraints are obtained to further ensure the consistency of vehicle motion.

[0130] S34, generating multi-vehicle synchronous motion instructions and outputting correction results.

[0131] The speed, acceleration and steering angle control quantities obtained after differential geometry dynamic correction are output as synchronous motion instructions actually executed by multiple vehicles, providing a more stable posture basis for the next stage of dynamic environment prediction and re-planning.

[0132] Based on the corrected kinematic parameters, a synchronization instruction set of multiple vehicles is uniformly constructed and encapsulated into a format that can be executed in an actual controller.

[0133] 1. According to the calibration result of step S33, define each AGV at time Synchronous motion instructions:

[0134] in It is used to convert the corrected speed, acceleration, steering angle and other quantities into the torque, speed or steering command required for the vehicle's underlying execution.

[0135] 2. Summarize all vehicles in the entire handling cycle The set of correction instructions within:

[0136] 3. Output the corrected multi-vehicle synchronous motion instructions and updated system status , ensuring the consistency of vehicle movement during linked handling, meeting the force and posture stability requirements of large-size materials, and providing a stable posture basis for the next stage of dynamic environment prediction and re-planning.

[0137] Step S4: predicting the state of collaborative handling and anticipating interference based on dynamic environment perception.

[0138] S41. Acquire and fuse multi-source sensor information.

[0139] During the collaborative handling of large-size materials, dynamic data from each automated guided vehicle and environmental sensors are collected and integrated in real time to provide real-time environmental basis for subsequent state prediction and interference anticipation.

[0140] That is, the output data of various sensors such as lidar, cameras, force sensors and ultrasonic ranging are integrated to carry out multi-dimensional fusion of interference sources (such as temporary obstacles, road congestion) and material force information.

[0141] 1. Let each vehicle be at time The collected multi-source sensor data constitutes a matrix:

[0142] in Indicates The environmental information vector collected by the automatic guided vehicle includes factors such as obstacle location, channel congestion, and mechanical load changes.

[0143] 2. For the time synchronization and network delay of each automated guided vehicle workshop, the timing alignment and weighted fusion method are used, and the fusion result is recorded as:

[0144] It is used to process the differences in timing and data accuracy among different vehicles and output globally consistent environmental perception information.

[0145] S42, performing linkage transport state prediction and interference prediction.

[0146] Based on the real-time fusion of environmental information, short-term predictions are made on the future status of vehicles and materials in the collaborative handling process, potential interference sources are identified and their impact on the movement of multiple vehicles is estimated.

[0147] The synchronous motion command corrected in step S3 is combined with the environmental perception data obtained by fusion in step S41, and a rolling prediction of the recent state is performed through the dynamic model, and the interference source is estimated and calibrated.

[0148] 1. Define the rolling forecast window , based on the dynamic equations of vehicle and material For the future moment The system status is predicted:

[0149] in It represents the prediction function based on the multi-vehicle and material coupling model, which comprehensively considers the dynamic influence of the corrected control input and environmental information.

[0150] 2. Estimate the interference sources such as obstacle movement, channel blockage and uneven load, and define the interference evaluation function:

[0151] It is used to quantify the risk of each interference source based on the current environment perception and future state prediction, and output the impact index on the linkage handling. .

[0152] 3. Feedback the prediction status and interference assessment results to the subsequent online re-planning mechanism and dynamic control link, so that the system can timely trigger local path adjustment or control parameter correction when potential risks are found.

[0153] Step S5: construct an online feedback correction mechanism for multi-vehicle collaboration and perform local path adjustment.

[0154] S51, monitor vehicle state deviation in real time and construct an online feedback error function.

[0155] In this sub-step, the system continuously obtains the multi-vehicle synchronous motion instructions output in step S3 Combined with the environmental perception data generated in step S4 For each vehicle at time The actual status With the ideal reference state The deviation between them is measured and the feedback error function is defined as:

[0156] in It represents the norm operation in a properly weighted space (combining multi-dimensional information of vehicle speed, acceleration, and steering angle).

[0157] If any Exceeding the set threshold , or detects obvious load imbalance or speed exceeding the limit, the system triggers the online feedback correction process, thereby fine-tuning the vehicle's motion parameters during operation and returning the error to within the tolerance range.

[0158] S52: Dynamically trigger and determine the abnormal level based on the interference prediction result.

[0159] In this sub-step, the system receives the interference assessment given in step S4 And the predicted status and compare it with the real-time error Perform a comprehensive evaluation. A grading function can be defined:

[0160] when When the system is in operation, it only needs to maintain routine online monitoring; when When the vehicle is moving in a certain direction, a moderate correction is triggered, such as a small range speed or steering angle adjustment; when , it indicates that there is serious interference or potential danger, and more mandatory response strategies such as local route replanning or forced stopping are required immediately.

[0161] S53. Constraints and optimization solutions for local path replanning.

[0162] Once it is determined that the interference has reached a level that requires path adjustment ( or 2 , then the local path replanning algorithm is started in this sub-step to ensure that the vehicle avoids or reduces the current interference as much as possible while maintaining the established global goal. The optimization objective function of the local path search is proposed and defined as:

[0163] in Indicates that multiple vehicles are in a local area space-time trajectory within; Used to measure the cost of obstacle avoidance; Measures the degree of deviation from the global reference path; It represents the cost such as energy consumption or mechanical load; is the weight coefficient. By performing minimization in the local feasible region Optimization algorithms (such as those based on fast search trees, Improved algorithms or nonlinear optimizers are both possible) to obtain the optimal local driving route for the vehicle under the current interference conditions.

[0164] S54, output the recalibrated motion command and close the loop with the online feedback.

[0165] Finally, this sub-step performs differential geometry dynamic correction on the local replanning results (refer to the coupling constraints in step S3 for the same reason) to obtain the corrected speed, acceleration and steering angle control values. . At the same time, the new reference state Sent to each vehicle's underlying controller:

[0166] in is the optimal trajectory discrete point after local path optimization, It is a control signal mapping function used to convert posture and kinematic information into vehicle execution instructions.

[0167] Through this instruction update, the system forms a closed loop with the aforementioned online feedback correction mechanism: if deviation or interference is detected again, the cycle process of S51 to S54 will continue to be executed, thereby achieving timely response and flexible adjustment to various types of interference on the entire transportation line.

[0168] Step S6: Execute high-precision handling actions based on the multi-load synchronization control strategy.

[0169] S61. Establish the coupled dynamic equations for synchronous control of multiple loads.

[0170] In this substep, the overall dynamic equation of the system is defined to uniformly describe the coupled dynamic relationship between multiple vehicles and materials.

[0171] make Representing the generalized coordinates of multiple vehicles and materials (refer to the manifold definition in step S3), the dynamic equation with coupling terms can be written:

[0172] in is the mass-inertia matrix, are the Coriolis force and centrifugal force terms, is the gravity and other generalized force vectors related to posture, It represents the total control force / torque vector (including driving, braking and steering torque, etc.) output by multiple vehicles. Through this equation, the coupled motion of vehicles when carrying large-sized materials can be described as a whole, laying the foundation for the design data of the multi-load synchronization control law in the next step.

[0173] S62. Design multi-load synchronization control law.

[0174] This substep uses a synchronous control strategy for the aforementioned coupled dynamic equations to ensure that the force, speed, and posture adjustments of each vehicle on the material remain consistent. represents the expected system trajectory (combined with the local path adjustment and online feedback results of step S5), while the actual system state is . Then the synchronization error is defined as:

[0175] In order to achieve fast and stable error convergence, a multi-load synchronization control law is proposed and designed:

[0176] in and is a positive definite gain matrix, corresponding to the feedback items of position / attitude error and velocity error respectively; and Used to compensate for coupling and gravity in dynamics.

[0177] Through the above control law, each vehicle will tend to follow the desired trajectory under the action of the resultant force / torque, achieving synchronous load-bearing and precise posture maintenance.

[0178] S63. Distribute vehicle output in real time and maintain multiple load force balance.

[0179] In this sub-step, to prevent extreme load distribution or inconsistent motion between vehicles, the total control force Perform real-time decomposition to obtain individual outputs for each vehicle :

[0180] in For the The current load proportionality factor of the vehicle satisfies .

[0181] The coefficient can be dynamically adjusted according to the load measurement of the real-time sensor, the vehicle dynamic performance, the steering demand, etc., so that the total force of the actual output is still consistent with the calculated by the synchronous control law. Keep consistent, thereby maintaining force balance between multiple loads and avoiding overloading or unbalanced dragging of the bicycle.

[0182] S64. Issue high-precision synchronous action instructions and execute the transport action.

[0183] In this sub-step, the aforementioned multiple load control laws and vehicle output allocation results are mapped to the underlying control system to form directly executable high-precision action instructions.

[0184] make Indicates The actual execution command vector of the vehicle (including the drive motor speed and steering wheel steering angle) is:

[0185] in represents the transformation function that converts the target force / torque into actual motor speed, steering angle and possible braking command.

[0186] That is, in the end, each vehicle The handling actions are performed synchronously, and under the constraints of multiple load synchronization control strategies, the precise loading and stable movement of large-sized materials are continuously maintained, thereby achieving the overall high-precision handling goal.

[0187] Compared with the prior art, the method of this embodiment has the following advantages: 1. This embodiment is based on the construction method of the global coordinate system and entity set, defines the geometric constraint relationship and dynamic state representation between multiple vehicles and large-sized materials, and forms a unified high-dimensional environment description vector. A real-time fusion strategy of multi-source sensor data (such as lidar, force sensor) is introduced to generate a comprehensive state matrix in conjunction with the vehicle kinematic characteristics. Compared with the prior art, the environmental modeling in multi-vehicle collaborative handling is usually limited to geometric constraints or simple kinematic features, and lacks comprehensive integration of dynamic sensor information. The method of this embodiment constructs a real-time updated high-dimensional environment description vector through the global coordinate system, multi-source sensor data fusion (such as lidar, force sensor) and vehicle kinematic characteristics, which can dynamically reflect the collaborative handling status of multiple vehicles and large-sized materials, and ensure the comprehensiveness and real-time nature of the high-dimensional environment description.

[0188] 2. This embodiment adopts an improved traditional diffusion model, introduces a linkage constraint feature channel, and dynamically adjusts the diffusion process through penalty functions (geometric collision, overload, channel width limit). Define a diffusion denoising mechanism for constraint optimization to gradually guide random disturbances to paths and action sequences that meet the requirements of collaborative handling of large-sized materials. Compared with the prior art, the path and action sequence generation often adopts static methods based on optimization or planning, which are difficult to handle complex linkage constraints, especially in large-sized material handling scenarios. The method of this embodiment proposes an improved conditional diffusion model, introduces a linkage constraint feature channel, and embeds geometric constraints, load limits, and channel width limits, which significantly improves the feasibility and coordination of the generated results.

[0189] 3. In the scenario of multiple vehicles and material handling, this embodiment constructs a generalized coordinate vector based on manifold theory to continuously describe the force and motion state of multiple vehicle linkage. By utilizing coupled motion constraints and dynamic mapping, the vehicle speed, acceleration and steering angle are optimized to ensure the dynamic consistency of the synchronous motion of multiple vehicles. Compared with the existing technology of multi-vehicle collaborative control, which mostly adopts a simple feedback control strategy and is difficult to fully characterize the complex force coupling relationship of multiple workshops, this embodiment makes the motion state of multiple vehicles and materials continuous through manifold theory, and combines coupling constraints and dynamic mapping to achieve precise correction of speed, acceleration and steering angle, ensuring the dynamic consistency of the synchronous motion of multiple vehicles.

[0190] 4. This embodiment integrates multi-source sensor data, and obtains globally consistent real-time environmental information through time alignment and weighted fusion. An interference assessment function is added to the rolling prediction model to quantify interference sources such as obstacles and load distribution imbalance, and output risk indicators. Compared with the existing technology, environmental perception is mostly limited to single sensor data, and the interference source identification and state prediction capabilities are limited. The method of this embodiment fuses multi-source sensor data, generates globally consistent dynamic environmental information through time alignment and weighted fusion, and cooperates with the rolling prediction model and interference assessment function to achieve accurate quantification and prediction of interference sources such as obstacles and load imbalance.

[0191] 5. This embodiment monitors the error between the vehicle and the ideal state in real time, defines a feedback error function, and dynamically triggers a multi-level interference response. A local path optimization algorithm is introduced to perform local path replanning based on space-time constraints while ensuring the consistency of the global goal. Compared with the existing technology, the path adjustment is mostly performed offline, the response speed is slow, and it is difficult to dynamically adapt to environmental changes. The method of this embodiment uses real-time error monitoring and a hierarchical feedback mechanism to dynamically trigger local path adjustments according to the interference level, and uses an optimization algorithm to quickly generate a local optimal path, ensuring that the system has high flexibility and robustness.

[0192] 6. This embodiment proposes a coupled dynamic model for multi-vehicle collaboration, which dynamically couples the movement between vehicles with the force of large-sized materials. Design a synchronous control law and an output distribution mechanism to ensure that the vehicles maintain synchronization and dynamic balance of load distribution during movement. Compared with the prior art, the mechanical distribution of multiple workshops mostly adopts empirical rules or equal distribution strategies, which is difficult to cope with complex dynamic force scenarios. The method of this embodiment designs multiple load synchronization control laws based on coupled dynamic equations, and achieves dynamic balance of load distribution by distributing vehicle outputs in real time, avoiding the problem of single-vehicle overload or drag imbalance.

[0193] 7. This embodiment takes into account the special optimization of large-size material handling. Compared with the existing technology, which fails to fully consider special constraints such as size boundaries, load distribution and factory channel width of large-size materials, the method of this embodiment introduces multiple constraints in the diffusion model and differential geometry control, especially geometric collision penalties, overload verification and factory channel restriction verification, and specifically optimizes path generation and synchronous control.

[0194] This method also has the following commercial value: 1. Wide application range and large market demand: It can be widely used in multiple scenarios such as smart manufacturing, logistics warehousing, port loading and unloading, especially in scenarios that require collaborative processing of large-size and heavy-weight materials, such as automobile manufacturing, aerospace assembly, and semiconductor equipment handling.

[0195] With the in-depth advancement of Industry 4.0, the market demand for multi-vehicle collaborative handling and automated logistics systems continues to grow. The method of the present invention empowers traditional logistics and manufacturing industries through intelligent and precise collaborative control.

[0196] 2. Significantly reduce operating costs: Through efficient path planning, dynamic control and real-time feedback, this technology reduces energy waste and the risk of misoperation in multi-vehicle collaborative transportation.

[0197] Improve equipment utilization, reduce reliance on manual intervention, and significantly reduce the labor and time costs of enterprises during warehousing and transportation.

[0198] 3. Improve safety and reliability: Introduce multi-source data fusion and dynamic interference prediction to perceive environmental changes in real time, actively avoid potential risks, and improve the safety of the handling process.

[0199] Special optimization for large-sized materials reduces the risk of accidents such as falling and collision during material handling and improves the overall reliability of the handling system.

[0200] 4. Meet the needs of complex scenarios and improve system flexibility: By linking constrained feature channels with high-dimensional differential geometry dynamic control, it supports refined operations of multi-vehicle collaborative handling in complex scenarios, especially the high-precision handling of large-size materials.

[0201] Dynamically adjust paths and control strategies to quickly adapt to dynamic demands such as production line changes and environmental changes, and enhance the flexibility of logistics and handling systems.

[0202] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for cooperatively controlling multiple AGVs for cooperatively transporting heavy objects, characterized in that: The steps include: 1) The kinematic characteristic information of multiple AGVs to be handled, the real-time feasible domain information of the multiple AGVs and the materials they jointly handle, and the high-dimensional environment description vector of the real-time environment information collected by sensors and the real-time load information of multiple AGVs are input as model conditions into the improved conditional diffusion model to obtain a preliminary plan including the initial path and action sequence of the multiple AGVs to be jointly handled; The improved conditional diffusion model is to add geometric collision penalty, overload penalty and factory aisle restriction penalty items for materials to the reverse prediction equation of the reverse denoising stage of the traditional conditional diffusion model; 2) Using the popular theory to describe the initial path and action sequence as a whole, and correcting the result of the whole continuous description, the correction value of the speed, acceleration and steering angle of each AGV when the constraint conditions are met in the manifold space is obtained, and the motion command of each AGV is determined according to the correction value to perform collaborative motion control on multiple AGVs; The constraint conditions are that the constraint function of the coupled motion of the force balance between the multiple AGVs and the materials is established, and the high-dimensional differential geometry constraint mapping constraint including multiple constraints on the AGV speed balance, load distribution balance and material posture stability is established; 3) In the collaborative motion control of multiple AGVs, the future state is predicted in the short term based on the real-time detection results of the environmental sensors, and the safety risks in the operation of multiple AGVs are evaluated based on the short-term prediction results. Different adjustment strategies are implemented according to the different safety risk assessment results to ensure the safety of multiple AGV operations; The multi-load synchronization control law is designed based on the coupled dynamic equations, and the dynamic balance of load distribution is achieved by allocating vehicle outputs in real time.

2. The collaborative control method of multiple AGVs according to claim 1, characterized in that: In step 1), the real-time feasible domain information includes load limit, material size, and the overall feasible area of ​​multiple AGVs and materials. When the load of multiple AGVs and the material size are within the feasible domain, the materials can be transported in a linked manner under the current geometry and load constraints.

3. The collaborative control method of multiple AGVs according to claim 1, characterized in that: In step 1), the reverse prediction equation of the reverse denoising stage of the improved conditional diffusion model is: ； ； in, represents the function that makes predictions based on the current state and conditional inputs in the traditional diffusion model, represents the penalty function for large-size material features, is a hyperparameter that balances the basic denoising objective of the model with the suppression strength of the large-size material linkage constraint feature channel. are the weight coefficients of the three penalty terms respectively, They correspond to the collision detection, load verification and channel width limit verification constraint functions respectively.

4. The collaborative control method of multiple AGVs according to claim 1, characterized in that: In step 2), the process of correcting the obtained overall continuous description result includes: defining a constraint function describing the coupled motion on the manifold, constructing a high-dimensional differential geometry constraint mapping for differential geometry control, dynamically correcting the preliminary plan based on the high-dimensional differential geometry method, and solving the motion correction value of each vehicle when the constraint conditions are met on the manifold space.

5. The collaborative control method of multiple AGVs according to claim 4, characterized in that: In step 2), the preliminary plan is dynamically corrected based on the high-dimensional differential geometry method: a projection correction is made along the negative gradient direction of the constraint function within the manifold to offset the components that violate synchronization and coordination.

6. The collaborative control method of multiple AGVs according to claim 5, characterized in that: In step 2), the constraint function of the coupled motion is: ,in is a time-varying curve of a generalized coordinate vector that describes the initial path and action sequence in an overall continuous manner, represents the time derivative of generalized coordinates; High-dimensional differential geometry constraint mapping for , In order to balance the AGV speed, load distribution and material posture, multiple constraints are satisfied in the manifold space. When , the coupled motion of multiple vehicles and materials is in an ideal synchronous equilibrium state; The dynamic correction formula for the preliminary scheme based on high-dimensional differential geometry method is: ； in, , , They are the corrected speed, acceleration and steering angular velocity of each AGV. represents the velocity vector calculated according to the preliminary scheme, represents the gradient of the constraint function; represents the corresponding Jacobian matrix The generalized inverse matrix or projection matrix of represents the step size factor, and It is an adjustable coefficient used to distribute the correction amplitude of acceleration and steering angle.

7. The collaborative control method of multiple AGVs according to claim 1, characterized in that: In step 3), the environmental sensor is a multi-source environmental sensor, and the real-time detection result of the environmental sensor is generated by fusing the results of the multi-source environmental sensors through time alignment and weighted fusion to generate globally consistent dynamic environmental information.

8. The collaborative control method of multiple AGVs according to claim 7, characterized in that: In step 3), short-term predictions of future states include: Combine the dynamic environment information with the motion instructions obtained in step 2) to estimate the interference sources of obstacle movement, channel blockage and uneven load; Combining the dynamic environment information with the motion instructions obtained in step 2), the deviation between the actual state of each AGV at the same time and the ideal reference state is measured.

9. The method for cooperative control of multiple AGVs according to claim 8, characterized in that: In step 3), the process of evaluating the safety risks during the operation of multiple AGVs based on the short-term prediction results includes: When the estimated result of the interference source is within the safe range and the measurement result of the deviation does not exceed the set threshold, the security risk level is 0; When the estimated result of the interference source reaches the warning level or the measurement result of the deviation exceeds the set threshold, the security risk level is 1; When the estimated result of the interference source reaches the emergency level of severe obstacles or emergencies, the security risk level is 2.

10. The collaborative control method of multiple AGVs according to claim 9, characterized in that: In step 3), different adjustment strategies are implemented according to different security risk assessment results: When the security risk level is 0, only regular online monitoring needs to be maintained; When the safety risk level is 1, a moderate correction of the speed or steering angle adjustment is triggered; When the safety risk level is 2, a more stringent response strategy of local route replanning or forced stopping is immediately implemented.

Images