Dynamic precision compensation method and compensation system for multi-AGV collaborative operation

Abstract

The invention relates to an AGV (Automatic Guided Vehicle), in particular to a dynamic precision compensation method and compensation system for cooperative operation of multiple AGVs. The method comprises the following steps: a plurality of AGVs construct a shared high-precision map through laser radar scanning; after the AGV-A detects ground vibration, vibration waveform characteristics are broadcasted through millimeter waves, and after the AGV-B receives data, torque output of a servo motor is adjusted in advance according to preset time; the servo motor torque output is adjusted according to the final control quantity, and the final control quantity is obtained through the following steps that when the multiple AGVs enter cooperative deviation rectification, distributed weighted PID control is adopted, and the combined control output quantity is obtained; for errors of ground vibration transmission, the compensation amount is predicted through a mechanical model; and the final control quantity is obtained by combining the control output quantity and the compensation quantity. According to the method, the multi-machine cooperative positioning precision can be remarkably improved, real-time correction of ultra-low delay is achieved, and the complex environment adaptability is enhanced.

Description

TECHNICAL FIELD

[0001] The application relates to an AGV, in particular to a dynamic precision compensation method and system for multi-AGV cooperative operation. BACKGROUND

[0002] AGV is an unmanned vehicle for automatic material handling based on navigation and control system, which is widely used in warehouse logistics, manufacturing industry and port terminal and the like. The core technologies include navigation technology, motion control, communication system and energy management and the like. Multi-AGV cooperative operation refers to efficient cooperation of multiple AGVs in a shared environment, and the key technologies include:

[0003] (1) task allocation and path planning: based on centralized (such as server scheduling) or distributed (such as reinforcement learning) job allocation algorithm, dynamic collision avoidance strategy is adopted to avoid deadlock and path conflict.

[0004] (2) group positioning and map sharing: joint SLAM technology is adopted to construct a globally consistent map through feature point fusion; at the same time, UWB or visual label assisted relative positioning is adopted to reduce cumulative error.

[0005] (3) communication cooperation: low-delay communication protocol (such as TSN time-sensitive network) is adopted to ensure real-time synchronization of instructions.

[0006] However, the prior art has the following problems: low cooperative positioning accuracy, long deviation correction response time and poor adaptability to complex environment. SUMMARY

[0007] To solve the above problems, the application provides a dynamic precision compensation method for multi-AGV cooperative operation, and the specific technical scheme is as follows:

[0008] A dynamic precision compensation method for multi-AGV cooperative operation, comprising the following steps:

[0009] Multiple AGVs construct a shared high-precision map through laser radar scanning;

[0010] When AGV-A detects ground vibration, it broadcasts the vibration waveform characteristics through millimeter wave, and when AGV-B receives the data, it adjusts the servo motor torque output in advance according to the predetermined time;

[0011] The adjustment of the servo motor torque output is based on the final control quantity, and the acquisition of the final control quantity comprises the following steps:

[0012] When multiple AGVs enter cooperative deviation correction, distributed weighted PID control is adopted to obtain joint control output;

[0013] For the error of ground vibration transmission, the compensation amount is predicted through a mechanical model.

[0014] Obtain final control quantity: joint control output quantity + compensation quantity.

[0015] Preferably, the predetermined time is 20-70 ms.

[0016] Preferably, when the broadcast vibration waveform features, the additional error caused by mutual interference is reduced by a group error propagation model, wherein the vibration transmission equation is:

[0017] [K]{δ}+[C]{δ′}+[M]{δ"}={F(t)};

[0018] In the formula: K is the ground stiffness matrix, C is the damping matrix, M is the AGV mass matrix, δ is the AGV pose error vector, and F(t) is the external excitation force.

[0019] Preferably, the distributed weighted PID control includes:

[0020]

[0021] In the formula, u joint (t) is the joint control output; e j (t) is the real-time error of the jth AGV; K p，j , K i，j , K d，j are the PID parameters of the jth AGV, respectively; and N is the number of AGVs participating in cooperation.

[0022] Further,

[0023] K p，j =w j ·K p，base ;

[0024]

[0025] In the formula, K p，base is the basic gain parameter of the proportional term in the PID control, used to determine the initial response strength of the system to the error, RMSE j is the historical root mean square error of the jth AGV; and ω j is the weight.

[0026] Further, the weight ω i :

[0027]

[0028] In the formula, α, β, γ, δ are adjustable parameters, d i is the distance, Stabilityi is the environmental stability evaluation, α is the distance attenuation factor, SNR is the strength of the communication signal, and Ehistory is the average error of the data source in the past period of time.

[0029] Further, the predicted compensation amount:

[0030] Δu ff = [H(ω)]·F(t-τ);

[0031] In the formula, Δuff is the feedforward compensation amount; H(ω) is a frequency response function matrix;

[0032] H ik (ω) = δ i (ω) / F k (ω), the influence of the kth AGV excitation on the ith AGV;

[0033] F(t-τ) is the excitation force after the time delay τ.

[0034] A dynamic precision compensation system for multi-AGV collaborative operation, comprising a perception layer, including a dual-frequency RTK-GNSS, a distributed pressure sensor array, and a millimeter wave communication module mounted on an AGV; a control layer for allocating communication resources according to the task urgency.

[0035] Compared with the prior art, the present application has the following beneficial effects:

[0036] (1) Significantly improve the multi-machine collaborative positioning accuracy, through a dynamic weight fusion algorithm and vibration feedforward compensation, the collaborative positioning accuracy is improved to within ±1mm (more than 90% improvement).

[0037] (2) Realize ultra-low delay real-time correction, adopt distributed edge computing and millimeter wave direct communication, the correction response time is compressed to <50ms.

[0038] (3) Enhance the adaptability to complex environment, laser / visual / millimeter wave complement each other, still maintain ±2mm accuracy when light suddenly changes (50-2000lux) or is temporarily blocked. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is the system framework diagram of the present application;

[0040] Figure 2 is a schematic diagram of a group error propagation mathematical model;

[0041] Figure 3 is a dynamic correction control flowchart;

[0042] Figure 4 is a block diagram of an automobile parts warehouse embodiment. DETAILED DESCRIPTION

[0043] The present application will be further described in conjunction with the drawings.

[0044] As Figures 1 to 4 shown, a dynamic precision compensation method for multi-AGV collaborative work includes the following steps:

[0045] Multiple AGVs construct a shared high-precision map through laser radar scanning;

[0046] When AGV-A detects ground vibration, it broadcasts the vibration waveform characteristics through millimeter wave. When AGV-B receives the data, it adjusts the servo motor torque output in advance according to the predetermined time;

[0047] Wherein, the adjustment of servo motor torque output is adjusted according to the final control quantity, and the acquisition of the final control quantity includes the following steps:

[0048] When multiple AGVs enter collaborative correction, distributed weighted PID control is adopted to obtain joint control output;

[0049] For the error of ground vibration transmission, the compensation amount is predicted through a mechanical model;

[0050] Obtain the final control quantity: joint control output + compensation amount.

[0051] (1) Group error propagation model

[0052] As Figure 2 shown, the group error propagation model is used to describe how individual errors spread and interact in the system when multiple AGVs work collaboratively. Based on the vibration theory of elasticity, the AGV cluster is regarded as a dynamically coupled system, and the error transmission law is analyzed through mathematical modeling.

[0053] Establish the vibration transmission equation based on elasticity: [K]{δ}+[C]{δ′}+[M]{δ″}={F(t)}

[0054] Where: [K] is the ground stiffness matrix (related to material, identified online by pressure sensor), [C] is the damping matrix (reflecting vibration attenuation characteristics), [M] is the AGV mass matrix, {δ} is the AGV pose error vector, {F(t)} is the external excitation force (such as the acceleration / braking of other AGVs)

[0055] (2) Adaptive weight fusion algorithm

[0056] Adaptive weight fusion algorithm is a dynamic data fusion method for multi-sensor or multi-AGV cooperative positioning. Its core idea is to dynamically adjust the weight according to the real-time reliability of different data sources to improve the accuracy and robustness of the overall system. Its core idea: In AGV cooperative work, each AGV may carry multiple sensors (such as laser radar, IMU, vision, etc.) or receive shared positioning data from other AGVs. Due to environmental interference, sensor error or communication delay, the reliability of different data sources will change dynamically. Adaptive weight fusion algorithm dynamically calculates the weight of each data source, so that high-reliability data has a greater impact on the final fusion result, while low-reliability data is suppressed.

[0057] The mathematical expression of the weight is:

[0058]

[0059] where α, β, γ, δ are adjustable parameters, d i is the distance, Stability i is the environmental stability evaluation. α is the distance attenuation factor, data close to the target is usually more reliable (such as near-distance TOF sensor is more accurate than long-distance laser radar); SNR is the strength of the communication signal (such as UWB or millimeter wave signal noise ratio); E history is the average error of this data source in the past period of time (such as IMU drifts over time, and the reliability decreases).

[0060] Data fusion generally uses federated Kalman filter (FKF) for fusion, each AGV runs Kalman filter locally, and outputs local optimal estimate The master node fuses all local estimates according to the weight ω i to get the global optimal X global .

[0061] As shown in Figure 3 , the cooperative work flow is:

[0062] (1) Initialization phase:

[0063] Multiple AGVs construct a shared high-precision map (global error <±2mm) through laser radar scanning.

[0064] (2) Running phase:

[0065] AGV-A detects ground vibration and broadcasts vibration waveform characteristics through millimeter wave.

[0066] AGV-B receives data and adjusts the servo motor torque output 50ms in advance.

[0067] (3) Correction phase:

[0068] Firstly, when multiple AGVs enter the collaborative correction phase (such as fork spacing <1m), a distributed weighted PID control is adopted:

[0069]

[0070] where u joint (t) is the joint control output (such as motor torque / steering angle adjustment); e j (t) is the real-time error of the jth AGV (such as the deviation between the actual pose and the target pose); K p，j , K i，j , K d，j are the PID parameters of the jth AGV (adjusted dynamically according to AGV mass and motion state); N is the number of AGVs participating in collaboration. d represents differentiation, t is time, and dt is the differential of time.

[0071] K p，j and the weight ω j are related, and the weight ω j depends on the credibility of the AGV (such as sensor accuracy, communication delay):

[0072]

[0073] where RMSE j is the historical root mean square error of the jth AGV.

[0074] K p，base is the basic gain parameter of the proportional term in PID control, used to determine the initial response strength of the system to the error.

[0075] Secondly, for the error of ground vibration transmission, the compensation amount is predicted through a mechanical model:

[0076] Δu ff = [H(ω)]·F(t-τ).

[0077] where Δuff is the feedforward compensation (added to the PID output); [H(ω)] is the frequency response function matrix (calibrated through experiments),

[0078] H ik (ω) = δ i (ω) / F k (ω) (the influence of the kth AGV excitation on the ith AGV); F(t-τ) is the excitation force after time delay τ (shared in real time through millimeter wave communication).

[0079] Finally, the final control amount is the PID feedback + vibration feedforward: u total (t) = u joint (t) + Δu ff .

[0080] Automotive parts warehouse scenarios, such as Figure 4 as shown:

[0081] (1) Environment configuration:

[0082] 3 AGVs cooperatively handling engine parts in a 4m wide aisle

[0083] The ground is a steel grid with a stiffness coefficient K = 2 x 10 6 N / m

[0084] AGV configuration:

[0085] AGV-A (trigger source): mass m1 = 350 kg, emergency stop acceleration a = -3 m / s 2 ;

[0086] AGV-B / C (affected units): mass m2 = m3 = 300 kg, velocity v = 1 m / s:

[0087] Rack positioning accuracy requirement: ±1 mm

[0088] (2) Implementation process:

[0089] AGV cooperatively handling, AGV-A emergency stop causing ground vibration, generating exciting force

[0090] F1 = m1 · a = 350 kg x (-3 m / s 2 ) = -1050 N

[0091] Then the ground vibration frequency is

[0092]

[0093] AGV-B performs vibration detection, and the amplitude attenuation at AGV-B is

[0094] A2 = A1 · e -βd = 0.3 mm · e -0.15×1.5 ≈ 0.24 mm

[0095] Where β = 0.15 is the attenuation coefficient of the steel grid, and d = 1.5 m is the AGV spacing.

[0096] At AGV-B, the theoretical displacement caused by vibration is

[0097] Δx theory = A2 · sin(2πft) = 0.24 mm · sin(2π · 12 · t)

[0098] The compensation amount calculated by the frequency response function (experimental calibration H(12Hz) = 0.2 mm / N) is

[0099] Δx comp = -H(f) · F1 = -0.2 x (-1050 N) = +1.0 mm

[0100] The actual compensation is

[0101] x target = x nominal + Δx comp = x nominal + 1.0 mm

[0102] AGV-B performs torque compensation, and the required holding torque is

[0103] τ = J · α + μ · F n = 0.5 kg · m 2 · 2 rad / s 2 + 0.1 x 3000 N = 1 N · m + 0.3 N · m

[0104] Where J is the moment of inertia, and μ is the friction coefficient.

[0105] AGV-C performs speed adjustment, and the speed reduction ratio is

[0106]

[0107] Where A3 = 0.2 mm is the amplitude at AGV-C, and A max = 1.2 mm is the maximum allowed value.

[0108] AGV-B / C adjusts 50 ms in advance according to the calculation results.

[0109] Collaborative working architecture of distributed pressure sensor and millimeter wave communication;

[0110] Pre-compensation algorithm based on ground vibration inversion, weight dynamic distribution strategy of multi-AGV joint PID control.

[0111] (1) Group error suppression: In the scene of multiple AGVs intensive operation, reduce the additional error caused by mutual interference;

[0112] (2) Dynamic coupling compensation: Establish the kinematic influence model between AGVs to eliminate the attitude angle deviation caused by ground vibration;

[0113] (3) Real-time collaborative control: Complete the fusion of group pose data and issue compensation instructions within the agreed time.

[0114] The multi-AGV collaborative system architecture includes a perception layer and a control layer.

[0115] The perception layer of each AGV is equipped with a dual-frequency RTK-GNSS (for open outdoor areas, accuracy ±2cm), a distributed pressure sensor array (detects ground vibration waveform, resolution 0.1N), and a millimeter wave communication module (60GHz frequency band, point-to-point transmission delay <5ms).

[0116] The control layer edge computing node deploys a federated learning framework, and each AGV locally processes 90% of the data; a dynamic priority arbitrator allocates communication resources according to the task urgency.

[0117] Among them, the TOF array is installed on the tines, and the dual-frequency RTK-GNSS is used for positioning.

[0118] The technical principles of the present application are described above in combination with specific embodiments. These descriptions are only for the purpose of explaining the principles of the present application and cannot be interpreted in any way as a limitation on the scope of protection of the present application. Based on the explanations here, those skilled in the art can think of other specific embodiments of the present application without having to exert creative effort, and these embodiments will fall within the scope of protection of the claims of the present application.

Claims

1. A dynamic accuracy compensation method for multi-AGV collaborative operation, characterized in that: The following steps are involved: Multiple AGVs build a shared high-precision map through lidar scanning; When AGV-A detects ground vibration, it broadcasts the vibration waveform characteristics via millimeter waves. When AGV-B receives the data, it adjusts the servo motor torque output in advance according to the predetermined time. The adjustment of the servo motor torque output is based on a final control amount, and obtaining the final control amount includes the following steps: When multiple AGVs enter collaborative deviation correction, distributed weighted PID control is used to obtain joint control output; For the error transmitted by ground vibration, the compensation amount is predicted through the mechanical model; The final control quantity is obtained: combined control output quantity + compensation quantity.

2. The dynamic accuracy compensation method for multi-AGV collaborative operation according to claim 1 is characterized in that: The predetermined time is 20-70 ms.

3. The dynamic accuracy compensation method for multi-AGV collaborative operation according to claim 1 is characterized in that: When broadcasting the vibration waveform characteristics, the group error propagation model is used to reduce the additional error caused by mutual interference, wherein the vibration transmission equation is: [K]{δ}+[C]{δ′}+[M]{δ″}={F(t)}; Where: K is the ground stiffness matrix, C is the damping matrix, M is the AGV mass matrix, δ is the AGV posture error vector, and F(t) is the external excitation force.

4. The dynamic accuracy compensation method for multi-AGV collaborative operation according to claim 1 is characterized in that: The distributed weighted PID control includes: Where u joint (t) is the joint control output; e j (t): real-time error of the jth AGV; K p,j ,K i,j ,K d,j are the PID parameters of the j-th AGV; N is the number of AGVs participating in the collaboration.

5. The dynamic accuracy compensation method for multi-AGV collaborative operation according to claim 4 is characterized in that: K p，j =w j ·K p，base ; Where K p,base It is the basic gain parameter of the proportional term in PID control, which is used to determine the initial response strength of the system to the error, RMSE j is the historical root mean square error of the jth AGV; ω j is the weight.

6. The method for dynamic accuracy compensation of multiple AGV collaborative operations according to claim 5, characterized in that: The weight ω i : In the formula, α, β, γ, δ are adjustable parameters, d i For distance, Stability i is the environmental stability assessment; α is the distance attenuation factor, SNR is the strength of the communication signal, E history The average error of the data source over the past period of time.

7. The method for dynamic accuracy compensation of multiple AGV collaborative operations according to claim 4, characterized in that: The predicted compensation amount: Thu ff =[H(ω)]·F(t-τ); Where Δuff is the feedforward compensation; H(ω) is the frequency response function matrix; H ik (ω)=δ i (ω) / F k (ω), the impact of the k-th AGV incentive on the i-th AGV; F(t-τ) is the excitation force after a time delay of τ.

8. The system of a dynamic accuracy compensation method for multi-AGV collaborative operation according to claim 1, characterized in that: include: The perception layer includes dual-frequency RTK-GNSS, a distributed pressure sensor array, and a millimeter-wave communication module on the AGV; The control layer is used to allocate communication resources according to the urgency of the task.

Images