AGV heavy load mobile robot combination dynamic adaptive variable gain control method

Abstract

The application discloses an AGV heavy-load mobile robot combined dynamic adaptive variable gain control method, which is developed on the basis of AGV technology, artificial intelligence technology and large-scale computing chip technology, and is characterized by the following steps: using a high-speed intelligent chip and a new algorithm, adopting a modular design idea, and comprehensively analyzing vehicle sensor data in real time during AGV operation, so that the conventional PID control process of AGV speed is simplified, a deep learning technology is adopted to adjust the high-speed and low-speed operation control process of the AGV, variable speed control and dynamic parameter adjustment are realized, and the control efficiency is improved.

Description

Technical Field

[0001] This invention relates to the technical field of AGV heavy-duty mobile robot control, and in particular to a dynamic adaptive variable gain control method for AGV heavy-duty mobile robot combinations. Background Technology

[0002] Automated Guided Vehicles (AGVs) are equipped with radar, electromagnetic, and optical sensors for automatic detection and judgment. In industrial, commercial, and logistics applications, they travel along predetermined work paths, performing autonomous or semi-autonomous guidance and possessing a certain degree of safety protection and obstacle avoidance capabilities. Current AGV automatic control methods are relatively simple, mostly employing constant speed control, meaning there are only two states: stopped and moving at a constant speed. When facing complex working conditions and obstacle avoidance, the multiple Mecanum wheels of heavy-duty AGVs generate complex state information, and each wheel operates independently, making control difficult. The distance, speed, and posture changes signals obtained from sensors, as well as the instantaneous and diverse changes in center of gravity distribution under different loads, are varied. Using fixed-gain control makes it difficult to account for all situations, resulting in low efficiency in acceleration control and adjustment, and insufficient adaptability.

[0003] Therefore, how to develop a new control method to solve the above problems has become an urgent issue. Summary of the Invention

[0004] In view of this, the present invention provides a dynamic adaptive variable gain control method for AGV heavy-duty mobile robots, in order to solve the problems of low acceleration control and adjustment efficiency and insufficient adaptability of previous control methods for AGV heavy-duty mobile robots.

[0005] The technical solution provided by this invention is specifically a dynamic adaptive variable gain control method for AGV heavy-duty mobile robot combinations, which includes the following steps:

[0006] Obtain the AGV's heavy-duty mobile robot's load transportation task M from the AGV task scheduling system. d And based on the load transport task M d The AGV operating speed setpoint V was obtained by calculating the speed at each unloading station. d and path setting value R d ;

[0007] Obtain information from the integrated control layer of the AGV heavy-duty mobile robot, and combine it with the load transportation task M. d Operating speed setpoint V d and path setting value R d The trajectory position offset E is calculated. rReal-time distribution of the center of gravity of the AGV after loading cargo D d ;

[0008] Set the running speed value V d Trajectory position offset E r Real-time distribution of the center of gravity of the AGV after loading cargo (D) d αE of variable gain nv βE r And based on the real-time distribution of the center of gravity of the AGV after it is loaded with cargo D d Speed ​​feedback v of the wheel set controller n The calculated predicted attitude value P of the loaded AGV nv The difference D between real-time acceleration and preset acceleration e Parameters are synthesized to form input parameters, where α and β are variable gain parameters for speed and trajectory control, reflecting the ability of trajectory deviation and real-time speed changes to affect the operating status of the AGV heavy-duty mobile robot.

[0009] Input parameters and current feedback I from the wheel set controller n Speed ​​feedback v n and torque feedback T n The input is fed into the AGV deep adversarial control model, and the control parameters ΔU of the wheel servo motors in the AGV heavy-duty mobile robot are generated by adversarial deep learning technology. n ;

[0010] The servo motor of the wheel assembly in the AGV heavy-duty mobile robot is controlled according to the parameter ΔU. n It completes the torque, direction and speed control of the wheel set servo motor.

[0011] Preferably, the AGV operating speed setpoint V d The calculation formula is:

[0012]

[0013] And the AGV running speed setpoint V d The constraints are:

[0014] Velocity constraint V d ∈[V min V max ], where V min For minimum operating speed, V max Maximum operating speed;

[0015] Stability constraints during cargo transportation;

[0016] The order requirements for the front and rear unloading stations.

[0017] Further preferably, the information in the integrated control layer of the AGV heavy-duty mobile robot includes: navigation and positioning, task operation rules, offset and trajectory prediction, and center distribution.

[0018] Further preferably, the trajectory position offset E r For: the navigation and positioning information of the AGV heavy-duty mobile robot and R d The positioning points of the pre-set path are compared to obtain the normalized values ​​of the left-right and front-back deviations of the positioning points.

[0019] Further preferably, α and β are variable gain parameters for speed and trajectory control, reflecting the ability of trajectory deviation and real-time speed changes to affect the AGV's operating status, where α∈(0,1), β∈(0,1), and α+β=1.

[0020] Further preferred, the predicted attitude value P of the load-bearing AGV nv =e γ δ (IMU measurement);

[0021] Where γ is the real-time acceleration value, and δ is the center of gravity coefficient, which is the average value of the center of gravity of the AGV cargo.

[0022] Further preferably, the AGV deep adversarial control model includes:

[0023] Servo drive operating parameter generator G1, AGV wheel group drive parameter discriminator D1, operating status generator G2, and AGV operating status discriminator D3;

[0024] The training process of the AGV deep adversarial control model includes:

[0025] S1: Extract features from a specified operating state in the AGV operating state library to obtain AGV operating features, and input the AGV operating features and AGV operating rules into the servo driver operating parameter generator G1 to generate AGV driving parameters;

[0026] S2: The loss Y2 between the AGV drive parameters generated by the AGV wheel set drive parameter discriminator D1 and the ideal control parameters;

[0027] S3: The AGV running status is generated by the running status generator G2 based on the AGV driving parameters generated by the servo driver running parameter generator G1;

[0028] S4: The loss Y4 between the generated AGV running state and the specified running state in the AGV running state library, as determined by the AGV running state discriminator D3;

[0029] S5: Based on the judgment results of steps S2 and S4, adjust the model parameters corresponding to the AGV wheel set drive parameter discriminator D1 until they meet the threshold requirements.

[0030] Further preferred, the objective function of the AGV deep adversarial control model is:

[0031]

[0032] More preferably, the servo driver operating parameter generator G1 includes:

[0033] The input layer is used to receive the input runtime parameters;

[0034] The feature sparse matrix is ​​used to process the input running parameters;

[0035] Fully connected layer: classifies the running parameters after processing the feature sparse matrix;

[0036] The Densenet121 network, as a first-level parameter generator, generates first-level control parameters based on the operating parameters input from the fully connected layer.

[0037] The LSTM network, acting as a second-level parameter generator, generates the nth set of control voltage ΔU based on the source parameter influence factor γ and the first-level control parameters generated by the Densenet121 network. n .

[0038] Further optimization yields the following formula for calculating the γ factor:

[0039]

[0040] In the formula, I is the initial maximum deviation parameter set, I t Let Pb be the squared deviation of the motor drive parameters in the current batch, Pb be the maximum squared deviation of the servo motor drive parameters at the initial training time, and Pb be the squared deviation of the servo motor drive parameters in the current training batch.

[0041] The AGV heavy-duty mobile robot combination dynamic adaptive variable gain control method provided by this invention is based on the development of AGV technology, artificial intelligence technology, and large-scale computing chip technology. It is a method for dynamic adaptive variable gain control of AGV heavy-duty mobile robot combination. By using high-speed intelligent chips and new algorithms, and adopting a modular design concept, the method comprehensively analyzes vehicle sensor data in real time during AGV operation, simplifies the conventional PID control process of AGV speed, and uses deep learning technology to adjust the high-speed and low-speed operation control process of AGV, realizing variable speed control and dynamic parameter adjustment, thereby improving control efficiency.

[0042] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit the disclosure of the present invention. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the control process of a dynamic adaptive variable gain control method for AGV heavy-duty mobile robot combination provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the generator's components.

[0047] Figure 3 This is a schematic diagram of the generator's components. Detailed Implementation

[0048] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of methods consistent with some aspects of the invention as detailed in the appended claims.

[0049] To address the shortcomings of previous control methods for heavy-duty AGV mobile robots, such as low efficiency in acceleration control and adjustment and insufficient adaptability, this implementation scheme provides a combined dynamic adaptive variable gain control method for heavy-duty AGV mobile robots. The specific control process is detailed in [link to relevant documentation]. Figure 1 The system uses four wheels as a group, with cross-coupling control within each group and vehicle-wide attitude control between groups, including the following steps:

[0050] Obtain the heavy-duty transportation task M of the AGV from the AGV task scheduling system of the AGV heavy-duty mobile robot. d And based on the load transport task M d The AGV operating speed setpoint V was obtained by calculating the speed at each unloading station. d and path setting value R d ;

[0051] Obtain information from the integrated control layer of the AGV heavy-duty mobile robot, and combine it with the load transportation task M. d Operating speed setpoint V d and path setting value R d The trajectory position offset E is calculated. r Real-time distribution of the center of gravity of the AGV after loading cargo D d ;

[0052] Set the running speed value V d Trajectory position offset E r Real-time distribution of the center of gravity of the AGV after loading cargo (D) d αE of variable gain nv βE r And based on the real-time distribution of the center of gravity of the AGV after it is loaded with cargo D d Speed ​​feedback v of the wheel set controller n The calculated predicted attitude value P of the loaded AGV nv The difference D between real-time acceleration and preset acceleration e Parameters are synthesized to form input parameters, where α and β are variable gain parameters for speed and trajectory control, reflecting the ability of trajectory deviation and real-time speed changes to affect the operating status of the AGV heavy-duty mobile robot.

[0053] Input parameters and current feedback I from the wheel set controller n Speed ​​feedback v n and torque feedback T n The input is fed into the AGV deep adversarial control model, and the control parameters ΔU of the wheel servo motors in the AGV heavy-duty mobile robot are generated by adversarial deep learning technology. n ΔU n To adjust the control voltage for each round of the nth group of the control model;

[0054] The servo motor of the wheel assembly in the AGV heavy-duty mobile robot is controlled according to the parameter ΔU. n It completes the torque, direction and speed control of the wheel set servo motor.

[0055] Among them, the predicted attitude value P of the load-bearing AGV nv This includes: yaw, tilt, and steering inertia.

[0056] AGV's load-carrying transportation task M d This data comes from transportation task data, including the weight, type, volume, unloading stations, and time requirements of the goods to be transported. d This is the preset speed of the AGV for this operation, that is, the average speed calculated based on the route, the speed performance of the AGV itself, the safety requirements for cargo transportation, and the route length. R dThe task server is based on M d The task requires the AGV to travel along its expected path.

[0057] AGV running speed setpoint V d The calculation formula is:

[0058]

[0059] And the AGV running speed setpoint V d The constraints are:

[0060] Velocity constraint V d ∈[V min V max ], where V min For minimum operating speed, V max Maximum operating speed;

[0061] Stability constraints during cargo transportation;

[0062] The order requirements for the front and rear unloading stations.

[0063] The information in the integrated control layer of the AGV heavy-duty mobile robot includes: navigation and positioning, task operation rules, offset and trajectory prediction, and center distribution.

[0064] The difference D between real-time acceleration and preset acceleration e = Current real-time acceleration - Current position preset acceleration.

[0065] The trajectory position offset E r For: the navigation and positioning information of the AGV heavy-duty mobile robot and R d The positioning points of the pre-set path are compared to obtain the normalized values ​​of the left-right and front-back deviations of the positioning points.

[0066] The α and β are variable gain parameters for speed and trajectory control, reflecting the ability of trajectory deviation and real-time speed changes to affect the AGV's operating status. α ∈ (0,1), β ∈ (0,1), and α + β = 1.

[0067] The predicted attitude value P of the loaded AGV nv =e γ δ (IMU measurement);

[0068] Where γ is the real-time acceleration value, and δ is the center of gravity coefficient, which is the average value of the center of gravity of the AGV cargo.

[0069] See Figure 1 From a module perspective, the above-mentioned dynamic adaptive variable gain control method for the AGV heavy-duty mobile robot combination is specifically as follows:

[0070] Task and Path Reference: Obtain the AGV's load transportation task M from the AGV task scheduling system. d The AGV operating speed setpoint V is calculated based on each unloading station. d and path setting value R d .

[0071] Integrated control module: based on M d V d R d Using the AGV's position sensors and body pressure sensors as input, E is calculated. r and D d and V d E r and D d It is passed to the control computing layer.

[0072] The control computation layer comprises two parts: parameter synthesis and a servo control parameter adversarial generative model. The parameter synthesis module will integrate V... d E r D d Adding the system's variable gains α and β to αE nv βE r , and P nv D e Parameter synthesis is completed to form the input parameter vector for the control model. A deep adversarial control model for the heavy-duty AGV is generated from the input parameters of the parameter synthesis module, and the control parameters ΔU of the servo motor are generated using adversarial deep learning technology. n .

[0073] Wheel set n-coupling drive control module: Receives the generated control parameter ΔU n It completes the torque, direction and speed control of the servo motor.

[0074] See Figure 2 The AGV deep adversarial control model includes: servo driver operating parameter generator G1, AGV wheel group drive parameter discriminator D1, operating status generator G2, and AGV operating status discriminator D3.

[0075] The training process of the AGV deep adversarial control model includes:

[0076] S1: Extract features from a specified operating state in the AGV operating state library to obtain AGV operating features, and input the AGV operating features and AGV operating rules into the servo driver operating parameter generator G1 to generate AGV driving parameters;

[0077] S2: The loss Y2 between the AGV drive parameters generated by the AGV wheel set drive parameter discriminator D1 and the ideal control parameters;

[0078] S3: The AGV running status is generated by the running status generator G2 based on the AGV driving parameters generated by the servo driver running parameter generator G1;

[0079] S4: The loss Y4 between the generated AGV running state and the specified running state in the AGV running state library, as determined by the AGV running state discriminator D3;

[0080] S5: Based on the judgment results of steps S2 and S4, adjust the model parameters corresponding to the AGV wheel set drive parameter discriminator D1 until they meet the threshold requirements.

[0081] Generator G1 takes the runtime state library as input to generate drive parameters; generator G2 takes the control parameters generated by G1 as input to generate virtual runtime state parameters. The model objective is to achieve consistency between the AGV's runtime state and the servo motor control parameters through iterative adversarial processing.

[0082] To minimize the loss Y4, ensure that the process of generating Y1 from X2 is correlated with X2.

[0083] Minimize the loss Y2 to ensure that Y1 generated by X2 is as close as possible to the ideal control parameters X3 of the drive.

[0084] The objective function of the AGV deep adversarial control model is:

[0085]

[0086] The objective is to minimize the difference between the reverse-generated runtime state parameters and the specified reference runtime state parameters.

[0087] See generator model Figure 3 The servo driver operating parameter generator G1 includes:

[0088] The input layer is used to receive the input runtime parameters;

[0089] The feature sparse matrix is ​​used to process the input running parameters;

[0090] Fully connected layer: classifies the running parameters after processing the feature sparse matrix;

[0091] The Densenet121 network, as a first-level parameter generator, generates first-level control parameters based on the operating parameters input from the fully connected layer.

[0092] The LSTM network, acting as a second-level parameter generator, generates the nth set of control voltage ΔU based on the source parameter influence factor γ and the first-level control parameters generated by the Densenet121 network. n .

[0093] Where γ is the influence factor of the source parameter. Run the parameter generator, with M... d Task, V d Speed ​​preset, R d Path preset, E r Trajectory position deviation, D d Center of gravity distribution, E nv Speed ​​difference, D e Difference in acceleration, P nv Attitude prediction, α and β variable gain parameters, I n Current feedback, v n Speed ​​feedback, T n Torque feedback, ΔU n In each round of the nth group, the control voltage is adjusted as the input, and the data structure is reorganized into a sparse matrix. A DenseNet121 network is used as the first-level parameter generator, and an LSTM network as the second-level parameter generator. To guide the generator's generation process directly in relation to the source data, a source parameter influence factor γ is set. As training progresses, the value of γ gradually decreases. Initially, γ is close to 1, maximizing the influence of the source parameters. Near the end of training, a generative model has been formed, and γ approaches 0. The formula for calculating the γ factor is:

[0094]

[0095] In the formula, I is the initial maximum deviation parameter set, I t Let Pb be the squared deviation of the motor drive parameters in the current batch, Pb be the maximum squared deviation of the servo motor drive parameters at the initial training time, and Pb be the squared deviation of the servo motor drive parameters in the current training batch.

[0096] At the initial stage of training, the deviation of the target motor drive parameters is the largest. As the training process progresses, the deviation value continuously decreases. Ideally, it will eventually decrease to 0, that is, the value of γ will automatically be between (0, 1).

[0097] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0098] It should be understood that the present invention is not limited to the content already described above, and various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A dynamic adaptive variable gain control method for AGV heavy load mobile robot combination, characterized in that, Includes the following steps: Obtain the AGV's load transportation task from the AGV task scheduling system of the AGV heavy-duty mobile robot. And according to the load transportation task The AGV operating speed setpoint was obtained by calculating the speed at each unloading station. and path settings ; Obtain information from the integrated control layer of the AGV heavy-duty mobile robot and combine it with the load transportation task. Running speed setting value and path settings The trajectory position offset is calculated. Real-time distribution of the center of gravity of the AGV after loading cargo ; Set the running speed value trajectory position offset Real-time distribution of the center of gravity of the AGV after loading cargo. Variable gain , And based on the real-time distribution of the center of gravity of the AGV after it is loaded with cargo. Speed ​​feedback from wheel set controller Calculated predicted values ​​of the load-bearing AGV attitude The difference between real-time acceleration and preset acceleration Parameters are synthesized to form input parameters, where, , The variable gain parameters for speed and trajectory control reflect the ability of trajectory deviation and real-time speed changes to affect the operating status of the AGV heavy-duty mobile robot. Feedback of input parameters and current from the wheel controller Speed ​​feedback and torque feedback The inputs are fed into the AGV deep adversarial control model, where adversarial deep learning technology generates the control parameters for the wheel servo motors in the AGV heavy-duty mobile robot. ; The servo motor of the wheel assembly in the AGV heavy-duty mobile robot is controlled according to parameters. It completes the torque, direction and speed control of the wheel set servo motor.

2. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 1, characterized in that, AGV running speed set value The calculation formula is: ； AGV running speed set value The constraint condition is: Speed ​​constraints ,in, For minimum operating speed, Maximum operating speed; Stability constraints during cargo transportation; The order requirements for the front and rear unloading stations.

3. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 1, characterized in that, The information in the integrated control layer of the AGV heavy-duty mobile robot includes: navigation and positioning, task operation rules, offset and trajectory prediction, and center distribution.

4. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 1, characterized in that, The trajectory position offset For: the navigation and positioning information of the AGV heavy-duty mobile robot and The positioning points of the pre-set path are compared to obtain the normalized values ​​of the left-right and front-back deviations of the positioning points.

5. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 1, characterized in that, The , These are variable gain parameters for speed and trajectory control, reflecting the ability of trajectory deviation and real-time speed changes to affect the AGV's operating status. , ,and .

6. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 1, characterized in that, The load AGV attitude prediction value ; wherein, is a real-time acceleration value, is a gravity center coefficient, the value of which is the average value of the gravity center of the AGV cargo.

7. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 1, characterized in that, The AGV deep adversarial control model includes: Servo drive operating parameter generator G1, AGV wheel group drive parameter discriminator D1, operating status generator G2, and AGV operating status discriminator D3; The training process of the AGV deep adversarial control model includes: S1: Extract features from a specified operating state in the AGV operating state library to obtain AGV operating features, and input the AGV operating features and AGV operating rules into the servo driver operating parameter generator G1 to generate AGV driving parameters; S2: The loss Y2 between the AGV drive parameters generated by the AGV wheel set drive parameter discriminator D1 and the ideal control parameters; S3: The AGV running status is generated by the running status generator G2 based on the AGV driving parameters generated by the servo driver running parameter generator G1; S4: The loss Y4 between the generated AGV running state and the specified running state in the AGV running state library, as determined by the AGV running state discriminator D3; S5: Based on the judgment results of steps S2 and S4, adjust the model parameters corresponding to the AGV wheel set drive parameter discriminator D1 until they meet the threshold requirements.

8. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 7, characterized in that, The objective function of the AGV deep adversarial control model is: 。 9. The AGV heavy load mobile robot combined dynamic self-adaptive variable gain control method according to claim 7, characterized in that, The servo driver operating parameter generator G1 includes: The input layer is used to receive the input runtime parameters; The feature sparse matrix is ​​used to process the input running parameters; Fully connected layer: classifies the running parameters after processing the feature sparse matrix; The Densenet121 network, as a first-level parameter generator, generates first-level control parameters based on the operating parameters input from the fully connected layer. The LSTM network, acting as a second-level parameter generator, generates the nth set of control voltages based on the source parameter influence factor γ and the first-level control parameters generated by the Densenet121 network. .

10. The AGV heavy-duty mobile robot combined dynamic adaptive variable gain control method according to claim 9, characterized in that, The formula for calculating the factor is: ； In the formula, I is the initial maximum deviation parameter set, I t Let Pb be the squared deviation of the motor drive parameters in the current batch, Pb be the maximum squared deviation of the servo motor drive parameters at the initial training time, and Pb be the squared deviation of the servo motor drive parameters in the current training batch.

Images