Neural network-based unmanned walking excavator body posture smooth control method

Abstract

The present application provides a kind of neural network-based unmanned walking excavator body posture stable control method, including the front terrain of walking excavator is scanned by laser radar and camera, generates terrain point cloud data, terrain point cloud data is handled, obtains terrain excitation data;Terrain excitation data and walking leg angle of walking excavator are as input, construct the neural network posture prediction model of walking excavator, output the pitch angle, roll angle and vertical displacement of walking excavator body;Based on the body posture change of walking excavator, construct the body stability optimal control model of walking excavator and solve, to optimize the angle of walking leg and chassis, realize the stable control of body posture under complex terrain conditions.The present application can efficiently adapt to complex terrain environment, improve the stability and safety of unmanned walking excavator in autonomous operation, and has important significance for improving operation efficiency and quality.

Description

TECHNICAL FIELD

[0001] The embodiment of the application relates to the field of engineering machinery control and intelligent optimization, and particularly relates to a neural network-based stable control method for a vehicle body posture of an unmanned walking excavator. BACKGROUND

[0002] The walking excavator is a full-terrain and multipurpose walking excavator and is widely used in operation tasks in complex terrain environments such as mountains, hills and forests.

[0003] Compared with a traditional crawler excavator, the chassis of the walking excavator is used to adjust the balance of the vehicle body through four independent walking mechanisms, so that the applicability and stability of the excavator are greatly improved. The stability of the walking excavator is crucial to operation safety and construction efficiency. Since the traditional kinematic model contains nonlinear terms of sine and cosine functions, the analytical solution can be obtained by simplification under a small angle obstacle, but such simplification cannot be performed when the pitch and roll angles are relatively large, so that it takes a long time to solve the numerical solution in the optimization model, and the walking excavator is difficult to realize real-time adjustment of the vehicle body balance under complex terrain conditions. SUMMARY

[0004] Therefore, the embodiment of the application provides a neural network-based stable control method for a vehicle body posture of an unmanned walking excavator, which aims to solve the problems of insufficient stability and inaccurate posture control of the existing walking excavator under complex terrain.

[0005] According to a first aspect of an embodiment of the present application, a neural network-based body posture stability control method for a walking excavator is provided, which comprises the following steps: S1, scanning the terrain in front of the walking excavator by means of a laser radar and a camera to generate terrain point cloud data, processing the terrain point cloud data to obtain terrain excitation data; S2, taking the terrain excitation data and the leg clamping angle of the walking excavator as inputs to construct a neural network posture prediction model of the walking excavator, and outputting the pitch angle, roll angle and vertical displacement of the body of the walking excavator to predict the body posture change of the walking excavator; S3, based on the body posture change of the walking excavator, constructing a body stability optimal control model of the walking excavator, including: optimization variables, objective function, model constraints; S4, assigning values to the parameters in the body stability optimal control model; S5, discretizing the variables of the body stability optimal control model at the collocation points to obtain processed variables; S6, assigning values to the processed variables at the collocation points to obtain the body pose, hydraulic cylinder angle and hydraulic cylinder speed time sequence trajectory; S7, according to the body pose, hydraulic cylinder angle and hydraulic cylinder speed time sequence trajectory, calculating the objective function and constraint violation degree of the body stability optimal control model under the current driving trajectory, including the hydraulic cylinder extension length, hydraulic cylinder extension speed and distance from the tire center to the ground; S8, judging whether the body stability optimal control model meets the optimization convergence condition, if yes, outputting the optimal control variables of the body stability optimal control model, if not, going to step S9; S9, single-step iterative optimization of the state variables and control variables of the body stability optimal control model, and going to step S6 until the optimal control variables of the body stability optimal control model are outputted, so as to realize the stability control of the body of the walking excavator.

[0006] In an implementation manner, the terrain excitation data in step S1 comprises slope, smoothness and concave-convex degree.

[0007] In another implementation manner, in step S1, the processing manner of the terrain point cloud data comprises at least one of statistical filtering, radius filtering denoising, Voxel Grid filtering downsampling and RANSAC algorithm plane segmentation.

[0008] In another implementation manner, in step S2, the neural network posture prediction model of the walking excavator is constructed in the following manner: based on the terrain excitation data and the leg clamping angle of the walking excavator, a training data set of the neural network posture prediction model is constructed, the input data of the neural network posture prediction model comprises terrain excitation (w1, w2, w3, w4) and leg clamping angle (θ1, θ2, θ3, θ4) of the walking leg and the chassis, and the output data is the body posture parameter, including: pitch angle θ s , roll angle and vertical displacement z sThe neural network posture prediction model adopts a multi-layer feedforward neural network structure, including an input layer, a plurality of hidden layers and an output layer, the number of nodes of the input layer is n+4, wherein n represents the number of terrain excitation features, and the four nodes represent the angles of the walking legs and the chassis, the hidden layer is composed of a plurality of neurons, the activation function adopts ReLU, and the output layer includes three nodes, respectively corresponding to the pitch angle θ s , the roll angle and the vertical displacement z s ; a loss function and an optimization objective are defined, and the neural network posture prediction model is trained by a supervised learning method, a mean square error loss function is selected to minimize the error between the prediction output of the neural network posture prediction model and the real posture data, and the mean square error loss function is defined as:

[0009]

[0010] , wherein, represents the output value predicted by the neural network posture prediction model, y i is the real posture value, and m is the sample number; the optimization method adopts an Adam optimization algorithm, the weights of the neural network posture prediction model are iteratively updated according to the gradient information, so as to accelerate the convergence; the neural network posture prediction model is trained by multiple rounds of iteration to update the weight parameters, so as to reduce the error between the prediction output and the real value; regularization and early stopping strategies are adopted, and the best model structure and hyperparameters of the neural network posture prediction model are determined through cross-validation, so as to prevent overfitting of the neural network posture prediction model; after the training is completed, the neural network posture prediction model can accurately predict the posture changes of the walking excavator according to the input terrain excitation data and the walking leg angle.

[0011] In another implementation mode, the Adam optimization algorithm dynamically adjusts the learning rate based on the gradient average values of the first and second moments, so as to accelerate the speed of the neural network posture prediction model to reach the optimal solution.

[0012] In another implementation mode, in step S3, the body stability optimal control model of the walking excavator is constructed as follows: taking the walking leg angle as the optimization variable and taking the stability of the body posture as the objective function, the body stability optimal control model is constructed, and the objective function is expressed as:

[0013]

[0014] , wherein, s represents the pitch angle, represents the roll angle, and z s represents the vertical displacement, t f represents the running time of the excavator.

[0015] Model constraints include: the extension and retraction speeds v of the hydraulic cylinders of the four support legs of the walking excavator during attitude adjustment. hi (i = 1, 2, 3, 4) ensures that the maximum allowable lower limit v is met. hmin and maximum allowed limit v hmax between:

[0016]

[0017] During the movement, the center of the four legs of the walking excavator will be displaced by z. ui (i = 1, 2, 3, 4) and the roadside elevation w corresponding to each wheel i The difference is kept at the maximum allowable lower limit ε min and maximum allowed upper limit ε max between:

[0018]

[0019] Therefore, the optimal control model for the vehicle stability of a walking excavator is expressed as follows:

[0020] findθ i (i = 1, 2, 3, 4)

[0021]

[0022] stC ij ≤0 (i=1,2,3,4,j=1,2,3,4)

[0023] t min ≤t f ≤t max

[0024] Where, θ i (i = 1, 2, 3, 4) represents the angle between the walking legs, θ s Indicates pitch angle, Indicates the roll angle, z s t represents vertical displacement. f Indicates runtime, t min t represents the minimum running time. max This indicates the maximum running time.

[0025] In another implementation, the discretization steps of the vehicle stability optimal control model in step S5 are as follows: First, the integral term in the objective function is discretized, and the time interval |0,t... f Divide the interval into N smaller intervals, and the integral can be approximated as a summation:

[0026]

[0027] in, For each time length of discrete step, the optimization problem of the body stability optimal control model is converted into solving the optimal combination of discrete variables by discretization; then genetic algorithm is used for solving, including: the clamping angle θ of the walking leg is coded as the gene of the individual, the population is initialized, and each individual represents an angle combination; the objective function is defined i (i = 1, 2, 3, 4) is coded as the gene of the individual, the population is initialized, and each individual represents an angle combination; the objective function is defined The fitness of each individual is calculated, and excellent individuals are retained and generated in selection, crossover and mutation operations; the penalty function method is used to reduce the fitness of individuals that do not meet the constraints to meet the range constraint condition of the clamping angle of the walking leg and the angular velocity constraint condition; the body stability optimal control model is iteratively evolved to obtain the optimal solution of the walking leg angle combination, and the stability control of the body of the walking excavator is realized.

[0028] According to a second aspect of the embodiment of the present application, a neural network-based body posture stability control system of an unmanned walking excavator is provided, comprising: a data acquisition module, configured to scan the terrain in front of the walking excavator through a laser radar and a camera, generate terrain point cloud data, process the terrain point cloud data, and obtain terrain excitation data; a neural network posture prediction model module, configured to take the terrain excitation data and the clamping angle of the walking leg of the walking excavator as inputs, construct a neural network posture prediction model of the walking excavator, and output the pitch angle, roll angle and vertical displacement of the body of the walking excavator to predict the body posture change of the walking excavator; a body stability optimal control model module, configured to construct a body stability optimal control model of the walking excavator based on the body posture change of the walking excavator, including: optimization variables, an objective function, model constraints, and assigning values to parameters in the body stability optimal control model; a body stability optimal control model solving module, configured to discretize the variables of the body stability optimal control model at the collocation points to obtain processed variables, assign values to the processed variables at the collocation points to obtain body posture, hydraulic cylinder angle and hydraulic cylinder speed time series trajectories, calculate the objective function and constraint violation degree of the body stability optimal control model under the current driving trajectory, including the hydraulic cylinder extension length, the hydraulic cylinder extension speed, the distance from the tire center to the ground, judge whether the body stability optimal control model meets the optimization convergence condition, if yes, output the optimal control variables of the body stability optimal control model, if not, one-step iteratively optimize the state variables and control variables of the body stability optimal control model, reassign values to the processed variables at the collocation points, until the optimal control variables of the body stability optimal control model are output, to realize the stability control of the body of the walking excavator.

[0029] According to a third aspect of the present invention, an electronic device is provided, including a processor and a memory storing a program. The program includes instructions that, when executed by the processor, cause the processor to perform the steps performed by the method of the first aspect described above.

[0030] According to a fourth aspect of the present invention, a computer storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method of the first aspect described above.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] This invention establishes a mapping relationship between road surface excitation, the angle between the walking legs and the chassis, and the vehicle posture. Using the walking leg angle as the optimization variable and the pitch angle, roll angle, and vertical displacement as optimization targets, an intelligent optimization algorithm is used to achieve precise adjustment of the walking leg angle. This significantly improves the posture stability and passability of walking excavators in complex terrain, which is of great significance for improving work efficiency and quality. Attached Figure Description

[0033] 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, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0034] Figure 1 This is a schematic diagram of a walking excavator to which this invention applies.

[0035] Figure 2 This is a schematic diagram of the terrain scanning in front of the lidar vehicle according to the present invention.

[0036] Figure 3 This is a schematic diagram of the excavator body dynamics parameters of the present invention.

[0037] Figure 4 This is a flowchart illustrating the steps of the neural network-based method for stabilizing the body posture of an unmanned walking excavator according to the present invention.

[0038] Figure 5 To and Figure 4 The overall flowchart of the corresponding neural network-based unmanned walking excavator body posture stabilization control method. Detailed Implementation

[0039] To provide a clearer understanding of the technical features, objectives, and effects of the embodiments of the present invention, specific implementation methods of the embodiments of the present invention will now be described with reference to the accompanying drawings.

[0040] In this document, “exemplary” means “serving as an example, illustration or description”, and any illustrations or implementations described herein as “exemplary” should not be construed as a more preferred or advantageous technical solution.

[0041] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art should fall within the protection scope of the present invention.

[0042] The specific implementation of the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0043] The neural network-based vehicle posture stabilization control method of this invention is applicable to, for example, Figure 1 The unmanned walking excavator shown achieves stable body posture control under complex road conditions by adjusting the hydraulic cylinders of the walking legs.

[0044] This invention utilizes multiple sensors, including LiDAR and cameras, to scan the terrain in front of the excavator, collecting multi-dimensional feature information such as height, slope, and unevenness. After data registration and filtering, accurate three-dimensional point cloud data is obtained, and then analyzed into terrain excitation data containing slope, smoothness, and stiffness using mathematical fitting methods such as least squares vector machines. Next, a neural network attitude prediction model is used. This model takes the terrain excitation and the angle between the walking leg and the chassis as inputs, and pitch angle, roll angle, and vertical displacement as outputs. Through deep learning training on a large amount of operational data, the model can accurately predict vehicle attitude changes under different terrains and walking leg configurations, constructing a nonlinear mapping relationship between the walking leg angle and vehicle attitude. Furthermore, the walking leg angle is used as the optimization variable, and the stability of the vehicle attitude (pitch angle, roll angle, and vertical displacement) is used as the objective function. The model sets multiple constraints, including the range of the walking leg angle and its angular velocity limit, to construct an optimal control model for vehicle stability. The optimal control model is transformed into a nonlinear programming problem through discretization. A nonlinear optimization algorithm is used to iteratively obtain the optimal output force of the hydraulic cylinder and the optimal trajectory of the vehicle body. A schematic diagram of the lidar front terrain scan is shown below. Figure 2 As shown in the figure, the excavator body dynamics parameters of the present invention are illustrated in the diagram. Figure 3 As shown.

[0045] This invention proposes a neural network-based method for stabilizing the vehicle posture of an unmanned walking excavator. The controlled object is an ET120 walking excavator, and the terrain scanning device is a Hesai Pandar 64-line lidar. The flowchart of the proposed method is as follows: Figure 4 , Figure 5 As shown, it includes the following steps:

[0046] S1. Scan the terrain in front of the walking excavator using lidar and camera to generate terrain point cloud data, process the terrain point cloud data to obtain terrain excitation data;

[0047] S2. Using terrain excitation data and the walking leg angle of the walking excavator as input, construct a neural network attitude prediction model for the walking excavator, and output the pitch angle, roll angle and vertical displacement of the walking excavator body to predict the attitude change of the walking excavator body.

[0048] S3. Based on the changes in the body posture of the walking excavator, construct an optimal control model for the body stability of the walking excavator, including: optimization variables, objective function, and model constraints.

[0049] S4. Assign values ​​to the parameters in the optimal control model for vehicle stability.

[0050] S5. Discretize the variables of the optimal control model for vehicle stability at the collocation points to obtain the processed variables.

[0051] S6. Assign values ​​to the processed variables at the points to obtain the vehicle body pose, hydraulic cylinder angle and hydraulic cylinder speed time trajectory.

[0052] S7. Based on the vehicle body posture, hydraulic cylinder angle, and hydraulic cylinder speed time-series trajectory, calculate the objective function and constraint violation degree of the optimal vehicle stability control model under the current driving trajectory, including the hydraulic cylinder extension length, hydraulic cylinder extension speed, and the distance from the tire center to the ground.

[0053] S8. Determine whether the optimal control model for vehicle stability meets the optimization convergence condition. If it does, output the optimal control variables of the optimal control model for vehicle stability. If it does not, proceed to step S9.

[0054] S9. Iterate and optimize the state variables and control variables of the optimal control model for vehicle stability step by step, and then proceed to step S6 until the optimal control variables of the optimal control model for vehicle stability are output, so as to achieve the stability control of the walking excavator body.

[0055] Step S1 is divided into the following two steps:

[0056] (1) The terrain features such as height, slope, and unevenness in front of the excavator are collected using LiDAR and cameras. The LiDAR generates 3D point cloud data, and the camera provides texture and color features. The data space is aligned using calibration and registration techniques, and noise is removed by statistical filtering and radius filtering. Voxel Grid filtering is used for downsampling, and the RANSAC algorithm is used to segment the plane to extract terrain features, i.e., terrain excitation data (such as slope and curvature).

[0057] The preprocessed terrain features are input into a least squares support vector machine (LS-SVR). Through kernel function mapping and optimization of the objective function, the relationship between terrain features and output targets (such as terrain type and slope) is learned to achieve terrain classification or representation, providing support for excavator operations.

[0058] (2) A nonlinear regression model of the relationship between terrain points x, y and elevation z is established using least squares support vector machine:

[0059] z = f(x, y) = ω Τ φ(x,y)+b (1)

[0060] Where φ(x,y) is the feature mapping function, mapping low-dimensional features to a high-dimensional space; ω is the regression coefficient, and b is the bias term. Modeling is performed by minimizing the following optimization objective function:

[0061]

[0062] The constraints are:

[0063] z i =ω Τ φ(x i ,y i )+b+ξ i ,i=1,2,...,N (3)

[0064] Where, ξ i γ is a slack variable, and γ is a regularization parameter.

[0065] The nonlinear regression model established based on LS-SVR can be substituted into formula (1) to calculate the road surface elevation information z under any horizontal coordinate (x,y) of the terrain in front of the vehicle.

[0066] Step S2, the modeling process of the neural network attitude prediction model for the walking excavator is as follows:

[0067] (1) Based on the terrain excitation data (including slope, smoothness, concavity, etc.) obtained in step S1 and the excavator's operation data (such as walking leg angle and chassis angle), a training dataset for the neural network attitude prediction model is constructed. The input data includes terrain excitation (w1, w2, w3, w4) and the angle between the walking leg and the chassis (θ1, θ2, θ3, θ4), and the output is the vehicle attitude parameters: pitch angle θ s yaw angle and vertical displacement z s .

[0068] (2) The attitude prediction model adopts a multi-layer feedforward neural network (FNN) structure, including an input layer, several hidden layers, and an output layer. The input layer has n+4 nodes, where n represents the number of terrain excitation features, and the four nodes represent the angle between the walking leg and the chassis. The hidden layer consists of multiple neurons, and the ReLU activation function is used to enhance the fitting ability to complex nonlinear relationships. The output layer contains three nodes, corresponding to the pitch angle θ. s yaw angle and vertical displacement z s .

[0069] (3) Define the loss function and optimization objective, and train the network using supervised learning. The Mean Squared Error (MSE) loss function is chosen to minimize the error between the network's predicted output and the actual pose data. The MSE loss function is defined as:

[0070]

[0071] in, y represents the output value predicted by the model. i is the true pose value, and m is the number of samples.

[0072] The optimization method employs the Adam optimization algorithm, which iteratively updates the network weights based on gradient information to accelerate convergence. The Adam algorithm dynamically adjusts the learning rate based on the average gradients of the first and second moments, enabling the network to reach the optimal solution more quickly.

[0073] (4) Through multiple rounds of iterative training of the neural network posture prediction model, the model continuously updates its weight parameters, gradually reducing the error between the predicted output and the true value. To prevent overfitting, regularization and early stopping strategies are adopted, and the optimal model structure and hyperparameters are determined through cross-validation. After training, the neural network can accurately predict vehicle posture changes based on the input terrain stimulus and walking leg angle. By inputting specific terrain stimulus and walking leg configuration, the model outputs the vehicle's pitch angle, roll angle, and vertical displacement, successfully establishing a nonlinear mapping relationship between walking leg configuration and vehicle posture.

[0074] Step S3, the construction of the optimal control model for the vehicle stability of the walking excavator can be divided into the following three steps:

[0075] (1) Using the walking leg angle as the optimization variable and the vehicle body posture stability as the objective function, an optimal control model for vehicle body stability is constructed. The objective function is expressed as:

[0076]

[0077] Where, θ s Indicates pitch angle, Indicates the roll angle, z s t represents vertical displacement. f This indicates the excavator's operating time.

[0078] (2) To ensure that the walking excavator obtains a feasible digging trajectory and input force during its movement, relevant geometric and performance constraints are imposed during the optimization process. These model constraints include the following aspects:

[0079] During posture adjustment, the extension and retraction speed of the hydraulic cylinders of the four support legs of the walking excavator is v. hi (i = 1, 2, 3, 4) ensures that the maximum allowable lower limit v is met. hmin and maximum allowed limit v hmax between:

[0080]

[0081] During the movement, the center of the four legs of the walking excavator will be displaced by z. ui (i = 1, 2, 3, 4) and the roadside elevation w corresponding to each wheel i The difference is kept at the maximum allowable lower limit ε min and maximum allowed upper limit ε max between:

[0082]

[0083] (3) Combining steps a and b above, the optimal control model for the vehicle stability of the walking excavator is expressed as follows:

[0084] findθ i (i = 1, 2, 3, 4)

[0085]

[0086] stC ij ≤0 (i=1,2,3,4,j=1,2,3,4)

[0087] t min≤t f ≤t max

[0088] Where, θ i (i = 1, 2, 3, 4) represents the angle between the walking legs, θ s Indicates pitch angle, Indicates the roll angle, z s t represents vertical displacement. f Indicates runtime, t min t represents the minimum running time. max This indicates the maximum running time.

[0089] Step S5, the discretization steps of the optimal control model for vehicle stability are as follows:

[0090] To solve the optimization problem of the optimal control model for vehicle stability, the integral term in the objective function is first discretized. The time interval |0,t... f Divide the interval into N smaller intervals, and the integral can be approximated as a summation:

[0091]

[0092] in, Let each variable be the time length from the walk. By discretizing, the optimization problem is transformed into solving for the optimal combination of a series of discrete variables.

[0093] The following steps utilize a genetic algorithm (GA) to solve the problem, including:

[0094] walking leg angle θ i (i = 1, 2, 3, 4) are encoded as the genes of an individual, and the population is initialized so that each individual represents a combination of angles;

[0095] By defining the objective function Calculate the fitness of each individual and retain and generate superior individuals through operations such as selection, crossover, and mutation;

[0096] The penalty function method is used to reduce the fitness of individuals that do not meet the constraints, so as to satisfy the walking leg angle range constraint and the angular velocity constraint.

[0097] The optimal control model for vehicle stability is iteratively evolved to output the optimal control variables of the optimal control model for vehicle stability, so as to obtain the optimal solution of the combination of walking leg angles and realize the stability control of the walking excavator body.

[0098] This invention introduces a neural network to establish a mapping relationship between road surface excitation, walking legs, and vehicle posture, constructs a predictive model of vehicle posture under arbitrary terrain, and further establishes an optimization method for solving the problem. This is of great significance for ensuring the safety of unmanned walking excavators during autonomous operation and improving the quality of operation.

[0099] This invention also provides a neural network-based vehicle body attitude stability control system for unmanned walking excavators, comprising four parts: a road feature recognition and data acquisition module, a neural network attitude prediction model module, a vehicle body stability optimal control model module, and a vehicle body stability optimal control model solution module.

[0100] The data acquisition module is used to scan the terrain in front of the walking excavator using lidar and cameras, generate terrain point cloud data, process the terrain point cloud data, and obtain terrain excitation data.

[0101] The neural network attitude prediction model module is used to take terrain excitation data and the walking leg angle of the walking excavator as input to build a neural network attitude prediction model for the walking excavator, and output the pitch angle, roll angle and vertical displacement of the walking excavator body to predict the attitude changes of the walking excavator body.

[0102] The vehicle stability optimal control model module is used to construct the optimal control model for the vehicle stability of a walking excavator based on the changes in the vehicle's posture. It includes: optimization variables, objective function, model constraints, and parameter assignment in the optimal control model for vehicle stability.

[0103] The vehicle stability optimal control model solution module is used to discretize the variables of the vehicle stability optimal control model at collocation points to obtain the processed variables. Values ​​are assigned to the processed variables at the collocation points to obtain the time-series trajectories of the vehicle pose, hydraulic cylinder angle, and hydraulic cylinder speed. Based on the vehicle pose, hydraulic cylinder angle, and hydraulic cylinder speed time-series trajectories, the objective function and constraint violation degree of the vehicle stability optimal control model under the current driving trajectory are calculated, including the hydraulic cylinder extension length, hydraulic cylinder extension speed, and the distance from the tire center to the ground. It is then determined whether the vehicle stability optimal control model meets the optimization convergence conditions. If it does, the optimal control variables of the vehicle stability optimal control model are output. If not, the state variables and control variables of the vehicle stability optimal control model are iteratively optimized step by step, and the processed variables at the collocation points are reassigned until the optimal control variables of the vehicle stability optimal control model are output, thereby achieving the stability control of the walking excavator vehicle body.

[0104] The first part of this invention is a data acquisition module that uses multiple sensors, such as LiDAR and cameras, to scan the terrain in front of the excavator, collecting multi-dimensional feature information such as height, slope, and concavity. After data registration and filtering, accurate three-dimensional point cloud data is obtained, and then analyzed into terrain excitation data containing slope, smoothness, and stiffness using mathematical fitting methods such as least squares support vector machine. The second part is a neural network attitude prediction model module, which takes terrain excitation and the angle between the walking leg and the chassis as inputs, and pitch angle, roll angle, and vertical displacement as outputs. Through deep learning training on a large amount of operational data, the model can accurately predict the vehicle attitude changes under different terrains and walking leg configurations, and construct a nonlinear mapping relationship between the walking leg angle and the vehicle attitude. The third part is a vehicle stability optimal control model module. The walking leg angle is the optimization variable, and the vehicle attitude stability (pitch angle, roll angle, and vertical displacement) is the objective function. The model sets multiple constraints, including the range of the walking leg angle and its angular velocity limit. The fourth part is a vehicle stability optimal control model solution module. The optimal control model is transformed into a nonlinear programming problem by discretization, and the optimal motion trajectory of the vehicle body is obtained through iterative steps using a nonlinear optimization algorithm.

[0105] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0106] This invention establishes a mapping relationship between road surface excitation, the angle between the walking legs and the chassis, and the vehicle posture. Using the walking leg angle as the optimization variable and the pitch angle, roll angle, and vertical displacement as optimization targets, an intelligent optimization algorithm is used to achieve precise adjustment of the walking leg angle. This significantly improves the posture stability and passability of walking excavators in complex terrain, which is of great significance for improving work efficiency and quality.

[0107] As another example, the present invention also provides an electronic device, which will now be described as an example of a hardware device that can be applied to various aspects of the present invention, serving as a server or client of the invention. The term "electronic device" is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0108] The electronic device may include a processor, a communications interface, memory, and a communication bus.

[0109] The processor, communication interface, and memory communicate with each other via a communication bus. The communication interface is used to communicate with other electronic devices or servers.

[0110] The processor is used to execute programs, specifically the relevant steps in the above method embodiments.

[0111] Specifically, the program may include program code, which includes computer operation instructions.

[0112] The processor may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in a smart device may be of the same type, such as one or more CPUs; or they may be of different types, such as one or more CPUs and one or more ASICs.

[0113] The memory is used to store programs. The memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0114] When the program is executed by the processor, it is used to enable electronic devices to perform the neural network-based method for smooth body posture control of unmanned walking excavators according to the present invention.

[0115] Furthermore, the specific implementation of each step in the program can be found in the corresponding descriptions of the steps and units in the above method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0116] An exemplary embodiment of the present invention also provides a computer storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the methods of the various embodiments of the present invention. The corresponding process descriptions in the foregoing method embodiments can be referred to, and will not be repeated here.

[0117] The methods described above according to embodiments of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be processed by software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0118] Specific embodiments of the invention have now been described. Other embodiments are within the scope of the appended claims. In some cases, the actions described in the claims can be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing can be advantageous.

[0119] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the patent protection scope of the embodiments of the present invention should be defined by the claims.

Claims

1. A method for stabilizing the body posture of an unmanned walking excavator based on neural networks, characterized in that, The specific steps are as follows: S1. Scan the terrain in front of the walking excavator using lidar and camera to generate terrain point cloud data, process the terrain point cloud data to obtain terrain excitation data; S2. Using terrain excitation data and the walking leg angle of the walking excavator as input, construct a neural network attitude prediction model for the walking excavator, and output the pitch angle, roll angle and vertical displacement of the walking excavator body to predict the attitude change of the walking excavator body. S3. Based on the changes in the body posture of the walking excavator, construct an optimal control model for the body stability of the walking excavator, including: optimization variables, objective function, and model constraints. The steps for constructing the optimal control model for the body stability of the walking excavator are as follows: Using the walking leg angle as the optimization variable and the vehicle body posture stability as the objective function, an optimal control model for vehicle body stability is constructed. The objective function is expressed as: in, Indicates pitch angle, Indicates the roll angle. Indicates vertical displacement. Indicates the excavator's operating time; Model constraints include: During posture adjustment, the extension and retraction speed of the hydraulic cylinders of the four support legs of the walking excavator is adjusted. Guarantee at the maximum allowable lower limit and maximum allowed limit between: During the movement, the center of the four legs of the walking excavator is displaced. Roadside elevation corresponding to each wheel The difference remains at the maximum allowable lower limit. and maximum allowed limit between: Therefore, the optimal control model for the vehicle stability of a walking excavator is expressed as follows: in, Indicates the angle between the walking legs. Indicates pitch angle, Indicates the roll angle. Indicates vertical displacement. Indicates runtime. Indicates the minimum running time. Indicates the maximum running time; S4. Assign values ​​to the parameters in the optimal control model for vehicle stability. S5. Discretize the variables of the optimal control model for vehicle stability at the collocation points to obtain the processed variables. S6. Assign values ​​to the processed variables at the points to obtain the vehicle body pose, hydraulic cylinder angle and hydraulic cylinder speed time trajectory. S7. Based on the vehicle body posture, hydraulic cylinder angle, and hydraulic cylinder speed time-series trajectory, calculate the objective function and constraint violation degree of the optimal vehicle stability control model under the current driving trajectory, including the hydraulic cylinder extension length, hydraulic cylinder extension speed, and the distance from the tire center to the ground. S8. Determine whether the optimal control model for vehicle stability meets the optimization convergence condition. If it does, output the optimal control variables of the optimal control model for vehicle stability. If it does not, proceed to step S9. S9. Iterate and optimize the state variables and control variables of the optimal control model for vehicle stability step by step, and then proceed to step S6 until the optimal control variables of the optimal control model for vehicle stability are output, so as to achieve the stability control of the walking excavator body.

2. The method according to claim 1, characterized in that, The terrain excitation data in step S1 includes slope, smoothness, and convexity.

3. The method according to claim 2, characterized in that, In step S1, the methods for processing the terrain point cloud data include at least one of statistical filtering, radius filtering for noise reduction, Voxel Grid filtering for downsampling, and RANSAC algorithm for plane segmentation.

4. The method according to claim 3, characterized in that, In step S2, the method for establishing the neural network attitude prediction model of the walking excavator is as follows: A training dataset for a neural network posture prediction model is constructed based on terrain stimulus data and the walking leg angle of the excavator. The input data for the neural network posture prediction model includes terrain stimulus data. and the angle between the walking leg and the chassis The output data consists of vehicle body attitude parameters, including pitch angle. yaw angle and vertical displacement ; The neural network pose prediction model employs a multi-layer feedforward neural network structure, including an input layer, several hidden layers, and an output layer. The number of nodes in the input layer is... ,in This indicates the number of terrain excitation features. The four nodes represent the angle between the walking leg and the chassis. The hidden layer consists of multiple neurons, and the activation function is ReLU. The output layer contains three nodes, corresponding to the pitch angle. yaw angle and vertical displacement ; Define the loss function and optimization objective, train the neural network pose prediction model using supervised learning, and select the mean squared error loss function to minimize the error between the predicted output of the neural network pose prediction model and the true pose data. The mean squared error loss function is defined as follows: in, This represents the output value predicted by the neural network pose prediction model. This represents the true pose value. The number of samples; The optimization method adopts the Adam optimization algorithm, which iteratively updates the weights of the neural network pose prediction model based on gradient information to accelerate convergence; By iteratively training the neural network pose prediction model multiple times and updating the weight parameters, the error between the predicted output and the true value can be reduced. Regularization and early stopping strategies are adopted, and cross-validation is used to determine the optimal model structure and hyperparameters of the neural network pose prediction model in order to prevent overfitting of the neural network pose prediction model. After training, the neural network posture prediction model can accurately predict the body posture changes of the walking excavator based on the input terrain excitation data and the walking leg angle.

5. The method according to claim 4, characterized in that, The Adam optimization algorithm dynamically adjusts the learning rate based on the average gradient of the first and second moments to accelerate the neural network pose prediction model to reach the optimal solution.

6. The method according to claim 4, characterized in that, In step S5, the discretization steps of the vehicle body stability optimal control model are as follows: First, the integral term in the objective function is discretized, and the time interval is... Divide the data into N smaller intervals, and the integral can be approximated as a summation: in, For each time elapsed before walking, the optimization problem of the vehicle stability optimal control model is transformed into solving for the optimal combination of discrete variables through discretization. Next, a genetic algorithm is used to solve the problem, including: Angle between walking legs The genes of an individual are encoded to initialize the population, so that each individual represents a combination of angles; By defining the objective function The fitness of each individual is calculated, and superior individuals are retained and generated during selection, crossover, and mutation operations. The penalty function method is used to reduce the fitness of individuals that do not meet the constraints, so as to satisfy the walking leg angle range constraint and the angular velocity constraint. The optimal control model for vehicle stability is iteratively evolved to obtain the optimal solution for the combination of walking leg angles, thereby achieving stability control of the walking excavator vehicle.

7. A neural network-based unmanned walking excavator body posture stabilization control system, characterized in that, include: The data acquisition module is used to scan the terrain in front of the walking excavator using lidar and cameras, generate terrain point cloud data, process the terrain point cloud data, and obtain terrain excitation data. The neural network attitude prediction model module is used to take terrain excitation data and the walking leg angle of the walking excavator as input to build a neural network attitude prediction model for the walking excavator, and output the pitch angle, roll angle and vertical displacement of the walking excavator body to predict the attitude changes of the walking excavator body. The vehicle stability optimal control model module is used to construct an optimal control model for the vehicle stability of a walking excavator based on the changes in the vehicle's posture. This includes: optimization variables, objective function, model constraints, and assigning values ​​to the parameters in the optimal control model. The construction steps for the optimal control model for the vehicle stability of the walking excavator are as follows: Using the walking leg angle as the optimization variable and the vehicle body posture stability as the objective function, an optimal control model for vehicle body stability is constructed. The objective function is expressed as: in, Indicates pitch angle, Indicates the roll angle. Indicates vertical displacement. Indicates the excavator's operating time; Model constraints include: During posture adjustment, the extension and retraction speed of the hydraulic cylinders of the four support legs of the walking excavator is adjusted. Guarantee at the maximum allowable lower limit and maximum allowed limit between: During the movement, the center of the four legs of the walking excavator is displaced. Roadside elevation corresponding to each wheel The difference remains at the maximum allowable lower limit. and maximum allowed limit between: Therefore, the optimal control model for the vehicle stability of a walking excavator is expressed as follows: in, Indicates the angle between the walking legs. Indicates pitch angle, Indicates the roll angle. Indicates vertical displacement. Indicates runtime. Indicates the minimum running time. Indicates the maximum running time; The vehicle stability optimal control model solution module is used to discretize the variables of the vehicle stability optimal control model at collocation points to obtain the processed variables. Values ​​are assigned to the processed variables at the collocation points to obtain the time-series trajectories of the vehicle pose, hydraulic cylinder angle, and hydraulic cylinder speed. Based on the vehicle pose, hydraulic cylinder angle, and hydraulic cylinder speed time-series trajectories, the objective function and constraint violation degree of the vehicle stability optimal control model under the current driving trajectory are calculated, including the hydraulic cylinder extension length, hydraulic cylinder extension speed, and the distance from the tire center to the ground. It is then determined whether the vehicle stability optimal control model meets the optimization convergence conditions. If it does, the optimal control variables of the vehicle stability optimal control model are output. If not, the state variables and control variables of the vehicle stability optimal control model are iteratively optimized step by step, and the processed variables at the collocation points are reassigned until the optimal control variables of the vehicle stability optimal control model are output, thereby achieving the stability control of the walking excavator vehicle body.

8. An electronic device, characterized in that, include: processor; Memory for stored programs; The program includes instructions that, when executed by the processor, cause the processor to perform the steps of the method as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-6.

Images