Design method of track planning and safety obstacle avoidance model for tracked vehicle under uncertain model

Abstract

The application provides a design method of track planning and safe obstacle avoidance model of a tracked vehicle under an uncertain model. First, a tracked dynamics model is divided into a modelable part and an unmodeled part, the same control quantity is applied to the two models, data sets are obtained, and online learning of hyperparameters is performed, and the learned mean value is fitted into dynamics constraints in a track planner; then, the learned variance is introduced into control obstacle function design to meet the obstacle avoidance hard constraint. Then, the dynamics constraint and the obstacle avoidance hard constraint are combined for track planning to establish a cost function for optimization, so that the tracked vehicle can realize track planning and safe obstacle avoidance under an inaccurate model, and a theoretical foundation is laid for the tracked vehicle to be applied to unstructured roads such as patrol and exploration.

Description

Technical Field

[0001] This invention belongs to the field of unmanned vehicles and autonomous driving technology, specifically relating to a design method for tracked vehicle trajectory planning and safety obstacle avoidance model under an uncertain model. Background Technology

[0002] With the continuous development of key technologies such as sensing, control, and computing, and the upgrading of hardware systems, unmanned vehicles have become an emerging product in the field of intelligence, and autonomous driving technology has become a hot research topic in the scientific research community. Unmanned vehicles are divided into wheeled and tracked vehicles. Wheeled vehicles are generally used on structured roads, while tracked unmanned vehicles have advantages over wheeled vehicles in unstructured roads in the wild, such as large support area, low ground pressure, good traction and adhesion performance, less prone to slippage, small turning radius, ability to overcome uneven ground, and good self-resetting and obstacle crossing capabilities. Therefore, they are often used in fields such as agriculture, search and rescue, military, fire fighting, forestry, mining, and planetary exploration. As an important direction of engineering vehicle technology, the research on safer and more efficient tracked engineering vehicles has important practical significance, and the realization of trajectory planning and safe obstacle avoidance are key technologies for unmanned tracked vehicles.

[0003] Structured roads generally refer to well-structured highways such as expressways and urban arterial roads. These roads have clear road markings, a relatively simple background environment, and obvious geometric features. Unstructured roads, on the other hand, generally refer to roads with lower levels of structure, such as urban non-arterial roads, rural streets, and mining areas. These roads lack lane lines and clear road boundaries. In addition, due to the influence of shadows and watermarks, it is difficult to distinguish between road areas and non-road areas. At the same time, unstructured roads are usually complex and diverse in terms of road type (such as snow, grassland, and gravel) and rugged (leading to changes in slope). How to ensure trajectory planning and safe obstacle avoidance under uncertain models is a key issue.

[0004] There is a wealth of research on autonomous driving planning. However, most of this research focuses on wheeled vehicles operating in highly structured environments, such as those on highways or urban roads. In contrast, unmanned tracked vehicles operating in unstructured, rugged mountainous and other road conditions play a significant role in reducing human resources and increasing automation in military, agriculture, transportation, and other sectors. Summary of the Invention

[0005] To address the shortcomings and deficiencies of existing technologies, this invention combines Gaussian regression with model predictive control. It utilizes the mean value of Gaussian regression learning to fit and compensate for unmodeled dynamics (time-varying parameter parts) to satisfy dynamic constraints. The variance of Gaussian regression learning is introduced into the design of the control obstacle function, and the feasibility of obstacle avoidance is proven. Finally, the effectiveness of the method is verified through simulation.

[0006] The purpose of this invention is to achieve trajectory planning and safe obstacle avoidance for tracked vehicles in applications such as patrolling and exploration on unstructured roads, so that tracked vehicles can reach the expected target point in the terrain without colliding with obstacles. A design method for tracked vehicle trajectory planning and safe obstacle avoidance model under uncertain model is proposed.

[0007] Considering the time-varying parameters of tracked vehicles on unstructured roads, which makes obtaining an accurate model difficult, this invention mainly addresses the trajectory planning and obstacle avoidance of tracked vehicles under inaccurate models, and mainly includes the following steps:

[0008] First, the tracked vehicle's dynamic model is divided into a modelable part and an unmodeled part. The same control variable is applied to both models to obtain datasets, and hyperparameters are learned online. The learned mean is then fitted to the dynamic constraints in the trajectory planner. Next, the learned variance is introduced into the design of the obstacle avoidance function to satisfy the obstacle avoidance hard constraints. Finally, the cost function is established by jointly using dynamic constraints and obstacle avoidance hard constraints for trajectory planning and optimization. This enables tracked vehicles to achieve trajectory planning and safe obstacle avoidance under an imprecise model, laying a theoretical foundation for the application of tracked vehicles in unstructured road environments such as patrol and exploration.

[0009] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0010] A design method for trajectory planning and obstacle avoidance model of tracked vehicles under uncertain models is characterized by the following steps: First, the tracked vehicle dynamic model is divided into a modelable part and an unmodeled part. Based on this, the same control variable is applied to both models, and the difference in state variables between the two models is used as a dataset. Hyperparameters are learned online to obtain a Gaussian regression model, and the learned mean is used to fit the dynamic constraints in the trajectory planner. Then, the learned variance is introduced into the design of the control obstacle function to satisfy the obstacle avoidance hard constraints. Finally, the dynamic constraints and obstacle avoidance hard constraints are combined with trajectory planning to establish a cost function for optimization and solution, thereby realizing trajectory planning and safe obstacle avoidance of tracked vehicles under uncertain models.

[0011] Furthermore, the mean obtained through the Gaussian regression model is fitted to the unknown dynamics to make the trajectory planning satisfy the dynamic constraints and set as hard constraints in the nonlinear programming problem, and the objective of the trajectory planning is designed as a cost function and used as a soft constraint.

[0012] Furthermore, the specific steps include:

[0013] Step 1: Establishing the cost function for tracked vehicle modeling and trajectory planning

[0014] First, the tracked vehicle dynamics model is divided into a modelable part and an unmodeled part. Based on this, the same control quantity is applied to the two models to obtain the difference in state quantities between the two models. The current state quantity and control quantity are used as inputs, and the difference in state quantities between the two models at the next moment is used as output to obtain the dataset and perform online learning of hyperparameters. Then, a cost function is established according to the objective of trajectory planning.

[0015] Step 2: Design of Unknown Dynamic Constraints

[0016] Based on dividing the tracked vehicle model into a modeled part and an unmodeled part in step one, the time-varying model is taken as the unknown dynamic part. The error of the two models is fitted by Gaussian regression model. At the same time, the mean and variance of the model prediction control in the time domain are calculated. The mean of the output in the Gaussian regression prediction time domain is compensated into the known model as a dynamic constraint.

[0017] Step 3: Design of obstacle avoidance constraints for the control obstacle function under probabilistic state variables

[0018] For the mean and variance obtained in the prediction time domain in step two, a scaling control obstacle function constraint method is introduced to achieve deterministic obstacle avoidance. The probabilistic state variables are transformed into deterministic constraints through the quantile function of the chi-square distribution, and the variance is introduced into the control obstacle function through the properties of the chi-square distribution. This transforms the state variable constraint of a probabilistic interval into a deterministic obstacle avoidance hard constraint, so as to achieve nonlinear model predictive control optimization solution under the premise of satisfying dynamic constraints and obstacle avoidance constraints.

[0019] Under the premise that the cost function satisfies the hard constraints of dynamics, it is necessary to ensure that obstacle avoidance is satisfied. The fitting output of the Gaussian regression model is a probabilistic state variable with mean and variance. This invention introduces a scaling control obstacle function constraint method to achieve deterministic obstacle avoidance. The probabilistic state variable is transformed into a deterministic constraint by the quantile function of the Kalfar distribution, and then designed into the control obstacle function as a hard constraint. Under the premise of satisfying the dynamic constraints and obstacle avoidance constraints, nonlinear model predictive control optimization is performed to solve the problem.

[0020] Furthermore, in step one, the control quantity is the torque of the motors on both sides, and the part that does not include time-varying parameters is regarded as the modelable part; the part that includes time-varying parameters involving steering resistance coefficient ξ and ground resistance coefficient κ is regarded as the unmodeled part.

[0021] Furthermore, in step two, the difference between the two models is fitted using machine learning methods, thereby achieving the fitting of the unknown dynamics to satisfy the dynamic constraints.

[0022] First, the same control input is applied to both the model without a time-varying model and the model with a time-varying model to obtain the difference between the two models at the next time step. Using the current state variable and the control input as input to the learning dataset, and the outputs of the two models at the next time step as the output of the offline dataset, we obtain the online learning dataset.

[0023] The output has a total of n y Each dimension is processed independently. After training on the training set, the predicted output on the test set is a Gaussian distribution with mean and variance.

[0024] To avoid affecting the real-time performance of tracked vehicle trajectory tracking control, training time needs to be reduced and the dataset updated. Sparse Gaussian regression is used for sparse approximation, and the dataset is updated online based on the current state variables. The model's predicted state variables and control variables in the future time domain are used as the induced input set to ensure a well-fitting output.

[0025] Sparse Gaussian regression was performed using a completely independent training condition method.

[0026] Furthermore, in step three, the control barrier function is as follows:

[0027]

[0028] and This refers to the mean of the x-coordinate and y-coordinate of the tracked vehicle at time k, r obs This refers to the radius of the obstacle, x obs and y obs This refers to the x-coordinate and y-coordinate of the obstacle. This refers to the marginal variance under the joint distribution of x and y; The larger the value, the more stringent the constraints, but the more confident that the actual state quantity is far from the obstacle; B is the width between the centers of the two tracks, and L refers to the length of the vehicle.

[0029] Compared with existing technologies, this invention and its preferred embodiment take tracked vehicles driven by dual independent motors as the research object. The proposed model design scheme can effectively solve the problem of obstacle avoidance failure caused by time-varying parameters of tracked vehicles on unstructured roads. This invention can ensure the safety and trajectory planning of tracked vehicles under uncertain models. Attached Figure Description

[0030] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0031] Figure 1 This is a simplified model diagram of a tracked vehicle from the background technology.

[0032] Figure 2 This is a model framework diagram of the proposed method according to an embodiment of the present invention;

[0033] Figure 3 This is a diagram illustrating the effect of learning model 1 according to an embodiment of the present invention;

[0034] Figure 4 This is a diagram illustrating the effect of learning model 2 according to an embodiment of the present invention;

[0035] Figure 5 This is a simulated three-dimensional ground map according to an embodiment of the present invention;

[0036] Figure 6 It is a graph of ordinary model predictive control and control obstacle function trajectory planning and obstacle avoidance task;

[0037] Figure 7 It is a distance map of the three obstacles under ordinary model predictive control and obstacle control function;

[0038] Figure 8 This is a diagram illustrating the Gaussian regression model predictive control and the trajectory planning and obstacle avoidance task of the control obstacle function, as described in this invention.

[0039] Figure 9 This is a distance map between the Gaussian regression model predicting control and the obstacle avoidance function in an embodiment of the present invention and three obstacle avoidance objects. Detailed Implementation

[0040] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below for detailed explanation:

[0041] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0042] like Figure 2 As shown in the embodiments of the present invention, the design scheme of tracked vehicle trajectory planning and safety obstacle avoidance model under uncertain model can be roughly divided into the establishment of modeling and trajectory planning objective function, the design of unknown dynamic constraints, and the design of obstacle avoidance constraints for control obstacle function under probabilistic state variables.

[0043] Step 1: Establishing the cost function for tracked vehicle modeling and trajectory planning

[0044] In this embodiment, Figure 1 The simplified model of the tracked vehicle shown can be represented as:

[0045]

[0046] Where w is the angular velocity of the tracked vehicle. x is the heading angle of the tracked vehicle, x0 is the abscissa of the vehicle's center of mass in the geodetic coordinate system, y0 is the ordinate of the vehicle's center of mass in the geodetic coordinate system, v is the instantaneous velocity in the geodetic coordinate system, and T is the y-coordinate of the vehicle's center of mass in the geodetic coordinate system. l It is the driving torque of the left motor, T r It is the driving torque of the right motor, and the ground resistance coefficient κ = sinα + f cosα, where α refers to the ground slope, sgn represents the sign function, and i g This represents the overall transmission ratio of the motor, ε refers to the coefficient of inertia, J refers to the moment of inertia of the tracked vehicle about the Z-axis, B is the width between the centers of the two tracks, g refers to the acceleration due to gravity, L refers to the vehicle length, m is the vehicle mass, and r refers to the radius of the drive wheel. The steering resistance coefficient is related to the road type, and λ refers to the longitudinal offset.

[0047] The research object of this invention is a tracked vehicle with independent electric drive on both sides. The control quantity is the torque of the motors on both sides. Therefore, expression (1) is written as U = [T] r ,T l ] T in

[0048]

[0049]

[0050] Where f(X,U) does not contain time-varying parameters and belongs to the modelable part, g(X) contains time-varying parameters, including the steering drag coefficient ξ and the ground drag coefficient κ. If the parameters in g(X) are known, the objective function of the trajectory planner with obstacle avoidance can be established, as shown in formula (4):

[0051]

[0052] N represents the prediction time domain of model predictive control, J t This refers to the cost function, X. t+k|t This refers to the state variables, U, that the model predicts and controls within the time domain k. t+k|t This refers to the control quantity of k in the time domain, X final The state variable of the final target point is represented by Q and R, which represent the weight of the difference between the state variable of the target point and the state variable of the target point. P refers to the terminal cost weight. The first constraint is the dynamic constraint, and the last constraint is the obstacle avoidance constraint. 1 > γ > 0 is the decay factor of the control obstacle function. It can be seen that the premise of the effectiveness of formula (4) is that an accurate model can be obtained. If there is an unknown dynamic part, the dynamic constraint and obstacle avoidance constraint cannot be satisfied.

[0053] Step 2: Design of Unknown Dynamic Constraints

[0054] In step one, the tracked vehicle model has been divided into a modeled part and an unmodeled part. In this embodiment, machine learning methods are used to fit the difference between the two models, thereby achieving the fitting of unknown dynamics to meet the dynamic constraints.

[0055] First, apply the same control quantity to... and The difference between the two models at the next time step is obtained from the two models. The current state variable and control variable are used as inputs to the learning dataset, and the outputs of the two models at the next time step are used as the outputs of the offline dataset to obtain the online learning dataset.

[0056]

[0057] The output has a total of n y Each dimension is processed independently. After training on the training set, the input X is used on the test set. t (X t (where n is a row vector) in dimension n (where n∈{1,......,n) y The predicted output of} is a Gaussian distribution with mean and variance:

[0058]

[0059] in k n (X,X t ) is k n (X t The transpose of (X), It is a positive definite matrix. Both μ and μ are variance hyperparameters, and both can be learned and trained. n (X t ) and Σ n (x t ) represents the predicted mean and variance.

[0060] To avoid affecting the real-time performance of tracked vehicle trajectory tracking control, it is necessary to reduce training time and update the dataset. To this end, this invention uses sparse Gaussian regression for sparse approximation and updates the dataset online based on the current state variables. Sparse Gaussian regression mainly establishes an approximation matrix by setting up an induced input point set, and the fitting accuracy of the induced input set is relatively high. Based on this property, this invention uses the state variables and control variables predicted by the model in the future time domain as the induced input set to ensure a better fit output.

[0061] Induced input N PFor the prediction time domain of MPC, this invention employs a fully independent training conditional (FITC) method to complete sparse Gaussian regression:

[0062]

[0063] in, and This refers to the predicted sparse mean and variance, and the generalized expression for the approximate matrix H is: G in Formula 7 is shown in Formula 8.

[0064]

[0065] To ensure the real-time requirements of the control task, in this invention, diag is a diagonal matrix. To train the hyperparameters, the online dataset is limited to a maximum of 100, while the dataset is updated continuously and the hyperparameters are learned.

[0066] After obtaining the mean and variance of the first future time domain, the mean and variance of the prediction time domain are obtained through formula (9):

[0067]

[0068] in μ g Σ represents the mean of the unknown dynamic state variables that need to be compensated. g The mean of the unknown dynamic state variables that need to be compensated represents the variance of the unknown dynamic state variables that need to be compensated. U refers to the mean state quantity predicted at time k+1. k This refers to the control quantity at time k; therefore, the dynamic constraint expression in this embodiment is:

[0069]

[0070] Step 3: Design of obstacle avoidance constraints for the control obstacle function under probabilistic state variables

[0071] In step two, the mean and variance in the prediction time domain are obtained. In this embodiment, the obtained variance is introduced into the control barrier function. The state variable learned through the Gaussian regression process is a probabilistic state variable. For the convenience of optimization, a deterministic state quantity (i.e., the mean) is usually set. However, this state quantity does not represent the true state quantity. The interval of the mean and variance represents the true state quantity falling into this interval with a corresponding probability. Here, this embodiment introduces a theorem: Set a two-dimensional random vector x ~ (μ, ∑) and a set B x (r)={x|||xs obsIf ||≥c}, Then we can conclude that P(x∈B) x (r))≥p,

[0072] The p-quantile function refers to the chi-square distribution of a two-dimensional variable, where p is the probability that the condition is met, and λ is the probability that the condition is met. max (∑) refers to the largest eigenvalue of ∑, and ∑ refers to the variance of the data distribution.

[0073] prove: It is the confidence interval of x at the p level. It can be directly derived from the properties of the chi-square distribution, therefore... Create a set and Then there is Then there is:

[0074]

[0075] if In this embodiment, P(||xx) can be obtained from formula (11). obs ||≥r)≥p.

[0076] Based on the above conclusions, the control obstacle function can be designed as follows in this embodiment:

[0077]

[0078] and This refers to the mean of the x-coordinate and y-coordinate of the tracked vehicle at time k, r obs This refers to the radius of the obstacle, x obs and y obs This refers to the x-coordinate and y-coordinate of the obstacle. Figure 1 It is shown in the middle, This refers to the marginal variance under the joint distribution of x and y; from the above theorem, we can conclude that... The larger the value, the more stringent the constraints, but the more certain it is that the true state quantity is far from the obstacle.

[0079] Therefore, the final expressions for the objective function and constraints obtained in this embodiment are designed as follows:

[0080]

[0081] This refers to the mean state predicted at time k, and h represents the designed control barrier function.

[0082] Simulation comparison

[0083] To verify the effectiveness of the method of the present invention, a method as follows was created. Figure 5 The three-dimensional ground shown represents an uneven, unstructured road. The changes in time-varying parameters represent different road conditions. The parameters of the tracked vehicle in this invention are set as shown in Table 1, and the changes in time-varying parameters are shown in Table 2. In the case of time-varying parameters, the slope is obtained through the three-dimensional ground. The model containing time-varying parameters represents the real tracked vehicle, and the model without time-varying parameters represents the uncertain model.

[0084] Table 1. Simulation Model Parameters of Tracked Vehicle

[0085] m 75 kg ε 1.05 / <![CDATA[i g ]]> 8.25 / r 0.265 m L 1.22 m B 0.45 m J 200 kg.m^2

[0086] Table 2 Time-varying parameters

[0087]

[0088]

[0089] This invention compares the designed method (represented by GP-MPC-CBF) with the methods of ordinary model predictive control and control barrier function (represented by MPC-CBF). The ordinary control barrier function method uses an uncertain model that does not include time-varying parameters. Figure 3 and Figure 4 The diagrams illustrate the state variable errors caused by time-varying parameters. It shows that the model fitted using Gaussian regression (GP) is roughly the same as the true model values, while inaccurate models introduce errors. In cases such as... Figure 5 Under the terrain shown, Figure 6 and Figure 7 This indicates the MPC-CBF method, from Figure 7 It can be seen that using ordinary model prediction control and obstacle control function under uncertain models will lead to obstacle avoidance failure; Figure 8 and Figure 9 The method used in this invention shows that the trajectory planning task was completed without collision with obstacles, verifying that this invention can solve the problem of potential obstacle avoidance failure in trajectory planning under uncertain models.

[0090] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0091] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0095] This patent is not limited to the above-described preferred embodiments. Anyone can derive design methods for tracked vehicle trajectory planning and safety obstacle avoidance models under various uncertain models based on the inspiration of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A method for designing a tracked vehicle trajectory planning and safe obstacle avoidance model under an uncertain model, characterized in that: First, the tracked vehicle dynamics model is divided into a modelable part and an unmodeled part. Based on this, the same control variable is applied to both models, and the difference in state variables between the two models is used as a dataset. Hyperparameters are then learned online to obtain a Gaussian regression model. The learned mean is used to fit the dynamic constraints in the trajectory planner. Then, the learned variance is introduced into the design of the obstacle avoidance function to meet the obstacle avoidance hard constraints. Finally, the dynamic constraints and obstacle avoidance hard constraints are combined with trajectory planning to establish a cost function for optimization, thereby enabling the tracked vehicle to achieve trajectory planning and safe obstacle avoidance under an inaccurate model. After obtaining the mean and variance of the first future time domain, the mean and variance in the prediction time domain are obtained using the following formula: in , This represents the mean of the unknown dynamic state variables that need to be compensated. The mean of the unknown dynamic state variables that need to be compensated represents the variance of the unknown dynamic state variables that need to be compensated. It refers to the mean state quantity predicted at time k+1. This refers to the control quantity at time k; The control barrier function is as follows: and This refers to the mean of the x and y coordinates of the tracked vehicle at time k. This refers to the radius of the obstacle. and This refers to the x-coordinate and y-coordinate of the obstacle. This refers to the marginal variance under the joint distribution of x and y; The larger the value, the more stringent the constraints, but the more confident that the actual state quantity is kept away from the obstacle; B is the width between the centers of the two tracks, and L refers to the vehicle length; It refers to The largest eigenvalue, This refers to the variance of the data distribution; The p-quantile function of the chi-square distribution of a two-dimensional variable. It refers to the probability that the condition is met; The expressions for the objective function and constraints used are as follows: This refers to the mean state predicted at time k, h represents the designed control barrier function, N represents the prediction time domain of the model predictive control, Q and R represent the weights of the state differences from the target point, and P refers to the terminal cost weight. It is the decay factor that controls the barrier function.

2. The design method for tracked vehicle trajectory planning and safe obstacle avoidance model under an uncertain model according to claim 1, characterized in that: The mean obtained by the Gaussian regression model is fitted to the unknown dynamics to make the trajectory planning satisfy the dynamic constraints and set as hard constraints in the nonlinear programming problem. The objective of the trajectory planning is designed as a cost function and used as a soft constraint.

3. The design method for tracked vehicle trajectory planning and safe obstacle avoidance model under an uncertain model according to claim 1, characterized in that: Specifically, the following steps are included: Step 1: Establishing the cost function for tracked vehicle modeling and trajectory planning First, the tracked vehicle dynamics model is divided into a modelable part and an unmodeled part. Based on this, the same control quantity is applied to the two models to obtain the difference in state quantities between the two models. The current state quantity and control quantity are used as inputs, and the difference in state quantities between the two models at the next moment is used as output to obtain the dataset and perform online learning of hyperparameters. Then, a cost function is established according to the objective of trajectory planning. Step 2: Design of Unknown Dynamic Constraints Based on dividing the tracked vehicle model into a modeled part and an unmodeled part in step one, the time-varying model is taken as the unknown dynamic part. The error of the two models is fitted by Gaussian regression model. At the same time, the mean and variance of the model prediction control in the time domain are calculated. The mean of the output in the Gaussian regression prediction time domain is compensated into the known model as a dynamic constraint. Step 3: Design of obstacle avoidance constraints for the control obstacle function under probabilistic state variables For the mean and variance obtained in the prediction time domain in step two, a scaling control obstacle function constraint method is introduced to achieve deterministic obstacle avoidance. The probabilistic state variables are transformed into deterministic constraints through the quantile function of the chi-square distribution, and the variance is introduced into the control obstacle function through the properties of the chi-square distribution. This transforms the state variable constraint of a probabilistic interval into a deterministic obstacle avoidance hard constraint, so as to achieve nonlinear model predictive control optimization solution under the premise of satisfying dynamic constraints and obstacle avoidance constraints.

4. The design method for tracked vehicle trajectory planning and safe obstacle avoidance model under an uncertain model according to claim 3, characterized in that: In step one, the control quantity is the torque of the two motors, and the part that does not include time-varying parameters is taken as the modelable part; this includes the part involving the steering resistance coefficient. and ground drag coefficient The time-varying parameters are treated as the unmodeled portion.

5. The design method for tracked vehicle trajectory planning and safe obstacle avoidance model under an uncertain model according to claim 4, characterized in that: In step two, the difference between the two models is fitted using machine learning methods, thereby achieving the fitting of the unknown dynamics to satisfy the dynamic constraints. First, the same control input is applied to both the model without a time-varying model and the model with a time-varying model to obtain the difference between the two models at the next time step. Using the current state variable and the control input as input to the learning dataset, and the outputs of the two models at the next time step as the output of the offline dataset, we obtain the online learning dataset. Output total Each dimension is processed independently. After training on the training set, the predicted output on the test set is a Gaussian distribution with mean and variance. To avoid affecting the real-time performance of tracked vehicle trajectory tracking control, training time needs to be reduced and the dataset updated. Sparse Gaussian regression is used for sparse approximation, and the dataset is updated online based on the current state variables. The model's predicted state variables and control variables in the future time domain are used as the induced input set to ensure a well-fitting output. Sparse Gaussian regression was performed using a completely independent training condition method.

Images