Intelligent vehicle lane changing scene small sample real vehicle data interactive world-like model construction method and system

Abstract

The invention discloses an intelligent vehicle lane changing scene small sample real vehicle data interactive world-like model construction method and system, and relates to the technical field of intelligent vehicle virtual testing, the technical scheme is characterized in that a genetic algorithm and Bayesian update joint algorithm is provided, real vehicle data is used to carry out parameter optimization on an intelligent driver model, and the real vehicle data is obtained; and a scene-based interactive control model is constructed. Meanwhile, a long-short-term memory network is used for identifying intentions of other vehicles, an interactive trajectory planning model is constructed, model parameters can be adjusted to achieve dynamic interaction with an interactive control model, richer simulation interaction scenes are achieved, and the effect of world model simulation interaction is achieved. According to the invention, a more real simulation world effect is provided, so that the trajectory planning capability of the intelligent vehicle in a lane changing interaction scene is finally improved.

Description

Technical Field

[0001] This invention relates to the field of virtual testing technology for intelligent vehicles, and more specifically, to a method and system for constructing an interactive world model based on small sample real vehicle data in lane-changing scenarios for intelligent vehicles. Background Technology

[0002] Virtual testing of intelligent vehicles enables large-scale scenario testing in a low-cost and efficient manner, and supports the safe reproduction of extremely dangerous scenarios, making it a key link in promoting its transition from simulation development to real-world application. However, accurately reproducing the dynamic interaction behavior between vehicles in a simulation environment remains a core problem that urgently needs to be solved.

[0003] Currently, widely used solutions mainly rely on two technical approaches: one is data playback-based simulation, which repeatedly plays back collected real traffic data. Its limitation lies in the fixed behavior of vehicles in the environment, making active interaction with intelligent vehicles (the host vehicle) impossible. The other is building a massive, globally comprehensive world model, relying on massive amounts of data to achieve interactive simulation. However, this approach typically faces challenges such as high training costs, long cycles, and complex parameter tuning. Especially for scenarios like lane changes, which are short-duration and highly interactive, using a global world model for modeling often lacks economic efficiency and practicality.

[0004] To address the aforementioned issues, this invention proposes a method and system for constructing an interactive world-like model based on small-sample real-vehicle data. Furthermore, this system can make greater use of replay-based datasets, adjusting the motion parameters of intelligent vehicles to adapt the motion behavior of traffic participants throughout the transportation system, thus making the original data interactive and providing richer interactive scene effects.

[0005] This method uses a small amount of real vehicle data to create interactive models of vehicles in the environment, enabling them to dynamically respond to the behavior of the host vehicle and have an autonomous interaction mechanism. While improving the realism, safety and human-likeness of the simulation, it effectively avoids the high complexity and high cost of traditional world models.

[0006] The system designs a trajectory planning sub-model and builds the application framework of the interactive model. By adjusting the sub-model parameters and interacting with the interactive environment model, the system generates diverse test scenarios, enhancing data reuse efficiency and reducing interaction complexity. This system achieves high-fidelity interactive simulation in a lightweight manner, ultimately improving the trajectory planning capabilities of intelligent vehicles in real lane-changing interaction scenarios. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for constructing an interactive world-like model based on small-sample real-vehicle data for lane-changing scenarios of intelligent vehicles. This invention builds an interactive control model based on an intelligent driver model, and proposes a joint optimization algorithm based on genetic algorithms and Bayesian updates to efficiently optimize model parameters using small-sample real-vehicle data. Furthermore, an interactive trajectory planning sub-model is constructed, and dynamic interaction with the interactive control model is achieved by adjusting the model parameters, thus forming a complete world-like model system. This method can fully utilize a large amount of existing real-vehicle test data, significantly improve the realism and interactive effect of the simulation environment by endowing the data with interactive capabilities, and ultimately effectively enhance the trajectory planning capability of intelligent vehicles in real lane-changing interactive scenarios.

[0008] The above-mentioned technical objective of the present invention is achieved through the following technical solution: The first aspect of this invention provides a method and system for constructing a world-like model based on small-sample real-vehicle data in intelligent vehicle lane-changing scenarios, comprising the following steps: Acquire driving status information, which includes longitudinal position, lateral position, speed, and acceleration; An interactive control model is constructed, which is based on the intelligent driver model and incorporates interactive factors. ; Based on a multi-objective parameter optimization algorithm, the parameters of the interactive control model are optimized using driving state information to obtain the optimal parameter model. An interactive trajectory planning sub-model is constructed, which includes an intent prediction module. The intent prediction module is used to predict the driving state of vehicles behind the target lane. The position and speed of both vehicles are input, and the driving intent is output. The vehicles behind are driven by an optimal parameter model. Based on the driving intention of the interactive vehicle, trajectory sampling is adaptively performed to generate a desired trajectory that integrates the perception of driving intention.

[0009] In conjunction with the first aspect, the present invention is further configured such that the interaction factor for

[0010] Among them, 0 represents a scenario with no preceding vehicle, 1 represents a scenario with a fixed preceding vehicle, and 2 represents a scenario where the preceding vehicle changes.

[0011] In conjunction with the first aspect, the present invention is further configured such that: the interactive control model is

[0012] in, Indicates the maximum acceleration. This indicates the current actual speed of the vehicle. Indicates the maximum expected speed. Indicates acceleration sensitivity (set to 4). Represents the minimum expected distance. Indicates the expected headway. This indicates the speed difference between the two vehicles. This indicates the current following distance of the vehicle.

[0013] In conjunction with the first aspect, the present invention is further configured such that: the multi-objective parameter optimization algorithm combines genetic algorithm and Bayesian update, and the parameters to be optimized are...

[0014] The cost function is

[0015] in, and These are the weighting coefficients. and The output values ​​are for interactive model simulation. and These are actual observed values.

[0016] In conjunction with the first aspect, the present invention is further configured as follows: ; in, These are constant coefficients. For the corresponding time. Taking the lane-changing and overtaking interaction scenario as an example, an overly aggressive lane-changing trajectory will cause vehicles behind in the target lane to slow down and give way to ensure safety. The lane-changing trajectory can be adjusted to reduce this impact.

[0017] In conjunction with the first aspect, the present invention is further configured such that: the new endpoint of the desired trajectory is ; in, This is the original endpoint location. To plan the time domain, The calculation method for the position change predicted based on driving intention is as follows: ; in, The attenuation coefficient is... This represents the predicted longitudinal acceleration of vehicles behind in the target lane.

[0018] In conjunction with the first aspect, the present invention is further configured such that: the method for obtaining driving status information is: to obtain real vehicle data of lane-changing scenarios in a real road environment, and to extract driving status information by taking the time point when the lane-changing vehicle crosses the lane line as the intermediate time.

[0019] A second aspect of the present invention also provides an apparatus / device / system for constructing a world-like model of small-sample real-vehicle data interaction in intelligent vehicle lane-changing scenarios, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the above method.

[0020] A third aspect of the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0021] A fourth aspect of the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0022] In summary, the present invention has the following beneficial effects: This invention utilizes a combined genetic algorithm and a Bayesian update algorithm to optimize the parameters of an intelligent driver model, constructing a scenario-based interactive control model. Simultaneously, a Long Short-Term Memory (LSTM) network is used for other vehicle intent recognition, and an interactive trajectory planning model is built. Adjustable model parameters enable dynamic interaction with the interactive control model, achieving richer simulation interaction scenarios and reaching the effect of world model-based interactive simulation. This invention provides a more realistic simulation world effect, ultimately improving the trajectory planning capability of intelligent vehicles in lane-changing interaction scenarios. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for constructing a small-sample real-vehicle data interactive world model for intelligent vehicle lane-changing scenarios according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of a method for constructing a world-like model of small sample real vehicle data interaction in a lane-changing scenario for intelligent vehicles, according to Embodiment 2 of the present invention. Detailed Implementation

[0024] The technical solutions of 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 without creative effort are within the scope of protection of the present invention.

[0025] Example 1: A method for constructing a world-like model based on small-sample real-vehicle data interaction in intelligent vehicle lane-changing scenarios, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect real vehicle data for lane-changing scenarios Real-world vehicle data from lane-changing scenarios in actual road conditions is acquired to optimize model parameters. The primary vehicle is defined as the intelligent vehicle changing lanes, and other vehicles are defined as environmental vehicles in the target lane. Key driving state information related to the primary vehicle and environmental vehicles is extracted, specifically including longitudinal position. Horizontal position ,speed acceleration This ensures a comprehensive depiction of the vehicle's dynamic interaction process. When extracting data, the time when a vehicle crosses the lane line during a lane change is used as the intermediate moment to extract vehicle driving data that covers the entire lane change process.

[0026] Step 2: Construct an interactive control model Interactive models can provide interactive responses that closely resemble real driving behavior for the research of intelligent vehicle interaction algorithms, and provide dynamic parameters for the models. For extracted lane-changing scenario driving data, given the driving trajectories of vehicles changing lanes in the data, the driving trajectory generated by the interactive model can simulate the actual driving trajectory of the following vehicles in the data as closely as possible. The interactive behavior generated by vehicle control models using fixed or empirical parameters cannot conform to the driving behavior of human drivers, and the vehicle's control behavior varies under different actual interaction scenarios, making it necessary to adjust the control parameters according to the interaction scenario.

[0027] This patent introduces interaction factors. Its definition is (1) Based on the intelligent driver model, a scenario-based intelligent driver model (SIDM) is established, which can adaptively adjust the model parameters according to the interaction scenario. Its mathematical expression is shown in equation (2).

[0028] (2) in, Indicates the maximum acceleration. This indicates the current actual speed of the vehicle. Indicates the maximum expected speed. Indicates acceleration sensitivity (set to 4). Represents the minimum expected distance. Indicates the expected headway. This indicates the speed difference between the two vehicles. This indicates the current following distance of the vehicle.

[0029] Step 3: Parameter Optimization To improve the driving behavior simulation accuracy of the interactive model obtained in step two, a multi-objective parameter optimization algorithm combining genetic algorithm and Bayesian update is adopted. Genetic algorithm is a powerful global optimization algorithm that mimics the natural selection process, effectively exploring a large number of possible parameter combinations in an unknown parameter space, avoiding getting trapped in local optima. Bayesian update, on the other hand, is a sequence update method based on Bayes' theorem, which can perform efficient local fine-tuning optimization based on the prior distribution provided by the genetic algorithm.

[0030] Step (1): Determine the optimization parameters and construct the objective function. Based on the interactive model constructed in step two, the parameters to be optimized are determined as follows: .

[0031] The goal of parameter optimization is to achieve the desired observation step size. At that time, the vehicle behavior driven by the interactive model should match the driving state information obtained in step one as closely as possible, specifically by minimizing the vehicle's following speed error and following distance error, with the error vector e. The definition is as follows: (3) The cost function can be defined as (4) in, and These are the weighting coefficients. and The output values ​​are for interactive model simulation. and The goal is to find the optimal parameters based on actual observations. Minimize the cost function.

[0032] Step (2): Determine the system state variables and update matrix Equation (2) shows that the system directly depends on the following distance. Driving speed Speed ​​difference between the two vehicles Therefore, the system state vector The input variables are defined as shown in equation (5). Speed ​​of the vehicle in front .

[0033] (5) Therefore, given a time interval The state update equation after linearization approximation of the system is constructed as follows: (6) State transition matrix The definition is as follows: (7) Input matrix The definition is as follows: (8) in, (9) Step (3): Genetic Algorithm (GA) Optimization Based on the parameters to be optimized determined in step (1), the population is initialized by uniformly and randomly sampling within its value range to obtain the initial population. .

[0034] (10) in, Indicates the first in the population Individual, Indicates population size.

[0035] For each individual in the population, its fitness is assessed, and the calculation method is shown in Equation (11).

[0036] (11) A tournament selection method is used, randomly sampling from the population to select the individual with the highest fitness as the parent, repeating this process until the desired target number of generations is reached. Two parent individuals are randomly selected, and a simulated binary crossover operation is used to generate two offspring individuals. Random perturbations are then applied to the offspring individuals' parameters with a certain probability to increase population diversity. This selection, crossover, mutation, and evaluation process is repeated until the maximum number of generations is reached, and the individual with the highest fitness is identified as the optimal parameter output by the genetic algorithm. Retain individuals including those with the best parameters A select few elite individuals are used for Bayesian updates, defined as follows: (12) Step (4): Bayesian Update First, according to the loss function Define the objective function Based on the distribution range of the parameters to be optimized, a Gaussian process is used to construct the prior distribution of the objective function. Then, the elite subset of individuals generated by GA. The initial observation data were obtained by constructing the corresponding cost value. Based on the Gaussian process modeling assumption, the observed data are defined. In the objective function The likelihood of the following It is important to note that the observation data It grows dynamically during the iterative optimization process. Finally, the posterior distribution is calculated. As shown in equation (13), where Marginal likelihood represents the observed data during the Gaussian process. The joint probability of occurrence.

[0037] (13) Define the acquisition function in the parameter space Its definition is shown in equation (14), where, Indicates the current observation data The minimum cost loss is determined by selecting the sampling point that maximizes this loss as the evaluation parameter. Then, simulation is performed based on the system update matrix from step (2), and its loss function is calculated. This loss function is then used as the new data point to update the observed data. And update the posterior distribution .

[0038] (14) This iterative process of selecting sampling points using the acquisition function, calculating the cost of system simulation, and updating data and Gaussian posterior is repeated until the optimal parameters are obtained. To obtain the optimal parameter model.

[0039] Step 4: Construct a trajectory planning sub-model based on an interactive model This step's design intent prediction module is used to predict the driving state of vehicles behind in the target lane. The vehicles behind in the target lane are driven by the optimal parameter model obtained in step three. This module introduces a Long Short-Term Memory (LSTM) network. Input features include the position and speed of both parties in the interaction. The network consists of three hidden layers and one fully connected output layer. Each hidden layer has 100 neurons with a random deactivation probability of 0.03, and the fully connected layer has two neurons activated using a linear activation function. The output is the predicted lateral and longitudinal acceleration values ​​(driving intent).

[0040] This step, based on the driving intent of the interactive vehicle, adaptively samples the trajectory to generate a desired trajectory that integrates driving intent perception. This guides the vehicle from its current state to the target state, achieving a more user-friendly lane-changing interaction. The generated lateral and longitudinal trajectory curves are obtained using fifth-order polynomials, respectively. (15) in, These are constant coefficients. For the corresponding time. Taking the lane-changing and overtaking interaction scenario as an example, an overly aggressive lane-changing trajectory will cause vehicles behind in the target lane to slow down and yield to ensure safety. The lane-changing trajectory can be adjusted to reduce this impact. When generating the trajectory, the new endpoint of the desired trajectory incorporates the longitudinal driving intention prediction result of other vehicles output by the intention prediction module, represented as: (16) in, This is the original endpoint location. To plan the time domain, The calculation method for the position change predicted based on driving intention is as follows: (17) in, The attenuation coefficient is... This represents the predicted longitudinal acceleration of vehicles behind in the target lane.

[0041] The following cost function is used to select the generated target trajectory: (18) in, These are the weight coefficients for each cost item, and the corresponding cost function for each weight is the path length cost. Used to avoid excessively long paths and the cost of lateral offset. To avoid excessive lateral deviation and longitudinal speed cost Used to optimize driving speed and acceleration cost Used to limit acceleration amplitude, optimize dynamic safety and energy consumption, and mitigate safety costs. Used to avoid track collisions, at the cost of comfort. Used to improve ride comfort. The calculation methods for each cost are as follows: (19) (20) (twenty one) (twenty two) (twenty three) (twenty four) in, It is the time domain of trajectory planning. , It represents the current trajectory's longitudinal and lateral positions in the road coordinate system. , Indicates longitudinal and lateral velocities. , Indicates longitudinal and lateral acceleration, and For longitudinal and lateral acceleration, It is the center position of the target lane and the desired speed. It is the set safe distance minus the trajectory and the first The difference in minimum distance between obstacles.

[0042] Example 2: A method for constructing a world-like model based on small-sample real-vehicle data interaction in intelligent vehicle lane-changing scenarios, such as... Figure 2 As shown, it includes the following steps: Includes the following steps: S100. Obtain driving status information, which includes longitudinal position, lateral position, speed, and acceleration; S200. Construct an interactive control model, and based on a multi-objective parameter optimization algorithm, use driving state information to optimize the parameters of the interactive control model to obtain the optimal parameter model; S300. Construct an interactive trajectory planning sub-model, which includes an intent prediction module. The intent prediction module is used to predict the driving state of vehicles behind the target lane. The position and speed of both vehicles are input, and the driving intent is output. The vehicles behind are driven by an optimal parameter model. Based on the driving intent of the interactive vehicles, trajectory sampling is adaptively performed to generate a desired trajectory that integrates driving intent perception.

[0043] In step S100 of this embodiment, the method for obtaining driving status information is as follows: real vehicle data of lane-changing scenarios in real road environments are obtained, and driving status information is extracted by taking the time point when the lane-changing vehicle crosses the lane line as the intermediate time.

[0044] In step S200 of this embodiment, the interactive control model is based on the intelligent driver model and introduces interactive factors. .

[0045] Interaction factor for ; Among them, 0 represents a scenario with no preceding vehicle, 1 represents a scenario with a fixed preceding vehicle, and 2 represents a scenario where the preceding vehicle changes. Interactive control model is ; in, Indicates the maximum acceleration. This indicates the current actual speed of the vehicle. Indicates the maximum expected speed. Indicates acceleration sensitivity (set to 4). Represents the minimum expected distance. Indicates the expected headway. This indicates the speed difference between the two vehicles. This indicates the current following distance of the vehicle.

[0046] The specific optimization steps include: S210: Determine the optimization parameters and construct the objective function. Based on the interactive model constructed in step two, the parameters to be optimized are determined as follows: ; The goal of parameter optimization is to achieve the desired observation step size. At that time, the vehicle behavior driven by the interactive model should match the driving state information obtained in step one as closely as possible, specifically by minimizing the vehicle's following speed error and following distance error, with the error vector e. The definition is as follows: (3) The cost function can be defined as (4) in, and These are the weighting coefficients. and The output values ​​are for interactive model simulation. and The goal is to find the optimal parameters based on actual observations. Minimize the cost function.

[0047] S220: Determine system state variables and update matrix This system relies directly on following distance. Driving speed Speed ​​difference between the two vehicles Therefore, the system state vector The input variables are defined as shown in equation (5). Speed ​​of the vehicle in front .

[0048] (5) Therefore, given a time interval The state update equation after linearization approximation of the system is constructed as follows: (6) State transition matrix The definition is as follows: (7) Input matrix The definition is as follows: (8) in, (9) S230: Genetic Algorithm (GA) Optimization Based on the parameters to be optimized determined in step S210, the population is initialized by uniformly and randomly sampling within its value range to obtain the initial population. .

[0049] (10) in, Indicates the first in the population Individual, Indicates population size.

[0050] For each individual in the population, its fitness is assessed, and the calculation method is shown in Equation (11).

[0051] (11) A tournament selection method is used, randomly sampling from the population to select the individual with the highest fitness as the parent, repeating this process until the desired target number of generations is reached. Two parent individuals are randomly selected, and a simulated binary crossover operation is used to generate two offspring individuals. Random perturbations are then applied to the offspring individuals' parameters with a certain probability to increase population diversity. This selection, crossover, mutation, and evaluation process is repeated until the maximum number of generations is reached, and the individual with the highest fitness is identified as the optimal parameter output by the genetic algorithm. Retain individuals including those with the best parameters A select few elite individuals are used for Bayesian updates, defined as follows: (12) S240: Bayesian Update First, according to the loss function Define the objective function Based on the distribution range of the parameters to be optimized, a Gaussian process is used to construct the prior distribution of the objective function. Then, the elite subset of individuals generated by GA. The initial observation data were obtained by constructing the corresponding cost value. Based on the Gaussian process modeling assumption, the observed data are defined. In the objective function The likelihood of the following It is important to note that the observation data It grows dynamically during the iterative optimization process. Finally, the posterior distribution is calculated. As shown in equation (13), where Marginal likelihood represents the observed data during the Gaussian process. The joint probability of occurrence.

[0052] (13) Define the acquisition function in the parameter space Its definition is shown in equation (14), where, Indicates the current observation data The minimum cost loss is determined by selecting the sampling point that maximizes this loss as the evaluation parameter. Then, simulation is performed based on the system update matrix from step (2), and its loss function is calculated. This loss function is then used as the new data point to update the observed data. And update the posterior distribution .

[0053] (14) This iterative process of selecting sampling points using the acquisition function, calculating the cost of system simulation, and updating data and Gaussian posterior is repeated until the optimal parameters are obtained. To obtain the optimal parameter model.

[0054] In step S300 of this embodiment, the multi-objective parameter optimization algorithm combines genetic algorithm and Bayesian update, and the parameters to be optimized are: ; The cost function is ; in, and These are the weighting coefficients. and The output values ​​are for interactive model simulation. and These are actual observed values.

[0055] In step S300 of this embodiment, ; in, These are constant coefficients. For the corresponding time. Taking the lane-changing and overtaking interaction scenario as an example, an overly aggressive lane-changing trajectory will cause vehicles behind in the target lane to slow down and give way to ensure safety. The lane-changing trajectory can be adjusted to reduce this impact.

[0056] The new endpoint of the desired trajectory is generated as follows: ; in, This is the original endpoint location. To plan the time domain, The calculation method for the position change predicted based on driving intention is as follows: ; in, The attenuation coefficient is... This represents the predicted longitudinal acceleration of vehicles behind in the target lane.

[0057] The following cost function is used to select the generated target trajectory: (18) in, These are the weight coefficients for each cost item, and the corresponding cost function for each weight is the path length cost. Used to avoid excessively long paths and the cost of lateral offset. To avoid excessive lateral deviation and longitudinal speed cost Used to optimize driving speed and acceleration cost Used to limit acceleration amplitude, optimize dynamic safety and energy consumption, and mitigate safety costs. Used to avoid track collisions, at the cost of comfort. Used to improve ride comfort. The calculation methods for each cost are as follows: (19) (20) (twenty one) (twenty two) (twenty three) (twenty four) in, It is the time domain of trajectory planning. , It represents the current trajectory's longitudinal and lateral positions in the road coordinate system. , Indicates longitudinal and lateral velocities. , Indicates longitudinal and lateral acceleration, and For longitudinal and lateral acceleration, It is the center position of the target lane and the desired speed. It is the set safe distance minus the trajectory and the first The difference in minimum distance between obstacles.

[0058] A device / equipment / system for constructing a world-like model of small-sample real-vehicle data interaction in intelligent vehicle lane-changing scenarios includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0059] A computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0060] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0061] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a world-like model based on small sample real-vehicle data in intelligent vehicle lane-changing scenarios, characterized by: Includes the following steps: Acquire driving status information, which includes longitudinal position, lateral position, speed, and acceleration; An interactive control model is constructed, and based on a multi-objective parameter optimization algorithm, driving state information is used to optimize the parameters of the interactive control model to obtain the optimal parameter model. An interactive trajectory planning sub-model is constructed, which includes an intent prediction module. The intent prediction module is used to predict the driving state of vehicles behind the target lane. The position and speed of both vehicles are input, and the driving intent is output. The vehicles behind are driven by an optimal parameter model. Based on the driving intention of the interactive vehicle, trajectory sampling is adaptively performed to generate a desired trajectory that integrates the perception of driving intention.

2. The method for constructing a world-like model of small-sample real-vehicle data interaction for intelligent vehicle lane-changing scenarios according to claim 1, characterized in that: The interactive control model is based on the intelligent driver model and introduces interactive factors. .

3. The method for constructing a world-like model of small-sample real-vehicle data interaction for intelligent vehicle lane-changing scenarios according to claim 2, characterized in that: The interaction factors for ； Among them, 0 represents a scenario with no preceding vehicle, 1 represents a scenario with a fixed preceding vehicle, and 2 represents a scenario where the preceding vehicle changes. The interactive control model is ； in, Indicates the maximum acceleration. This indicates the current actual speed of the vehicle. Indicates the maximum expected speed. Indicates acceleration sensitivity, Represents the minimum expected distance. Indicates the expected headway. This indicates the speed difference between the two vehicles. This indicates the current following distance of the vehicle.

4. The method for constructing a world-like model based on small sample real-vehicle data interaction in intelligent vehicle lane-changing scenarios according to claim 1, characterized in that: The multi-objective parameter optimization algorithm combines genetic algorithm and Bayesian update. The parameters to be optimized are: ； The cost function is ； in, and These are the weighting coefficients. and The output values ​​are for interactive model simulation. and These are actual observed values.

5. The method for constructing a world-like model of small-sample real-vehicle data interaction for intelligent vehicle lane-changing scenarios according to claim 1, characterized in that: The new endpoint of the desired trajectory is ； in, This is the original endpoint location. To plan the time domain, The calculation method for the position change predicted based on driving intention is as follows: ； in, The attenuation coefficient is... This represents the predicted longitudinal acceleration of vehicles behind in the target lane.

6. The method for constructing a world-like model of small-sample real-vehicle data interaction for intelligent vehicle lane-changing scenarios according to claim 5, characterized in that: ； in, These are constant coefficients. For the corresponding time.

7. The method for constructing a world-like model of small-sample real-vehicle data interaction for intelligent vehicle lane-changing scenarios according to claim 1, characterized in that: The method for obtaining driving status information is as follows: real vehicle data of lane-changing scenarios in real road environments are obtained, and driving status information is extracted by taking the time point when the lane-changing vehicle crosses the lane line as the intermediate time.

8. A device / equipment / system for constructing a world-like model of small-sample real-vehicle data interaction in intelligent vehicle lane-changing scenarios, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

Images