Driving track planning method and device based on vehicle cloud cooperation

Abstract

The invention relates to the technical field of vehicle auxiliary driving, in particular to a driving track planning method and device based on vehicle cloud collaboration.The method comprises the steps that an initial surrounding vehicle probability model is updated based on obtained target sensing data of a vehicle and surrounding vehicles, and a surrounding vehicle probability model is obtained; the parameters and parameter uncertainty of the driver model and the vehicle lane changing model and the intention and intention uncertainty of the vehicle are obtained; constructing a self-vehicle closed-loop strategy tree based on the parameters, the parameter uncertainty, the intention of the self-vehicle and the intention uncertainty, searching the self-vehicle closed-loop strategy tree in combination with the self-vehicle closed-loop strategy tree and a parallel computing mode, and generating a behavior sequence decision of the self-vehicle; and generating a spatio-temporal trajectory of the own vehicle based on the behavior sequence decision and the trajectory optimization model. Therefore, the problems of low availability of a planning result, low prediction precision and the like caused by limitation of a view field of a single vehicle, relatively small observable range, limited calculation power of the single vehicle and difficulty in effectively solving a long-time-domain planning problem in related technologies are solved.

Description

Technical Field

[0001] The present application relates to the field of vehicle assisted driving technology, and in particular to a driving trajectory planning method and device based on vehicle-cloud collaboration. Background Art

[0002] Assisted driving technology is crucial for ensuring safe driving and improving traffic efficiency, especially in an era of low penetration of connected vehicles. Long-term assisted driving technology uses information from the road ahead and surrounding vehicles to plan lanes and speeds in advance, enabling vehicles to avoid potential large-scale emergencies and ensuring safe and efficient operation. However, some challenges remain, which can be addressed through vehicle-cloud collaboration.

[0003] Related technologies are mainly divided into two categories: predictive driving planning methods based on single-vehicle intelligence and predictive driving planning methods based on vehicle-cloud collaboration. The predictive driving planning method based on single-vehicle intelligence obtains information about surrounding vehicles and road conditions through sensors on the ego vehicle, performs predictive planning and solves the problem, and then performs further short-term trajectory planning or driving control. The predictive driving planning method based on vehicle-cloud collaboration uses cloud-based perception information to model the surrounding vehicles. After predicting the surrounding vehicles, the ego vehicle's behavior is predicted. During the planning process, the surrounding vehicles are not affected by the ego vehicle, and the planning time domain is generally less than 10 seconds.

[0004] However, in related technologies, the predictive driving planning method based on single-vehicle intelligence is restricted by the field of view of a single vehicle, the observable range is generally small, and the computing power of a single vehicle is limited. It is generally difficult to effectively solve long-term predictive planning problems, resulting in unreasonable problem design, low availability of planning results, and low prediction accuracy. The predictive driving planning method based on vehicle-cloud collaboration does not consider the interaction between the vehicle and surrounding vehicles, or implicitly considers such interaction, making long-term planning unusable. Summary of the Invention

[0005] The present application provides a driving trajectory planning method and device based on vehicle-cloud collaboration to solve the problems in related technologies, such as being restricted by the field of view of a single vehicle, having a small observable range, and having limited computing power of a single vehicle, making it difficult to effectively solve long-term planning problems, resulting in low availability of planning results and low prediction accuracy.

[0006] The first aspect of the present application provides a driving trajectory planning method based on vehicle-cloud collaboration, which includes the following steps: acquiring target perception data of the self-vehicle and surrounding vehicles, and updating an initial surrounding vehicle probability model based on the target perception data to obtain a surrounding vehicle probability model, so as to obtain parameters and parameter uncertainties in a driver model and a vehicle lane change model, and the intention and intention uncertainty of the self-vehicle based on the surrounding vehicle probability model; constructing a closed-loop strategy tree for the self-vehicle based on the parameters and parameter uncertainties, and the intention and intention uncertainty of the self-vehicle, and searching the closed-loop strategy tree for the self-vehicle in combination with the self-vehicle closed-loop strategy tree and a parallel computing method to generate a behavior sequence decision for the self-vehicle; and generating a spatiotemporal trajectory of the self-vehicle based on the behavior sequence decision and the trajectory optimization model.

[0007] Optionally, in one embodiment of the present application, the target perception data of the self-vehicle and the surrounding vehicles are obtained, and an initial surrounding vehicle probability model is updated based on the target perception data to obtain a surrounding vehicle probability model, including: segmenting and extracting road map information along the route according to the destination of the self-vehicle and the global path planning information, so as to generate discretized road map information of the self-vehicle according to different positions of the self-vehicle, for establishing the initial surrounding vehicle probability model; at the beginning of each planning cycle, the target perception data of the self-vehicle and the surrounding vehicles are obtained according to the real-time traffic twin; in each planning cycle, the discretized road map information is retrieved according to the position of the self-vehicle, and the initial surrounding vehicle probability model is updated in combination with the discretized road map information and parameters and the intention recognition algorithm to obtain a surrounding vehicle probability model, wherein the surrounding vehicle probability model includes at least one of the surrounding vehicle intention and its probability, the driver model and the vehicle lane change model.

[0008] Optionally, in one embodiment of the present application, the parameter and intention recognition algorithm is based on Bayesian inference, wherein the Bayesian inference includes: selecting a prior distribution of parameters in the driver model based on an initial hypothesis; constructing a posterior distribution of parameters in the intention, the driver model and the surrounding vehicle lane change model based on the prior distribution; when the posterior distribution cannot be analyzed, using the MCMC (Markov Chain Monte Carlo) method to sample the parameters to be identified to obtain sample data of the parameters to be identified; and obtaining the probability of the intention to be identified, parameters and their uncertainty in the parameters to be identified based on the sample data.

[0009] Optionally, in one embodiment of the present application, the constructing of the ego-vehicle closed-loop strategy tree and the searching of the ego-vehicle closed-loop strategy tree in combination with the ego-vehicle closed-loop strategy tree and the parallel computing method to generate the behavior sequence decision of the ego-vehicle include: constructing an ego-vehicle decision sequence including multiple lane-changing behaviors in a long time domain; screening out surrounding vehicles that interact with the ego-vehicle in combination with the ego-vehicle decision sequence and the confidence reachable set; obtaining the interaction information between the ego-vehicle and the surrounding vehicles by using a forward deduction simulation mechanism; performing the task allocation in the parallel computing method based on the interaction information and the ego-vehicle closed-loop strategy tree, and dynamically decomposing the tasks according to the number of available cores to obtain the behavior sequence decision.

[0010] Optionally, in one embodiment of the present application, the parameters and parameter uncertainties in the driver model and the vehicle lane changing model, and the intention and intention uncertainty of the ego vehicle include: the first-order uncertainty of the parameters in the driver model; the first-order uncertainty of the preceding vehicle changing lanes once; the uncertainty of the ego vehicle's intention entropy; and the cumulative effect of uncertainty over time.

[0011] Optionally, in one embodiment of the present application, the generating the spatiotemporal trajectory of the ego vehicle based on the behavior sequence decision and trajectory optimization model includes: selecting a high-order polynomial based on the number of constraints on the starting point and the ending point of the ego vehicle, wherein the high-order polynomial is greater than 5th order; generating the boundary constraints of the ego vehicle according to the behavior sequence decision; optimizing the residual degree of freedom coefficient of the boundary constraints using the trajectory optimization model to construct a trajectory optimization model for the rollover state index; solving the trajectory optimization model to obtain the optimal trajectory of the ego vehicle, and obtaining the spatiotemporal trajectory based on the optimal trajectory.

[0012] The second aspect of the present application provides a driving trajectory planning device based on vehicle-cloud collaboration, and the device includes: a generation module for obtaining target perception data of the self-vehicle and surrounding vehicles, and updating the initial surrounding vehicle probability model based on the target perception data to obtain the surrounding vehicle probability model, so as to obtain the parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the intention and intention uncertainty of the self-vehicle based on the surrounding vehicle probability model; a construction module for constructing a closed-loop strategy tree for the self-vehicle based on the parameters and parameter uncertainties, the intention and intention uncertainty of the self-vehicle, and searching the closed-loop strategy tree for the self-vehicle in combination with the self-vehicle closed-loop strategy tree and a parallel computing method to generate a behavior sequence decision of the self-vehicle; a planning module for generating the spatiotemporal trajectory of the self-vehicle based on the behavior sequence decision and the trajectory optimization model.

[0013] Optionally, in one embodiment of the present application, the generation module includes: a first generation unit, used to segment and extract road map information along the route according to the destination of the vehicle and the global path planning information, so as to generate discretized road map information of the vehicle according to different positions of the vehicle, for establishing the initial surrounding vehicle probability model; an acquisition unit, used to obtain target perception data of the vehicle and the surrounding vehicles according to the real-time traffic twin at the beginning of each planning cycle; an update unit, used to retrieve the discretized road map information according to the position of the vehicle in each planning cycle, and update the initial surrounding vehicle probability model in combination with the discretized road map information and parameters and the intention recognition algorithm to obtain a surrounding vehicle probability model, wherein the surrounding vehicle probability model includes at least one of the surrounding vehicle intention and its probability, the driver model and the vehicle lane change model.

[0014] Optionally, in one embodiment of the present application, the parameter and intention recognition algorithm is based on Bayesian inference, wherein the Bayesian inference includes: selecting a prior distribution of parameters in the driver model based on an initial hypothesis; constructing a posterior distribution of parameters in the intention, the driver model and the surrounding vehicle lane change model based on the prior distribution; when the posterior distribution cannot be resolved, using the MCMC method to sample the parameters to be identified to obtain sample data of the parameters to be identified; and obtaining the probability of the intention to be identified, parameters and their uncertainty in the parameters to be identified based on the sample data.

[0015] Optionally, in one embodiment of the present application, the construction module includes: a first construction unit, used to construct a self-vehicle decision sequence including multiple lane changing behaviors in a long time domain; a first screening unit, used to combine the self-vehicle decision sequence and the confidence reachable set to screen out the surrounding vehicles that interact with the self-vehicle; a second generation unit, used to obtain the interaction information between the self-vehicle and the surrounding vehicles using a forward deduction simulation mechanism; and a third generation unit, used to perform task allocation in the parallel computing mode based on the interaction information and the self-vehicle closed-loop strategy tree, and dynamically decompose the tasks according to the number of available cores to obtain the behavior sequence decision.

[0016] Optionally, in one embodiment of the present application, the parameters and parameter uncertainties in the driver model and the vehicle lane changing model, and the intention and intention uncertainty of the ego vehicle include: the first-order uncertainty of the parameters in the driver model; the first-order uncertainty of the preceding vehicle changing lanes once; the uncertainty of the ego vehicle's intention entropy; and the cumulative effect of uncertainty over time.

[0017] Optionally, in one embodiment of the present application, the planning module includes: a second screening unit, used to select a high-order polynomial based on the number of starting point and end point constraints of the ego vehicle, wherein the high-order polynomial is greater than 5th order; a fourth generation unit, used to generate the boundary constraints of the ego vehicle according to the behavior sequence decision; a second construction unit, used to optimize the residual degree of freedom coefficient of the boundary constraints using the trajectory optimization model to construct a trajectory optimization model for the rollover state index; a solving unit, used to solve the trajectory optimization model to obtain the optimal trajectory of the ego vehicle, and obtain the spatiotemporal trajectory based on the optimal trajectory.

[0018] The third aspect of the present application provides a server, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the driving trajectory planning method based on vehicle-cloud collaboration as described in the above embodiment.

[0019] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned driving trajectory planning method based on vehicle-cloud collaboration.

[0020] The fifth aspect of the present application provides a computer program product, including a computer program, which, when executed, implements the above-mentioned driving trajectory planning method based on vehicle-cloud collaboration.

[0021] The embodiment of the present application can obtain a surrounding vehicle probability model based on the acquired target perception data of the ego vehicle and surrounding vehicles, and then obtain the parameters and parameter uncertainties in the driver model and the vehicle lane change model, the ego vehicle's intention and intention uncertainty, and construct an ego vehicle closed-loop policy tree based on the parameters and parameter uncertainties, the ego vehicle's intention and intention uncertainty. The ego vehicle closed-loop policy tree is searched using a parallel computing method to obtain behavior sequence decisions, generate a spatiotemporal trajectory, and plan the ego vehicle's spatiotemporal trajectory. The uncertainty of the driver's parameters and intentions is clearly reflected through the surrounding vehicle probability model. The uncertainty propagation mechanism avoids the logical error of "prediction first, decision planning later" while considering uncertainty information, thereby improving the credibility and feasibility of the planning results. The ego vehicle closed-loop policy tree achieves the highest solution efficiency and ensures real-time performance. Good performance can be obtained under long-term predictive decision-making tasks. The trajectory optimization model is used to ensure the rollover safety of the ego vehicle, solve the lateral manipulation safety problem of the ego vehicle from the planning layer, and output a feasible, safe and stable trajectory. This solves the problems in related technologies, such as being restricted by the field of view of a single vehicle, a small observable range, and limited computing power of a single vehicle, making it difficult to effectively solve long-term planning problems, resulting in low availability of planning results and low prediction accuracy.

[0022] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0024] Figure 1 This is a flowchart of a driving trajectory planning method based on vehicle-cloud collaboration provided according to an embodiment of the present application;

[0025] Figure 2 A schematic diagram of an example of discretized road map information provided according to an embodiment of the present application;

[0026] Figure 3 A flowchart of constructing a weekly vehicle probability model according to one embodiment of the present application;

[0027] Figure 4 A flowchart for obtaining lane change probability using the MLE (Maximum Likelihood Estimation) and MOBIL (Minimize Overall Braking Induced by Lane Change) models according to one embodiment of the present application;

[0028] Figure 5 A flowchart of a predictive driving planning algorithm provided according to one embodiment of the present application;

[0029] Figure 6 A flowchart of constructing a variable time step vehicle closed-loop strategy tree according to one embodiment of the present application;

[0030] Figure 7 A schematic diagram of an example of an operating scenario provided according to an embodiment of the present application;

[0031] Figure 8 A block diagram of interactive cycle vehicle extraction results and real-time cycle vehicle deduction provided according to one embodiment of the present application;

[0032] Figure 9 A schematic block diagram of scene division based on trajectory configuration according to one embodiment of the present application;

[0033] Figure 10 A flow chart of a distribution trajectory configuration provided according to one embodiment of the present application;

[0034] Figure 11A flowchart of generating a spatiotemporal trajectory of a vehicle according to an embodiment of the present application;

[0035] Figure 12 A block diagram of a rollover prevention lane change trajectory planning result provided according to one embodiment of the present application;

[0036] Figure 13 A block diagram of a predictive, safe, and efficient driving trajectory planning result provided according to one embodiment of the present application;

[0037] Figure 14 A block diagram of a driving trajectory planning system provided according to one embodiment of the present application;

[0038] Figure 15 A block diagram of a driving trajectory planning device based on vehicle-cloud collaboration provided according to an embodiment of the present application;

[0039] Figure 16 A schematic diagram of the structure of a server provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0041] The following describes the driving trajectory planning method and device based on vehicle-cloud collaboration according to the embodiment of the present application with reference to the accompanying drawings. In view of the problem mentioned in the above background technology that it is limited by the field of view of a single vehicle, the observable range is small, the computing power of a single vehicle is limited, and it is difficult to effectively solve the planning problem in the long time domain, resulting in low availability of planning results and low prediction accuracy, the present application provides a driving trajectory planning method based on vehicle-cloud collaboration. In this method, a probability model of the surrounding vehicles can be obtained based on the target perception data of the self-vehicle and the surrounding vehicles, and then the parameters and parameter uncertainties, the intentions and intention uncertainties of the self-vehicle in the driver model and the vehicle lane change model can be obtained, and a closed-loop strategy tree of the self-vehicle can be constructed based on the parameters and parameter uncertainties, the intentions and intention uncertainties of the self-vehicle, and a parallel computing method is used. The system searches the closed-loop policy tree of the ego vehicle, obtains behavioral sequence decisions, generates a spatiotemporal trajectory, and plans the spatiotemporal trajectory of the ego vehicle. The uncertainty of the driver's parameters and intentions is clearly reflected through the weekly vehicle probability model. The uncertainty propagation mechanism avoids the logical error of "prediction first, decision planning later" while considering uncertainty information, improving the credibility and feasibility of the planning results. The closed-loop policy tree of the ego vehicle achieves the highest solution efficiency and guarantees real-time performance. It can achieve good performance in long-term predictive decision-making tasks. The trajectory optimization model ensures the safety of the ego vehicle's rollover, solves the ego vehicle's lateral manipulation safety problem at the planning level, and outputs a feasible, safe and stable trajectory. This solves the problem in related technologies that are limited by the field of view of a single vehicle, the small observable range, and the limited computing power of a single vehicle, making it difficult to effectively solve long-term planning problems, resulting in low availability of planning results and low prediction accuracy.

[0042] Specifically, Figure 1 This is a flowchart of a driving trajectory planning method based on vehicle-cloud collaboration provided according to an embodiment of the present application.

[0043] like Figure 1 As shown, the driving trajectory planning method based on vehicle-cloud collaboration includes the following steps:

[0044] In step S101, target perception data of the ego vehicle and surrounding vehicles are obtained, and the initial surrounding vehicle probability model is updated based on the target perception data to obtain the surrounding vehicle probability model, so as to obtain the parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the intention and intention uncertainty of the ego vehicle based on the surrounding vehicle probability model.

[0045] It can be understood that the target perception data of the embodiment of the present application may include but is not limited to the perception data of the roadside perception unit, the perception data of the intelligent connected vehicle, etc., which can be specifically set by technical personnel in this field according to actual conditions, and this application does not impose any specific restrictions.

[0046] Among them, in the embodiment of the present application, the target perception data of the vehicle itself and surrounding vehicles can be obtained through roadside perception units, intelligent connected vehicles, and information sharing of traffic management platforms, and the overall traffic twins in each small area can be established in parallel after partitioning the map, and then the global numbers of the vehicle itself and surrounding vehicles can be obtained with the help of vehicle ReID (Person Re-identification) technology. The global number is unique and can be retrieved through the database within a short period of time without infringing on the personal privacy of road users, and parameters and intentions can be updated for dynamic deduction.

[0047] In addition, it should be noted that the vehicle-circle probability model in the embodiment of the present application may include, but is not limited to, a driver model and a vehicle lane-changing model, which may be specifically configured by those skilled in the art based on actual conditions, and this application does not impose any specific restrictions.

[0048] As a possible implementation method, after obtaining the target perception data of the own vehicle and surrounding vehicles, the embodiment of the present application can update the initial surrounding vehicle probability model based on the target perception data to obtain the surrounding vehicle probability model, and then obtain the parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the intention and intention uncertainty of the own vehicle.

[0049] Optionally, in one embodiment of the present application, target perception data of the own vehicle and surrounding vehicles are obtained, and an initial surrounding vehicle probability model is updated based on the target perception data to obtain a surrounding vehicle probability model, including: segmenting and extracting road map information along the route according to the destination of the own vehicle and global path planning information, so as to generate discretized road map information of the own vehicle according to different positions of the own vehicle, which is used to establish an initial surrounding vehicle probability model; at the beginning of each planning cycle, target perception data of the own vehicle and surrounding vehicles are obtained according to real-time traffic twins; in each planning cycle, discretized road map information is retrieved according to the position of the own vehicle, and the initial surrounding vehicle probability model is updated in combination with the discretized road map information and parameters and the intention recognition algorithm to obtain a surrounding vehicle probability model, wherein the surrounding vehicle probability model includes at least one of the surrounding vehicle intention and its probability, the driver model and the vehicle lane change model.

[0050] Among them, in the embodiment of the present application, the discretized road map information can be understood as lane-level structured map information, which can include but is not limited to the number of drivable lanes, lane restrictions for different types of vehicles, lane width, scattered position information of lane centerlines, road level, road speed limit or lane-level speed limit, road connection relationship and traffic rules information, etc., which can be specifically set by technicians in this field according to actual conditions. For example, the discretized road map information in the embodiment of the present application can be expressed in the form of a data dictionary, specifically as follows Figure 2 shown.

[0051] As a possible implementation method, the process of constructing the weekly vehicle probability model in the embodiment of the present application is as follows: Figure 3 As shown, the following steps are included:

[0052] Step S301: Segment and extract the road map information along the route according to the vehicle's destination and global path planning information, so as to generate discretized road map information of the vehicle according to different positions of the vehicle, which is used to establish an initial vehicle-circle probability model.

[0053] Step S302: At the beginning of each planning cycle, target perception data of the vehicle and surrounding vehicles are obtained based on the real-time traffic twin.

[0054] Step S303: In each planning cycle, the discretized road map information is retrieved based on the vehicle's location, and the initial vehicle-period probability model is updated by combining the discretized road map information and parameters with the intention recognition algorithm to obtain a vehicle-period probability model.

[0055] It can be understood that the embodiment of the present application updates the initial vehicle-circling probability model, such as updating the driver parameter information of the driver model (such as the parameters and parameter variance of the driver model), the lane changing parameter information of the vehicle lane changing model (such as the parameters and parameter variance of the vehicle lane changing model) and the vehicle intention information of the ego vehicle (the intention and intention variance of the ego vehicle), and then obtains the vehicle-circling probability model, wherein the vehicle-circling probability model may include but is not limited to at least one of the vehicle intention and its probability, the driver model and the vehicle lane changing model.

[0056] The expression of the driver model in the embodiment of the present application can be, but is not limited to,:

[0057]

[0058] Among them, T hw represents the expected headway, a max Indicates the maximum acceleration, b comf Indicates comfortable deceleration, s min Indicates the minimum distance, v des,ego Indicates the expected speed.

[0059] In addition, it should be noted that the embodiment of the present application uses softplus(*) instead of ReLU(*) to make the driver model continuous and differentiable everywhere, and then the analytical gradient of each parameter can be obtained, where the parameter θ to be estimated is T hw 、a max 、b comf 、s min 、v des,ego The expression of the softplus(*) function can be, but is not limited to, the following:

[0060] softplus(x)=log(1+e x )

[0061] , (3)

[0062] Optionally, in one embodiment of the present application, the parameter and intention recognition algorithm is based on Bayesian inference, wherein the Bayesian inference includes: selecting the prior distribution of parameters in the driver model based on the initial hypothesis; constructing the posterior distribution of parameters in the intention, driver model and surrounding vehicle lane change model based on the prior distribution; when the posterior distribution cannot be resolved, using the MCMC method to sample the parameters to be identified to obtain sample data of the parameters to be identified; and obtaining the probability of the intention to be identified, parameters and their uncertainty in the parameters to be identified based on the sample data.

[0063] In some embodiments, the parameter and intention recognition algorithm of the driver model in the embodiment of the present application can be, but is not limited to, based on Bayesian inference, wherein, in the embodiment of the present application, the steps based on Bayesian inference include: first, selecting the prior distribution of the parameters in the driver model based on the initial hypothesis; second, constructing a likelihood function that reflects the observed data under given parameters; then, combining the prior distribution and the likelihood function through Bayes' theorem to obtain the posterior distribution of the parameters; then, when the posterior distribution cannot be resolved, use the MCMC method to sample it to obtain sample data of the parameters to be identified; finally, through the sample data, calculate the variance of the model parameters to quantify the uncertainty, and further determine the credible interval to characterize the potential range of variation of the parameters.

[0064] The expression of the prior distribution p(θ) in the embodiment of the present application can be, but is not limited to,:

[0065]

[0066] Among them, μ θ is the mean of the prior distribution of the parameter, is the prior variance of the parameter.

[0067] The expression of the constructed likelihood function p(data|θ) can be, but is not limited to:

[0068]

[0069] Among them, x i is the i-th observation value, σ 2 is the variance of the observation noise, and N is the number of observations.

[0070] Furthermore, according to Bayes' theorem, the posterior distribution p(θ|data) in the embodiment of the present application is obtained by multiplying the prior distribution and the likelihood function, and its expression can be, but is not limited to,:

[0071]

[0072] Among them, p(data) is a normalization constant, which is ignored in actual processing and does not depend on the specific model parameters θ.

[0073] In addition, when the posterior distribution cannot be analyzed, the embodiment of the present application uses the MCMC method to sample it, wherein the MCMC method of the embodiment of the present application constructs a Markov chain and samples a series of θ from the posterior distribution to construct the probability distribution of the parameters. The θ1, θ2, ..., θ obtained by sampling M , and then estimate the distribution of the parameters. For example, if the embodiment of the present application samples T hw Sample value of The sample mean can be used as the estimated value of the parameter, the sample variance can be used as the uncertainty of the parameter, and the sample covariance matrix can be used to represent the linear dependence between the parameters. That is, the embodiment of the present application calculates the covariance between each pair of samples sampled for each parameter, and then obtains the sample covariance matrix. Among them, the element C of the covariance matrix is ij The calculation formula can be but not limited to:

[0074]

[0075] Where M is the number of samples in Markov Chain Monte Carlo, is the i-th parameter value of the m-th sample, is the sample mean of the i-th parameter.

[0076] In addition, it should be noted that the diagonal elements of the covariance matrix in the embodiment of the present application are the variances of the parameters, and the diagonal elements represent the correlation between different parameters.

[0077] In some embodiments, the parameter and intention recognition algorithm for the driver model of the present application can be, but is not limited to, a heuristic algorithm. The variance of the parameters is determined by simulating the driver model after the parameters have been identified. The preceding vehicle's speed and relative distance information are input at a fixed time, initialized with real historical speed data, and the simulated vehicle's speed is obtained. The squared difference between the simulated speed data and the real speed data is then calculated frame by frame. The average of the squared differences is calculated and multiplied by a defined conversion factor to determine the overall parameter variance. The fixed time can be set by those skilled in the art based on actual circumstances and is not specifically limited in this application.

[0078] In some embodiments, the parameters and intention recognition algorithm of the driver model of the embodiment of the present application can be, but is not limited to, a combination of rule-based intention possibility recognition and Bayesian inference. The intention bifurcation point of the vehicle within a set distance (which can be set by technicians in this field according to actual conditions, and is not specifically limited in this application) is determined through map information and traffic rules, and different intentions are set based on the identified driver model (each intention has its own prior information, which can be inferred through road structure and vehicle position). The historical data of the weekly vehicle in the window within the past set time (which can be set by technicians in this field according to actual conditions, and is not specifically limited in this application) is used as input for simulation to obtain the acceleration output of the vehicle under different intention assumptions. Bayesian inference is still used for parameter update, and the likelihood function of each different intention is calculated under the given identified driver model parameters and observation data conditions. After inputting the evaluation function, the probability increment of the different intentions is obtained, which is added to the intention probability of the previous control cycle to update the new intention information.

[0079] Furthermore, the vehicle lane-changing model in the embodiment of the present application can use the MLE method or Bayesian inference to perform probability calculation and parameter identification and update.

[0080] For example, the embodiment of the present application is based on the MLE method and the MOBIL model, and the steps for obtaining the probability of the vehicle changing left lane L, lane keeping K, and right lane changing R are as follows: Figure 4 As shown, the main steps include:

[0081] Step S401: Obtain lane-changing acceleration.

[0082] It can be understood that the embodiment of the present application can use the MOBIL model concept to obtain the lane change acceleration vector a, which can include but is not limited to Δa L,ego 、a K,ego =0, Δa R,ego , Δa L,f , Δa K,f =0, Δa R,f The expected acceleration benefits of the ego vehicle and the following vehicle after the lane change corresponding to the ego vehicle's LKR action (calculated using the identified driver models of the ego vehicle and surrounding vehicles) are: for the ego vehicle, it is the difference between the acceleration after the lane change and the acceleration before the lane change; for the following vehicle, it is the sum of the loss of the new following vehicle after the lane change due to the ego vehicle's lane change and the benefit of the original following vehicle before the lane change due to the ego vehicle's lane change.

[0083] Step S402: Design an evaluation function.

[0084] The evaluation function f(a,p) in the embodiment of the present application can reflect the preferences of the ego vehicle and surrounding vehicles for different lane change options. Generally, the greater the acceleration, the more inclined the ego vehicle and surrounding vehicles are to choose this behavior. It can be expressed as, but not limited to:

[0085]

[0086] Among them, α L ,β L ,α R ,β R ,p A total of 5 parameters are model parameters to be identified. The common parameter p is the politeness coefficient of the corresponding MOBIL model, which reflects the degree to which this car attaches importance to the impact of its own car on other cars.

[0087] Step S403: Obtain lane-changing probability.

[0088] Furthermore, in the embodiment of the present application, the output of the evaluation function can be normalized using a softmax(*) function to obtain the probabilities of left lane change L, lane keeping K, and right lane change R at this moment, which can be expressed as, but not limited to:

[0089]

[0090] For example, the embodiment of the present application uses MLE to perform parameter identification and update. The main content is: the embodiment of the present application can use historical data sets to construct data sets, which can include but are not limited to Among them, d i ∈{L,K,R} represents the actual lane-changing decision of the i-th sample. The main goal is to find the parameter α L ,β L ,a R ,β R ,p makes the probability predicted by the model best match the actual observed decisions.

[0091] It is understood that in the embodiment of the present application, the goal of MLE is to maximize the product of the decision probabilities predicted by the model under all data samples, and its log-likelihood function can be expressed as, but not limited to:

[0092]

[0093] Among them, the embodiment of the present application can use the first-order gradient descent method, or the second-order quasi-Newton method to solve and update the parameters.

[0094] Furthermore, in the embodiment of the present application, the parameters of the vehicle lane-changing model and the intention recognition algorithm can be, but are not limited to, a combination of rule-based intention possibility recognition and Bayesian inference, and the intention bifurcation point of the vehicle within a given distance in the future (not specifically limited in the present application) is determined through map information and traffic rules (M possible intentions are identified, for example, if the vehicle is currently driving on a highway and there is a ramp 500m ahead, then the possible intentions include going straight and getting off the ramp; for example, if the vehicle is currently at a traffic light intersection, the possible intentions include going straight, turning left, turning right, and turning around), and based on the identified driver model, the vehicle is set to different intentions (each intention has its own prior information, which can be determined by the road structure, vehicle position, etc. Reasonable inference is given based on the position. For example, at the ramp entrance, P(straight) = 0.7, P(down ramp) = 0.3) can be given based on the usual traffic flow + the current lane position of the vehicle + the current acceleration trend of the vehicle. The historical data of weekly vehicles in the window within a given time in the past (this application does not impose specific restrictions) is used as input for simulation to obtain the acceleration output of the vehicle under different intention assumptions. The Markov chain Monte Carlo under the Bayesian inference just given is still used for parameter update. The likelihood function of each different intention is calculated under the given identified driver model parameters and observation data conditions. After inputting the evaluation function, the probability increment of different intentions is obtained, which is added to the intention probability of the previous control cycle to update the new intention information.

[0095] The embodiment of the present application relies on real-time traffic twins to establish a weekly vehicle probability model, fully considering the uncertainty of driver parameters and intentions, accurately simulating the driving environment, and providing the vehicle with more realistic and accurate environmental information, thereby further improving the safety and accuracy of the vehicle's decision-making.

[0096] Optionally, in one embodiment of the present application, the parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the vehicle's intention and intention uncertainty include: the first-order uncertainty of the parameters in the driver model; the first-order uncertainty of the preceding vehicle changing lanes once; the uncertainty of the vehicle's intention entropy; and the cumulative effect of uncertainty over time.

[0097] It will be understood by those skilled in the art that the parameters and parameter uncertainties, the intentions of the vehicle and the uncertainty of intentions in the driver model and the vehicle lane change model in the embodiment of the present application may include, but are not limited to: the first-order uncertainty of the parameters in the driver model, the first-order uncertainty of the preceding vehicle changing lanes once, the uncertainty of the entropy of the vehicle's intentions, the cumulative effect of uncertainty over time, etc., which may be specifically set by those skilled in the art according to actual conditions, and this application does not impose any specific restrictions.

[0098] In step S102, a closed-loop strategy tree for the ego vehicle is constructed based on parameters and parameter uncertainty, and the ego vehicle's intention and intention uncertainty. The closed-loop strategy tree for the ego vehicle is searched by combining the ego vehicle's closed-loop strategy tree with a parallel computing method to generate a behavior sequence decision for the ego vehicle.

[0099] In some embodiments, the embodiments of the present application use a forward deduction method to consider the interaction between the ego vehicle and the surrounding vehicles, and use the uncertain surrounding vehicle information to assist in trajectory decision-making through an uncertainty accumulation mechanism. While considering the uncertainty information, it avoids the logical error of "predicting first, then decision-making and planning", improves the credibility and feasibility of the planning results, and can be used to make long-term credible ego vehicle decisions and consider the potential dangers brought by the uncertainty of the surrounding vehicles.

[0100] It can be understood that the embodiment of the present application provides an uncertainty propagation mechanism, as well as an interaction method between the vehicle and the surrounding vehicles based on the uncertainty propagation mechanism. In the embodiment of the present application, the interaction methods can be divided into two types. The first is a multi-scenario Monte Carlo simulation method based on sampling, referred to as Monte Carlo simulation; the second is a single-scenario simulation method that considers uncertainty propagation, referred to as single-scenario simulation.

[0101] In addition, it should be noted that, in the embodiment of the present application, the simulation of all scenarios is performed based on the action sequence conditions selected by the ego vehicle, taking into account the interaction between the ego vehicle and the surrounding vehicles.

[0102] In some embodiments, Monte Carlo simulation does not consider the accumulation of uncertainty. Random sampling is used to sample different parameter values ​​of the surrounding vehicles and scenarios with different specific intentions based on the estimation of the mean and variance of the model parameters of the surrounding vehicles. Parameter values ​​and intentions with higher probabilities are more likely to be sampled. Each scenario contains a group of surrounding vehicles with specific parameter values ​​and intentions. By performing forward simulation and deduction on each surrounding vehicle, each specific parameterized scenario instance with the same probability obtains a deterministic surrounding vehicle prediction result. The surrounding vehicle position prediction results of all scenarios are weighted according to the same probability to obtain the probability position distribution of the surrounding vehicle under the selected ego-vehicle behavior sequence.

[0103] In some embodiments, the results of a single-scenario simulation are based on the parameters and intentions identified with the highest probability, but the accumulation and propagation of uncertainty are calculated during each forward simulation step, and deterministic events (such as a static obstacle on the road, which has a 100% probability of existing here) will cause the uncertainty to gradually converge.

[0104] In some embodiments, the uncertainty propagation mechanism is an uncertainty propagation mechanism after linearizing vehicle motion, and the propagation of uncertainty between vehicles is considered by Taylor expanding the impact of the driver model and lane change model parameters on the model output results. The specific implementation method is as follows:

[0105] The driver model in the embodiment of the present application can be, but is not limited to, represented as:

[0106] a pred =f(v p ,v ego ,d intvl ,θ)

[0107] , (11)

[0108] When there is no preceding vehicle, it can be simplified to free flow driving, which can be expressed as, but not limited to:

[0109] a pred =f(θ)

[0110] , (12)

[0111] The parameters of the driver model are identified with variance. Through Taylor expansion, the influence of the uncertainty of the leading vehicle state and the uncertainty of the driver parameter θ of the following vehicle on the speed can be approximated as:

[0112]

[0113] The uncertainty of the spacing variance is the sum of the uncertainty of the position states of the two vehicles, which can be expressed as, but not limited to:

[0114]

[0115] Because the preceding vehicle may change lanes and leave (1-P(K)) or may stay in the lane (P(K)), the calculation formula for the longitudinal uncertainty including the lane change uncertainty can be expressed as, but not limited to:

[0116]

[0117] The subscript pp represents the preceding vehicle of the preceding vehicle.

[0118] When considering intention uncertainty, since we can assume with good reason that the intentions of different vehicles are independent of each other, the calculation of uncertainty increases the product of intention entropy. Let:

[0119]

[0120] Then the following formula holds:

[0121]

[0122]

[0123] The calculation formula of the intention entropy H of a certain car is shown in formula (18). The intention can include but is not limited to the N possible intentions of this car I = {I1, I2, ..., I N}.

[0124]

[0125] Furthermore, the embodiment of the present application can assume that the intentions of different vehicles are independent of each other because the macro intention depends on the path planning of the vehicle to the destination, and basically no mutual influence between vehicles will be generated.

[0126] In addition, in the embodiment of the present application, in the absence of state updates, the propagation value of uncertainty depends on the prediction step, which is achieved through the following steps: within each time step Δt, the state of the vehicle is predicted by the acceleration model, and the state vector x includes position and velocity. The state transition can be expressed as, but not limited to:

[0127]

[0128] Right now

[0129] x=Ax+Ba pred

[0130] , (20)

[0131] Perform ZOH discretization to obtain the discrete state equation:

[0132] x k+1 =Gx k +Ha pred,k

[0133] ,(twenty one)

[0134] At each time step, the uncertainty of the state will propagate. This uncertainty affects the uncertainty of position and velocity through the state transfer equation. Therefore, the uncertainty propagation of the leading vehicle can be expressed as, but not limited to:

[0135] P k+1 =GP k G+Q

[0136] ,(twenty two)

[0137] Where Q is the process noise covariance matrix, which can be expressed as, but not limited to:

[0138]

[0139] in, Represents the effect of acceleration on position variance, because position is the double integral of acceleration and the variance is proportional to the fourth power of sampling time; represents the effect of acceleration on the unknown and velocity covariance, because velocity is the first integral of acceleration and position is the second integral of acceleration, resulting in a cubic time dependence; Δt 2 This indicates that velocity depends on the square of acceleration.

[0140] The embodiment of the present application uses a forward deduction method to comprehensively consider the interaction between the vehicle and surrounding vehicles, and introduces an uncertainty accumulation mechanism to fully utilize the uncertainty information of surrounding vehicles to assist in trajectory decision-making, avoid the traditional logical error of "predict first, then decision-making and planning", ensure the correctness of the planning results, and improve the credibility of decision-making and planning.

[0141] Optionally, in one embodiment of the present application, a closed-loop strategy tree for the ego vehicle is constructed, and the closed-loop strategy tree for the ego vehicle is searched in combination with a parallel computing method to generate a behavior sequence decision for the ego vehicle, including: constructing a decision sequence for the ego vehicle including multiple lane-changing behaviors in a long time domain; combining the decision sequence for the ego vehicle and a confidence reachable set to screen out surrounding vehicles that interact with the ego vehicle; utilizing a forward deduction simulation mechanism to obtain interaction information between the ego vehicle and the surrounding vehicles; performing task allocation in a parallel computing method based on the interaction information and the closed-loop strategy tree for the ego vehicle, and dynamically decomposing tasks according to the number of available cores to obtain a behavior sequence decision.

[0142] In some embodiments, the embodiments of the present application construct a closed-loop strategy tree for the ego vehicle based on parameters and parameter uncertainties, the ego vehicle's intentions and intention uncertainties, consider the ego vehicle decision sequence including multiple lane changing behaviors in the long time domain, use the confidence reachable set to screen the interacting surrounding vehicles, design a forward deduction simulation mechanism to focus on the interaction between the ego vehicle and related surrounding vehicles, and design a load-adaptive ego vehicle behavior tree search algorithm based on parallel computing. The task is dynamically decomposed according to the number of available cores to achieve the highest solution efficiency and ensure real-time performance. Then, through heuristic pruning and other methods, the solution efficiency is improved while ensuring safe and efficient driving, and good performance can be achieved in predictive decision-making tasks in the long time domain.

[0143] For example, the embodiment of the present application can generate behavior sequence decisions by constructing a closed-loop strategy tree for the self-vehicle, interactive vehicle screening, self-vehicle behavior tree pruning, parallel computing and task allocation, etc. The process is as follows: Figure 5 As shown, the main steps can be:

[0144] Step S501: Utilize a single-scenario simulation of the interaction model between the ego vehicle and surrounding vehicles to obtain a cost value as the upper limit of the cost.

[0145] Step S502: Select a weekly vehicle that meets certain interaction conditions.

[0146] Among them, certain interaction conditions can be set by technicians in this field according to actual conditions, and this application does not impose any specific restrictions.

[0147] It can be understood that the embodiment of the present application uses the set of surrounding vehicles that interact with the ego vehicle to be screened by the confidence reachable set, and in the forward simulation process, the surrounding vehicles that do not interact with the ego vehicle completely use the data obtained by simulation in step S501. (For example, in the embodiment of the present application, the trajectories generated by the same lane changing operation are regarded as a trajectory configuration set, and L and R are used to represent the left lane changing and right lane changing operations, respectively. Then, in the prediction time domain L (which can be set by those skilled in the art according to actual conditions and is not specifically limited in this application), there is a trajectory configuration representing the set of all possible trajectories that complete the left lane change once and only once in the prediction time domain; RLR represents the set of all possible trajectories that complete the right, left, and right lane changes once in sequence in the pre-test domain, such as [K, R, R, R, R, K, K, K, L, L, L, K, R, R, R, R, K, K] in 20 seconds, indicating that the lane change behavior to the right, left, and right was performed for a duration of 4 seconds at the 2nd, 10th, and 15th seconds, respectively).

[0148] In addition, it should be noted that the embodiment of the present application can use a confidence reachable set to screen the interaction method of the surrounding vehicles that interact with the own vehicle, that is, according to a certain percentile of the comfortable acceleration in the driver parameters of the surrounding vehicles as the maximum acceleration, and a certain percentile of the comfortable deceleration as the maximum deceleration, the speed of the vehicle at the start time of the prediction is used as the initial speed, and the nearest and farthest distances reached by the vehicle within the confidence range are calculated. The distance in between is its confidence reachable set, and the vehicles whose confidence reachable set at the final moment of the prediction time domain intersects with the confidence reachable set of the own vehicle are screened as interacting vehicles.

[0149] For example, in this embodiment of the application, the 85th percentile of the comfortable acceleration in the surrounding vehicle driver parameters can be used as the maximum acceleration, and the 85th percentile of the comfortable deceleration as the maximum deceleration. The vehicle's speed at the start of the prediction is used as the initial speed. The closest and farthest distances reached by the vehicle within the confidence range are calculated, and the distances in between are its confidence reachable set. Vehicles whose confidence reachable set at the final moment of the prediction time domain intersects with the confidence reachable set of the ego vehicle are selected as interacting vehicles.

[0150] Step S503: Build a closed-loop strategy tree for the ego-vehicle and obtain a trajectory configuration set.

[0151] Among them, in the embodiment of the present application, a closed-loop strategy tree of the vehicle can be constructed based on domain knowledge, and a feasible trajectory configuration set can be constructed according to the current state and traffic rules.

[0152] In addition, the closed-loop strategy tree of the self-driving vehicle in the embodiment of the present application can be constructed in a variable time step manner, specifically as follows Figure 6 As shown, set a time limit t N , when time t <tN When the discrete step length is Δt1; when time t>t N When , the discrete step length is taken as Δt2. Among them, Δt1<Δt2, which reflects that the uncertainty of the prediction results increases with time. When the prediction time domain is very long, a multi-stage time step can be adopted, and a larger time discrete step length can be used in the longer time domain to save computing resources. The principle of step size selection ensures that the sum of all time steps can converge to a certain constant, that is, when t→∞, Δt i →∞, and

[0153] Step S504: perform task allocation in a parallel computing manner and dynamically decompose tasks according to the number of available cores to obtain a behavior sequence decision.

[0154] It can be understood that the embodiment of the present application performs task allocation in a parallel computing manner, and performs parallel search and pruning at the same time. The parallel search initializes M cores and performs parallel forward deduction based on different trajectory configurations. Each core is responsible for searching a subset of the trajectory configuration.

[0155] In the embodiment of the present application, pruning of the vehicle behavior tree can be performed using both heuristic and rule-based methods. The heuristic method may include, but is not limited to, training a heuristic function using a learning-based approach to estimate the cost of a particular situation; the rule-based method may include, but is not limited to, setting a rule such as automatically pruning a scenario if the incremental cost of a single scenario simulation exceeds a certain value.

[0156] For example, in the embodiment of the present application, the pruning of the vehicle behavior tree is based on a rule-based approach. The rule is set as follows: if the cumulative cost of a scene exceeds the recorded cost upper limit, the scene will be automatically pruned. For example, if the cost increment of a single scene simulation exceeds a value of 100 (related to the weight setting), the scene will be automatically pruned.

[0157] The task allocation in the parallel computing mode in the embodiment of the present application is the computing task allocation in the tree search algorithm based on the trajectory configuration parallelism. The cloud parallel core can adopt a means of parallel computing such as CPU (Central Processing Unit) or GPU (Graphics Processing Unit). Based on the feasibility, N different trajectory configurations are preliminarily screened. When considering the maximum number of lane change decisions, there is an upper limit N for the trajectory configuration. max , the number of parallel cores M≥N maxBased on the discrete rules of the ego vehicle's closed-loop policy tree, the number of scenarios for each trajectory configuration can be calculated in advance. The following principles are employed: Principle 1: Minimize the maximum number of scenarios across all cores; Principle 2: Distribute all cores as evenly as possible. Parallel core trajectory allocation is performed using a heuristic greedy algorithm. Each core is assigned a subset of trajectory configurations containing several specific decision-making scenarios. Using dynamic programming, the common portion of the ego vehicle's behavior tree within each core can be calculated only once. The increased computational effort resulting from the different specific behavioral decisions within each trajectory configuration is reflected in the increased heterogeneity of scenarios compared to a scenario without these decisions.

[0158] For example, the task allocation in the parallel computing mode in the embodiment of the present application is the computing task allocation in the tree search algorithm based on the trajectory configuration parallelism, and the cloud parallel core adopts the 24-core Intel@i9 CPU for parallel computing. Specifically, Figure 7 The figure shows a scenario at the beginning of a specific planning cycle. The vehicle has just passed the intersection and there are 11 surrounding vehicles within a radius of 200 meters. Using the confidence reachable set method, the number of interactive surrounding vehicles is 7 surrounding vehicles, as shown in the figure. Figure 8 As shown in the figure, N different feasible trajectory configurations are preliminarily screened out according to feasibility. In the embodiment of the present application, N = 6 (trimmedSet = {R, RL, RR, RRR, RLR, RRL}). When the maximum number of lane change decisions is 3, there is an upper limit N on the trajectory configuration. max =14, number of parallel cores M = 24 ≥ N max =14, such as Figure 9 As shown. Based on the discrete rules of the ego vehicle closed-loop policy tree, the number of scenarios for each trajectory configuration is calculated in advance, and the following principles are used:

[0159] Principle 1: Minimize the maximum number of scenes among all cores.

[0160] Principle 2: Distribute all cores as evenly as possible and perform parallel core trajectory allocation.

[0161] The embodiment of the present application adopts a heuristic greedy algorithm to assign a subset of trajectory configurations to each core, which contains several specific possible decision-making behaviors. The goal of the algorithm is to reasonably divide the lane change scenarios (such as "RLR" represents lane changes to the right, left, and right in sequence) and evenly distribute them to the parallel computing cores to achieve load balancing. The specific process is as follows: Figure 10 As shown, the main steps include:

[0162] Step S1001: Input scene initialization.

[0163] Among them, the embodiment of the present application initializes the number of lane changes and the corresponding time period (such as the moment of each lane change) for each scene, generates a scene set, and the length of the scene set is discretized based on a fixed time step, and the length of the time series corresponding to each scene is recorded.

[0164] Step S1002: scene segmentation preparation.

[0165] Among them, the embodiment of the present application converts the number of lane changes corresponding to the scene into the length of the "plank", where the length represents the computational workload required for each scene, and records the start and end time points (index) of each scene to facilitate subsequent segmentation.

[0166] Step S1003: Calculate the required number of splits.

[0167] Among them, the embodiment of the present application determines the target core number M, calculates the difference between the current number of wooden boards and the target core number, and determines the number of times the wood boards need to be cut.

[0168] Step S1004: Greedy algorithm segmentation.

[0169] Among them, the embodiment of the present application splits the longest wooden board (i.e., the scenario with the largest computational workload): calculates its possible cutting points (accumulated workload); selects a cutting point closest to the average workload to ensure load balancing; divides the scene into two parts, and adds the newly cut part to the wooden board collection.

[0170] Step S1005: Update scene information.

[0171] Among them, the embodiment of the present application updates the start and end indexes of the segmented scenes to ensure that the scenes allocated to each core include the workload after segmentation.

[0172] Step S1006: Allocate cores.

[0173] Among them, the embodiment of the present application allocates each sub-scenario to a parallel core according to the segmentation result of the scene, and outputs the allocation result, which may include but is not limited to the scene information and corresponding workload of each core.

[0174] Step S1007: Output the balancing result.

[0175] Among them, the embodiment of the present application outputs the scene name and workload finally processed by each core to ensure that the load of each core is balanced and the total computing power is optimal.

[0176] For example, the scene name assigned according to the above process in the embodiment of the present application, such as RRR_a00b01, represents the RRR scene in which the first lane change is completed within the time interval [0,1]. For the 6 trajectory configurations of trimmedSet given in this example, a subset of 20 trajectory configurations is refined according to the number of parallel cores 20: balancedSet = 20×1 array {'R_a00b17'}, {'RL_a00b02'}, {'RR_a00b02'}, {'RRR_a00b01'}, {'RLR_a00b01'}, {'RRL_a00b01'}, {'RRR_a01b02'}, {'RLR_a00b02'} '}, {'RRL_a01b02'}, {'RRR_a02b03'}, {'RLR_a02b03'}, {'RRL_a02b03'}, {'RL_a02b05'}, {'RR_a02b05'}, {'RRR_a03b06'}, {'RLR_a03b17'}, {'RRL_a03b17'}, {'RL_a05b17'}, {'RR_a05b17'}, {'RRR_a06b17'}.

[0177] Furthermore, in the embodiments of the present application, through the concept of dynamic programming, the common part of the closed-loop strategy tree of the ego-vehicle in each core can be calculated only once, and the increase in computation brought about by the different specific behavioral decisions in each trajectory configuration is reflected in the increased heterogeneity of scenarios compared to the case without including secondary decisions.

[0178] The embodiment of the present application constructs a closed-loop self-vehicle strategy tree and uses a confidence reachable set to screen interactive vehicles. A load-adaptive self-vehicle closed-loop strategy tree search algorithm for parallel computing is designed. The algorithm can dynamically decompose tasks according to the number of available cores. Heuristic pruning further improves the solution efficiency, ensuring the real-time and high efficiency of the system in predictive decision-making tasks involving multiple lane changes over a long time domain, thereby further improving the efficiency and real-time performance of self-vehicle decision-making.

[0179] In step S103 , the spatiotemporal trajectory of the ego vehicle is generated based on the behavior sequence decision and trajectory optimization model.

[0180] In some embodiments, the embodiments of the present application derive a trajectory optimization model that minimizes the lateral load transfer rate based on a simplified vehicle roll model, which is used to output a predictive driving space-time trajectory based on the results of smooth behavioral decisions, ensure the rollover safety of the vehicle, solve the lateral manipulation safety problem of the vehicle from the planning level, and output a feasible, safe and smooth space-time trajectory.

[0181] As a possible implementation method, the embodiment of the present application can use the behavior sequence decision and trajectory optimization model to generate the spatiotemporal trajectory of the vehicle.

[0182] Optionally, in one embodiment of the present application, a spatiotemporal trajectory of the ego vehicle is generated based on a behavior sequence decision and a trajectory optimization model, including: selecting a high-order polynomial based on the number of constraints on the starting point and the ending point of the ego vehicle, wherein the high-order polynomial is greater than 5th order; generating boundary constraints of the ego vehicle according to the behavior sequence decision; optimizing the residual degree of freedom coefficient of the boundary constraints using the trajectory optimization model to construct a trajectory optimization model for the rollover state index; solving the trajectory optimization model to obtain the optimal trajectory of the ego vehicle, and obtaining the spatiotemporal trajectory based on the optimal trajectory.

[0183] In some embodiments, the process of generating the spatiotemporal trajectory of the vehicle in the embodiment of the present application is as follows: Figure 11 As shown, the main contents are:

[0184] Step S1101: Select a high-order polynomial (greater than 5th order).

[0185] For example, the high-order polynomial selected in the embodiment of the present application is a 6th-order polynomial, which can be expressed as, but not limited to:

[0186]

[0187] Among them, a0, a1, ..., a6 and b0, b1, ..., b6 represent the unknown coefficients, which are 14 in total.

[0188] Step S1102: Generate boundary constraints of the vehicle based on the behavior sequence decision.

[0189] For example, the embodiment of the present application can generate boundary constraints for each lane change process curve based on the calculated decision, such as the starting point constraint and the ending point constraint, etc., which are not specifically limited in the present application. In particular, the embodiment of the present application assumes that all lane changes last for 4 seconds, that is, t f =4 (assuming the lane is changed from t0 seconds, the time axis is shifted to t0 seconds to facilitate problem construction).

[0190] In addition, the embodiment of the present application can determine the lateral positions of the starting and ending points based on the original lane and the target lane of the lane change, and can determine the longitudinal positions of the starting and ending points based on the speed information, and constrain the lateral speed and acceleration of the starting and ending points to be 0 to meet the trajectory feasibility and driving smoothness.

[0191] The starting point constraint can be expressed as, but not limited to:

[0192]

[0193] End point constraints can be expressed as, but are not limited to:

[0194]

[0195] Step S1103: constructing a trajectory optimization model that minimizes LTR (Lateral-load Transfer Ratio) based on the remaining degrees of freedom;

[0196] Step S1104: Solve the trajectory optimization model to obtain a space-time trajectory that meets certain load transfer rate conditions.

[0197] Furthermore, the embodiment of the present application brings the 6th order polynomial into the starting point constraint, and the coefficients a0, a1, a2 and b0, b1, b2 can be directly solved, which can be: a0 = x0, b0 = y0, a1 = v 0x ,b1=v 0y ,

[0198] Furthermore, in the embodiment of the present application, the coefficients a3, a4, a5 and b3, b4, b5 are coupled with the remaining coefficients a6, b6. By introducing the termination point constraint, the coefficients a3, a4, a5 and b3, b4, b5 can be expressed by the remaining coefficients a6, b6, which can be expressed as, but not limited to:

[0199]

[0200] in,

[0201]

[0202] make

[0203]

[0204] [b5 b4 b5] T =A -1 cA -1 db6

[0205] , (31)

[0206] Therefore, only the remaining coefficients a6 and b6 in x(t) and y(t) are to be determined. Once they are determined, a specific curve can be obtained immediately, and the lateral acceleration a of the trajectory can be calculated after the trajectory is obtained. y That is, the embodiment of the present application can obtain the lateral acceleration according to the speed curve and the curvature curve.

[0207] The expressions of the velocity curve and the curvature curve can be, but are not limited to, expressed as:

[0208]

[0209] The expression of lateral acceleration can be, but is not limited to, expressed as:

[0210]

[0211] Furthermore, the embodiment of the present application simplifies the LTR calculation method by approximating the LTR as the second-order inertia link of the lateral acceleration of the space-time trajectory, which can be expressed as, but not limited to:

[0212]

[0213] Furthermore, the embodiment of the present application identifies the coefficients ζ and ω of the simplified LTR model by recursive least squares method. n , and the other coefficients generated for the given remaining coefficients a6,b6 can be directly obtained as a y (t), according to the input a y (t) Solving equation (31) yields LTR(t).

[0214] It should be noted that the embodiment of the present application adds vehicle steering actuator constraints:

[0215] The calculation formula of the trajectory front wheel angle can be expressed as but not limited to:

[0216] δ(t)=atan(Lk(T))

[0217] , (36)

[0218] The front wheel angle constraint and the rotation angular velocity constraint can be expressed as, but not limited to:

[0219]

[0220] Furthermore, the trajectory optimization model constructed in the embodiment of the present application can be expressed as, but not limited to:

[0221]

[0222] st(35) system dynamics

[0223] (31)(34) Calculation of lateral acceleration

[0224] (37) Actuator constraints

[0225] , (38)

[0226] Among them, the embodiment of the present application can use various numerical methods to quickly solve the problem, and obtain the optimal residual coefficient after solving. It is used to generate a smooth and executable trajectory that minimizes LTR. In this embodiment of the application, for the underdamped LTR case, the trajectory solved is as follows Figure 12As shown in the figure, the optimal LTR lane-changing trajectory tends to be an asymmetric trajectory that is slow at first and then intense. Under different speed conditions, the maximum LTR is reduced by about 10% to 20% compared with the symmetrical trajectory of the fifth-order polynomial. The smoothing of all lane-changing trajectories is specifically as follows: Figure 13 As shown, it is the planned smooth space-time trajectory.

[0227] The embodiment of the present application can generate a smooth space-time trajectory based on the trajectory optimization model, thereby ensuring the stability and safety of the vehicle during lateral operation, effectively preventing vehicle rollover, improving lateral control safety at the planning level, and enhancing vehicle control safety.

[0228] The following combination Figure 14 As shown, the driving trajectory planning method based on vehicle-cloud collaboration proposed in the embodiment of this application is introduced in detail using a specific embodiment.

[0229] in, Figure 14 A block diagram of a driving trajectory planning system provided according to one embodiment of the present application.

[0230] exist Figure 14 In the embodiment of the present application, the driving trajectory planning system is composed of traffic (the vehicle itself, surrounding vehicles, traffic participants, etc.), road infrastructure and roadside intelligent devices, a cloud control platform, a communication network, and related supporting platform information (map platform, traffic management platform, etc., providing relevant information support). Among them, the vehicle itself (vehicle-side) and the cloud control platform (cloud) each contain an information space and a physical space. The cloud information space contains the vehicle predictive planning algorithm and establishes a real-time digital twin of the surrounding traffic. The vehicle-side continuously uploads the vehicle's status information and receives predictive driving information from the cloud, which is used for re-planning and trajectory tracking control.

[0231] According to the driving trajectory planning method based on vehicle-cloud collaboration proposed in the embodiment of the present application, a surrounding vehicle probability model can be obtained based on the target perception data obtained from the ego vehicle and surrounding vehicles, and then the parameters and parameter uncertainties in the driver model and the vehicle lane change model, the ego vehicle's intention and intention uncertainty are obtained. Based on the parameters and parameter uncertainties, the ego vehicle's intention and intention uncertainty, a closed-loop strategy tree for the ego vehicle is constructed. A parallel computing method is used to search the closed-loop strategy tree for the ego vehicle, and then behavior sequence decisions are obtained, a spatiotemporal trajectory is generated, and the spatiotemporal trajectory of the ego vehicle is planned. The uncertainty of the driver's parameters and intentions is clearly reflected through the surrounding vehicle probability model. The uncertainty propagation mechanism avoids the logical error of "prediction first, decision planning later" while considering uncertainty information, thereby improving the credibility and feasibility of the planning results. The closed-loop strategy tree for the ego vehicle achieves the highest solution efficiency and ensures real-time performance. Good performance can be obtained under long-term predictive decision-making tasks. The trajectory optimization model is used to ensure the rollover safety of the ego vehicle, and the lateral manipulation safety problem of the ego vehicle is solved from the planning layer, and a feasible, safe and smooth trajectory is output. This solves the problems in related technologies, such as being restricted by the field of view of a single vehicle, a small observable range, and limited computing power of a single vehicle, making it difficult to effectively solve long-term planning problems, resulting in low availability of planning results and low prediction accuracy.

[0232] Next, a driving trajectory planning device based on vehicle-cloud collaboration proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0233] Figure 15 This is a block diagram of a driving trajectory planning device based on vehicle-cloud collaboration provided according to an embodiment of the present application.

[0234] like Figure 15 As shown, the driving trajectory planning device 10 based on vehicle-cloud collaboration includes: a generation module 100, a construction module 200 and a planning module 300.

[0235] Among them, the generation module 100 is used to obtain the target perception data of the own vehicle and the surrounding vehicles, and update the initial surrounding vehicle probability model based on the target perception data to obtain the surrounding vehicle probability model, so as to obtain the parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the intention and intention uncertainty of the own vehicle based on the surrounding vehicle probability model.

[0236] The construction module 200 is used to construct a closed-loop policy tree for the ego vehicle based on parameters and parameter uncertainties, the ego vehicle's intention and intention uncertainty, and to search the closed-loop policy tree for the ego vehicle in combination with a parallel computing method to generate a behavioral sequence decision for the ego vehicle.

[0237] The planning module 300 is used to generate the spatiotemporal trajectory of the ego vehicle based on the behavior sequence decision and trajectory optimization model.

[0238] Optionally, in one embodiment of the present application, the generation module 100 includes: a first generation unit, an acquisition unit, and an update unit.

[0239] Among them, the first generation unit is used to segment and extract the road map information along the route according to the destination of the vehicle and the global path planning information, so as to generate the discretized road map information of the vehicle according to the different positions of the vehicle, which is used to establish the initial vehicle probability model.

[0240] The acquisition unit is used to obtain the target perception data of the vehicle and surrounding vehicles based on the real-time traffic twin at the beginning of each planning cycle.

[0241] An updating unit is configured to retrieve discretized road map information based on the vehicle's location in each planning cycle, and update an initial surrounding vehicle probability model in combination with the discretized road map information and parameters and the intention recognition algorithm to obtain a surrounding vehicle probability model, wherein the surrounding vehicle probability model includes at least one of the surrounding vehicle intention and its probability, the driver model, and the vehicle lane change model.

[0242] Optionally, in one embodiment of the present application, the parameter and intention recognition algorithm is based on Bayesian inference, wherein the Bayesian inference includes: selecting the prior distribution of parameters in the driver model based on the initial hypothesis; constructing the posterior distribution of parameters in the intention, driver model and surrounding vehicle lane change model based on the prior distribution; when the posterior distribution cannot be resolved, using the MCMC method to sample the parameters to be identified to obtain sample data of the parameters to be identified; and obtaining the probability of the intention to be identified, parameters and their uncertainty in the parameters to be identified based on the sample data.

[0243] Optionally, in one embodiment of the present application, the construction module 200 includes: a first construction unit, a first screening unit, a second generation unit and a third generation unit.

[0244] The first construction unit is used to construct a vehicle decision sequence including multiple lane-changing behaviors in a long time domain.

[0245] The first screening unit is used to combine the ego vehicle decision sequence and the confidence reachable set to screen out the surrounding vehicles that interact with the ego vehicle.

[0246] The second generation unit is used to obtain the interaction information between the vehicle and the surrounding vehicles by using the forward deduction simulation mechanism.

[0247] The third generation unit is used to perform task allocation in a parallel computing manner based on the interaction information and the ego-vehicle closed-loop policy tree, and dynamically decompose the tasks according to the number of available cores to obtain behavior sequence decisions.

[0248] Optionally, in one embodiment of the present application, the parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the vehicle's intention and intention uncertainty include: the first-order uncertainty of the parameters in the driver model; the first-order uncertainty of the preceding vehicle changing lanes once; the uncertainty of the vehicle's intention entropy; and the cumulative effect of uncertainty over time.

[0249] Optionally, in one embodiment of the present application, the planning module 300 includes: a second screening unit, a fourth generating unit, a second constructing unit and a solving unit.

[0250] The second screening unit is configured to select a high-order polynomial based on the number of constraints on the starting point and the ending point of the vehicle, wherein the high-order polynomial is greater than 5th order.

[0251] The fourth generation unit is used to generate boundary constraints of the vehicle according to the behavior sequence decision.

[0252] The second construction unit is used to optimize the residual degree of freedom coefficient of the boundary constraint condition by using the trajectory optimization model to construct a trajectory optimization model for the rollover state index.

[0253] The solving unit is used to solve the trajectory optimization model to obtain the optimal trajectory of the vehicle and obtain the spatiotemporal trajectory based on the optimal trajectory.

[0254] It should be noted that the above explanation of the embodiment of the driving trajectory planning method based on vehicle-cloud collaboration is also applicable to the driving trajectory planning device based on vehicle-cloud collaboration in this embodiment, and will not be repeated here.

[0255] According to the driving trajectory planning device based on vehicle-cloud collaboration proposed in the embodiment of the present application, a surrounding vehicle probability model can be obtained based on the target perception data of the self-vehicle and surrounding vehicles, and then the parameters and parameter uncertainties in the driver model and the vehicle lane change model, as well as the intention and intention uncertainty of the self-vehicle are obtained. A closed-loop strategy tree for the self-vehicle is constructed based on the parameters and parameter uncertainties, the intention and intention uncertainty of the self-vehicle, and a parallel computing method is used to search the closed-loop strategy tree for the self-vehicle to obtain behavior sequence decisions, generate a spatiotemporal trajectory, and plan the spatiotemporal trajectory of the self-vehicle. The uncertainty of the driver's parameters and intentions is clearly reflected through the surrounding vehicle probability model. The uncertainty propagation mechanism avoids the logical error of "prediction first, decision planning later" while considering uncertainty information, thereby improving the credibility and feasibility of the planning results. The self-vehicle closed-loop strategy tree can achieve the highest solution efficiency and ensure real-time performance. Good performance can be obtained under long-term predictive decision-making tasks. The trajectory optimization model is used to ensure the rollover safety of the self-vehicle, solve the lateral manipulation safety problem of the self-vehicle from the planning layer, and output a feasible, safe and smooth trajectory. This solves the problems in related technologies, such as being restricted by the field of view of a single vehicle, a small observable range, and limited computing power of a single vehicle, making it difficult to effectively solve long-term planning problems, resulting in low availability of planning results and low prediction accuracy.

[0256] Figure 16 This is a schematic diagram of the structure of a server provided according to an embodiment of the present application. The server may include:

[0257] Memory 1601 , processor 1602 , and computer programs stored in the memory 1601 and executable on the processor 1602 .

[0258] When the processor 1602 executes the program, the driving trajectory planning method based on vehicle-cloud collaboration provided in the above embodiment is implemented.

[0259] Furthermore, the server further includes:

[0260] The communication interface 1603 is used for communication between the memory 1601 and the processor 1602 .

[0261] The memory 1601 is used to store computer programs that can be run on the processor 1602 .

[0262] The memory 1601 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0263] If the memory 1601, the processor 1602, and the communication interface 1603 are implemented independently, the communication interface 1603, the memory 1601, and the processor 1602 can be connected to each other via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 16 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0264] Optionally, in a specific implementation, if the memory 1601, the processor 1602 and the communication interface 1603 are integrated on a chip, the memory 1601, the processor 1602 and the communication interface 1603 can communicate with each other through an internal interface.

[0265] The processor 1602 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0266] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the driving trajectory planning method based on vehicle-cloud collaboration as described above is implemented.

[0267] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, implements the above-mentioned driving trajectory planning method based on vehicle-cloud collaboration.

[0268] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0269] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0270] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0271] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or N wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program can be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing it in other suitable ways as necessary, and then storing it in a computer memory.

[0272] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented using hardware, as in another embodiment, it can be implemented using any one or a combination of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0273] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0274] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0275] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A driving trajectory planning method based on vehicle-cloud collaboration, characterized in that: The following steps are involved: Acquire target perception data of the ego vehicle and surrounding vehicles, and update an initial surrounding vehicle probability model based on the target perception data to obtain a surrounding vehicle probability model, so as to obtain parameters and parameter uncertainties in a driver model and a vehicle lane change model, and the intention and intention uncertainty of the ego vehicle based on the surrounding vehicle probability model; Constructing a self-vehicle closed-loop strategy tree based on the parameters and parameter uncertainties, the intention of the self-vehicle and the intention uncertainty, and searching the self-vehicle closed-loop strategy tree in combination with the self-vehicle closed-loop strategy tree and a parallel computing method to generate a behavior sequence decision of the self-vehicle; A spatiotemporal trajectory of the ego vehicle is generated based on the behavior sequence decision and trajectory optimization model.

2. The method according to claim 1, characterized in that The step of acquiring target perception data of the vehicle and surrounding vehicles, and updating an initial surrounding vehicle probability model based on the target perception data to obtain a surrounding vehicle probability model includes: Segmenting and extracting the road map information along the route according to the destination of the vehicle and the global path planning information, so as to generate the discretized road map information of the vehicle according to the different positions of the vehicle, so as to establish the initial vehicle probability model; At the beginning of each planning cycle, target perception data of the self-vehicle and the surrounding vehicles are obtained according to the real-time traffic twin; In each planning cycle, the discretized road map information is retrieved through the position of the ego vehicle, and the initial surrounding vehicle probability model is updated in combination with the discretized road map information and parameters and the intention recognition algorithm to obtain a surrounding vehicle probability model, wherein the surrounding vehicle probability model includes at least one of the surrounding vehicle intention and its probability, the driver model and the vehicle lane change model.

3. The method according to claim 2, characterized in that The parameter and intention recognition algorithm is based on Bayesian inference, wherein the Bayesian inference includes: Selecting a prior distribution of parameters in the driver model based on an initial assumption; Constructing a posterior distribution of parameters in the intention, the driver model, and the surrounding vehicle lane change model based on the prior distribution; When the posterior distribution cannot be resolved, a Markov chain Monte Carlo (MCMC) method is used to sample the parameters to be identified to obtain sample data of the parameters to be identified; Based on the sample data, the probability of the intention to be identified in the parameters to be identified, the parameters and their uncertainty are obtained.

4. The method according to claim 1, characterized in that: The constructing of the self-vehicle closed-loop strategy tree, and searching the self-vehicle closed-loop strategy tree in combination with the self-vehicle closed-loop strategy tree and a parallel computing method to generate a behavior sequence decision of the self-vehicle, includes: Constructing a decision sequence of the ego vehicle that includes multiple lane-changing behaviors over a long time domain; Combining the self-vehicle decision sequence and the confidence reachable set to screen out surrounding vehicles that interact with the self-vehicle; Using a forward deduction simulation mechanism to obtain the interaction information between the self-vehicle and the surrounding vehicles; Based on the interaction information and the self-driving vehicle closed-loop strategy tree, the parallel computing method is used to allocate tasks, and the tasks are dynamically decomposed according to the number of available cores to obtain the behavior sequence decision.

5. The method according to claim 1, characterized in that The parameters and parameter uncertainties in the driver model and the vehicle lane change model, and the intention and intention uncertainty of the vehicle include: first-order uncertainty in parameters of the driver model; The first-order uncertainty of the preceding vehicle changing lanes once; The uncertainty of the ego-vehicle intention entropy; The cumulative effect of uncertainty over time.

6. The method according to claim 1, characterized in that The generating the spatiotemporal trajectory of the vehicle based on the behavior sequence decision and trajectory optimization model includes: Selecting a high-order polynomial based on the number of constraints on the starting point and the ending point of the vehicle, wherein the high-order polynomial is greater than 5th order; Generating boundary constraints of the ego vehicle according to the behavior sequence decision; The trajectory optimization model is used to optimize the residual degree of freedom coefficient of the boundary constraint condition to construct a trajectory optimization model for the rollover state index; The trajectory optimization model is solved to obtain an optimal trajectory of the ego vehicle, and the spatiotemporal trajectory is obtained based on the optimal trajectory.

7. A driving trajectory planning device based on vehicle-cloud collaboration, characterized in that: The device comprises: A generation module is used to obtain target perception data of the ego vehicle and surrounding vehicles, and update an initial surrounding vehicle probability model based on the target perception data to obtain a surrounding vehicle probability model, so as to obtain parameters and parameter uncertainties in a driver model and a vehicle lane change model, and the intention and intention uncertainty of the ego vehicle based on the surrounding vehicle probability model; A construction module, configured to construct a self-vehicle closed-loop strategy tree based on the parameters and parameter uncertainties, the intention of the self-vehicle and the intention uncertainty, and search the self-vehicle closed-loop strategy tree in combination with the self-vehicle closed-loop strategy tree and a parallel computing method to generate a behavior sequence decision of the self-vehicle; A planning module is used to generate a spatiotemporal trajectory of the ego vehicle based on the behavior sequence decision and trajectory optimization model.

8. A server, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the driving trajectory planning method based on vehicle-cloud collaboration as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the driving trajectory planning method based on vehicle-cloud collaboration as described in any one of claims 1-6.

10. A computer program product, characterized in that It includes a computer program which, when executed, is used to implement the driving trajectory planning method based on vehicle-cloud collaboration as described in any one of claims 1-6.

Images