Automatic driving behavior optimization method and device, equipment and storage medium
By obtaining the perceived data of the bicycle driving and performing parameter screening and probability model regression analysis using functional finite state machine and maximum likelihood estimation, the autonomous driving behavior is optimized, and the decision-making system update and maintenance problems based on rule definition are solved, and performance and safety are improved sex.
Patent Information
- Application Number
- CN202411006468.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-05
AI Technical Summary
The autonomous driving decision system based on rules definition needs to be continuously updated and maintained, and there are redundant rules in the system, which affects performance.
By obtaining the bicycle driving perception data, parameter screening and probability model regression analysis are performed using functional finite state machines and maximum likelihood estimation, optimize autonomous driving behavior, reduce parameter redundancy, and improve algorithm performance and interpretability.
It reduces parameter redundancy in decision-making of autonomous driving behavior, improves algorithm performance and safety, and enhances the transparency of the model and problem positioning efficiency.
Smart Images

Figure CN118898053A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of autonomous driving technology, and in particular to autonomous driving behavior optimization methods, devices, equipment and storage media. Background Art
[0002] In recent years, autonomous driving technology has become the focus of major car companies due to its safe and convenient characteristics, and the functional scenarios are becoming more and more complex, which also brings difficulties to the decision-making planning of autonomous driving. At present, the mainstream decision-making planning is based on rules such as speed control, following, and steering. Its logic is based on human driving behavior and has the advantages of clear structure, safety, interpretability, and easy implementation. However, due to the complexity and diversity of real driving scenarios, rule-based planning methods have encountered a large bottleneck. The decision-making system defined by rules needs to be constantly updated and maintained, and the rules in the system are also redundant, which affects its performance. The decision-making planning method based on learning can adaptively adjust the decision-making mechanism by interacting with unknown environments and has stronger environmental adaptability. However, the decision-making method based on learning lacks interpretability and has poor problem location efficiency. At the same time, there are high requirements for the amount of test data. For location scenarios outside the training data, the model may not be able to make effective decisions, and the security of the obtained decision model may not be better than that of the rule-based expert system.
[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention
[0004] The main purpose of this application is to provide an autonomous driving behavior optimization method, device, equipment and storage medium, aiming to solve the technical problems that the current decision-making system based on rule definition needs to be constantly updated and maintained, and the rules within the system are also redundant, which affects the performance of autonomous driving.
[0005] To achieve the above objectives, the present application proposes an automatic driving behavior optimization method, which includes: Acquire self-driving perception data; Selecting the initial input and output parameter set of the prediction model according to the functional finite state machine conditions and prior rules; Parameter screening is performed based on the self-driving perception data to obtain the input and output parameter sets; Based on the input and output parameter sets and the maximum likelihood estimation probability, a probability model regression analysis is performed to obtain the model parameters; The probability model is updated according to the model parameters to obtain a new probability model; Analyze driving behavior based on the self-vehicle perception data based on the new probability model to obtain the analysis results; The analysis result is compared with the result obtained based on the finite state machine condition to obtain a comparison result, and the autonomous driving behavior is optimized according to the comparison result.
[0006] In one embodiment, the parameter screening based on the self-driving perception data to obtain the input and output parameter sets includes: Obtaining the number of target parameters, and obtaining a total feature parameter set according to the self-vehicle driving perception data; Screening the features in the total feature parameter set based on the target parameter quantity to obtain a target parameter set; A joint mutual information analysis is performed on the target parameter set and the pre-selected driving behavior parameters to obtain an input and output parameter set.
[0007] In one embodiment, the target parameter set and the pre-selected driving behavior parameters are subjected to joint mutual information analysis to obtain an input and output parameter set, including: Calculating joint mutual information between each parameter in the target parameter set and the preselected driving behavior; When the mutual information meets the preset feature requirements, the parameters in the target parameter set whose mutual information meets the requirements are used as the input and output parameter sets.
[0008] In one embodiment, the probability model regression analysis is performed based on the input and output parameter sets and the maximum likelihood estimation probability to obtain the model parameters, including: Input the probability model according to the input and output parameter set to obtain initial observation values and initial actions; Obtaining a potential state based on the initial observation value and the initial action; The latent state is model trained according to the latent state expectation vector and the latent state covariance matrix to obtain model parameters.
[0009] In one embodiment, the driving behavior is analyzed based on the new probability model according to the vehicle perception data to obtain the analysis result, including: Calculating maximum likelihood estimates of model parameters based on the new probability model according to the ego-vehicle perception data; Performing Gaussian mixture model regression analysis based on the maximum likelihood estimate to obtain a multimodal Gaussian distribution; The analytical results of the model parameters are determined according to the multimodal Gaussian distribution.
[0010] In one embodiment, the comparing the analysis result with the result obtained based on the finite state machine condition to obtain a comparison result, and optimizing the autonomous driving behavior according to the comparison result includes: Obtaining a real driving behavior according to the self-vehicle perception data; Comparing the actual driving behavior with the driving behavior to obtain a comparison result; According to the comparison result, the self-vehicle perception data is reduced in redundancy to obtain the self-vehicle driving perception data after redundancy reduction; The automatic driving behavior is optimized according to the reduced-redundancy self-driving perception data.
[0011] In one embodiment, the step of obtaining the self-vehicle driving perception data includes: Obtain driving information of the vehicle, surrounding intelligent agents and environment information; Slicing the self-vehicle driving information to obtain driving behavior and driving behavior habit parameters; Performing cluster analysis on the driving behavior and the driving behavior habit based on the surrounding intelligent agent information to obtain static parameters; The static parameters are divided into driving habits based on the environmental information to obtain self-vehicle driving perception data.
[0012] In addition, to achieve the above-mentioned purpose, the present application also proposes an automatic driving behavior optimization device, the automatic driving behavior optimization device comprising: A parameter acquisition module is used to obtain the driving perception data of the vehicle; The parameter acquisition module is further used to perform parameter screening on the self-vehicle driving perception data according to the maximization of the joint mutual information to obtain the predicted characteristic parameters; The parameter acquisition module is further used to observe the preset model state according to the predicted characteristic parameters to obtain model parameters; A driving behavior analysis module, used for performing regression analysis based on the model parameters and maximum likelihood estimation probability to obtain driving behavior; The driving behavior analysis module is further used to analyze the driving behavior based on the vehicle perception data, obtain an analysis result, and realize the analysis of the vehicle driving behavior.
[0013] In addition, to achieve the above-mentioned purpose, the present application also proposes an autonomous driving behavior optimization device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the autonomous driving behavior optimization method as described above.
[0014] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the autonomous driving behavior optimization method described above are implemented.
[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the autonomous driving behavior optimization method described above.
[0016] One or more technical solutions proposed in the present application have at least the following technical effects: analyzing the driving behavior of the self-vehicle through the perception data of the self-vehicle driving process, screening parameters according to the analysis results, and streamlining the parameters related to the self-vehicle driving behavior, which can reduce the parameter redundancy of the autonomous driving behavior when making behavioral decisions, improve the algorithm performance, and ensure safety at the same time, wherein the parameters of the self-vehicle driving perception data are screened according to the maximization of the joint mutual information to obtain predicted feature parameters; the preset model state observation is performed according to the predicted feature parameters to obtain the model parameters, which improves the interpretability of the learning method, improves the transparency of the model, and facilitates problem location. Parameter redundancy can be reduced according to the driving style, and the algorithm performance is improved while ensuring safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0019] Figure 1 A flowchart of the first embodiment of the automatic driving behavior optimization method of the present application is provided; Figure 2 A flowchart of the second embodiment of the automatic driving behavior optimization method of the present application is provided; Figure 3 This is a schematic diagram of the module structure of the automatic driving behavior optimization device according to an embodiment of the present application; Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the autonomous driving behavior optimization method in the embodiment of the present application.
[0020] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0021] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.
[0022] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.
[0023] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an automatic driving behavior optimization device, etc. The following takes the automatic driving behavior optimization device as an example to illustrate this embodiment and the following embodiments.
[0024] Based on this, the present application embodiment provides an automatic driving behavior optimization method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the autonomous driving behavior optimization method of the present application.
[0025] In this embodiment, the automatic driving behavior optimization method includes steps S10 to S50: Step S10, obtaining driving perception data of the vehicle.
[0026] It is understandable that the self-vehicle driving perception data may include self-vehicle driving information, surrounding intelligent agent information, environmental information, etc.
[0027] It should be understood that the acquisition of the self-vehicle driving perception data can be obtained by collecting data from various sensors of the self-vehicle and further processing the data, such as image acquisition devices, radars, etc.
[0028] In a feasible implementation, step S10 may include steps A11 to A14. For specific steps, please refer to Figure 2 : Step A11, obtaining the vehicle driving information, surrounding intelligent agent information and environment information.
[0029] It is understandable that the self-driving information may include vehicle status information, such as speed, acceleration, angular velocity and other information; the self-driving information may be obtained through the vehicle's own sensors, such as speed sensors, steering angle sensors and other devices to obtain the vehicle's current speed, steering angle, acceleration and other status information; the self-driving information may also include vehicle positioning and navigation information, which can obtain the vehicle's real-time location information through the positioning system and combine it with map data; the self-driving information may also include vehicle control information, such as: throttle, brake, light and other control information, which is directly obtained through the vehicle's control system.
[0030] It is understandable that surrounding intelligent agents may include people, cars and other objects around the vehicle; surrounding intelligent agent information may include detecting objects around the vehicle through radar and laser radar (LiDAR) and measuring their distance and relative speed to the vehicle; visual information around the vehicle can be obtained through the camera; combined with image recognition and deep learning technology, pedestrians, vehicles, traffic signs, etc. can be identified and their status and behavior can be understood; through wireless communication between vehicles, information such as the driving status and intention of other vehicles can be obtained.
[0031] It should be noted that environmental information can include weather and road condition information, geographic and environmental data, and integrated environmental information. The information can be obtained through weather information obtained through meteorological services, and road condition information obtained through traffic management systems or third-party services, including road congestion, accidents, construction, etc.; map data and geographic information systems (GIS) can also be used to obtain information such as the geometry, slope, curvature of the road, as well as surrounding environmental features such as buildings and green belts; the data from different sensors can then be fused and processed to obtain more comprehensive and accurate environmental information.
[0032] In the specific implementation, based on the actual vehicle driving data, the vehicle perception data is extracted, including the vehicle (coordinates, speed, acceleration) and surrounding intelligent agent information and environmental information. The intelligent agent is denoted as: A={a i}, i=1,2,3···N; the historical trajectory information of each agent is recorded as a h =(x th ,y th , v x th , v y th , a x th , a y th , type), including coordinates, speed, acceleration and agent category, etc. Environmental information includes lanes, traffic signs, etc.
[0033] Step A12, slicing the driving information of the vehicle to obtain driving behavior and driving behavior habit parameters.
[0034] It is understandable that data slicing of self-driving information is a data processing process that aims to divide continuous driving data into multiple segments or subsets according to specific dimensions or conditions.
[0035] It should be understood that there are many ways to perform data slicing on the self-vehicle driving information, such as time slicing, space slicing, vehicle state slicing, driver behavior slicing, etc.
[0036] It should be noted that the driving behavior in the driving behavior and driving behavior habit parameters can be a driving behavior composed of parameters such as the speed, acceleration, angular velocity, etc. of the vehicle; the driving behavior habit can be the maximum angular velocity, maximum acceleration, maximum jerk, etc. obtained based on multiple driving behaviors.
[0037] It should be emphasized that in this embodiment, data slicing of the vehicle driving information can be performed according to driving behavior and parameters corresponding to the driver's specific operations, which can better analyze the driver's driving behavior habits.
[0038] Step A13: performing cluster analysis on the driving behavior and the driving behavior habit based on the surrounding intelligent agent information to obtain static parameters.
[0039] It is understandable that the cluster analysis of the driving behavior and the driving behavior habit based on the surrounding intelligent agent information may first be performed on the surrounding intelligent agent information by performing data cleaning, data feature extraction and data normalization.
[0040] It should be understood that data cleaning of surrounding intelligent agent information can be data deduplication, interference data removal, etc., and data feature extraction can be extracting features related to the vehicle's driving behavior and driving habits from the surrounding intelligent agent information, such as the type, speed, acceleration, direction, distance, etc. of the surrounding intelligent agent, normalizing each feature, and then clustering the normalized data.
[0041] It should be understood that cluster analysis can be performed by clustering algorithm or by setting clustering parameters. This embodiment performs cluster analysis by clustering algorithm, and other clustering methods can also be selected according to actual conditions, which is not limited in this embodiment.
[0042] It should be noted that clustering based on agent information can obtain clustering results, extract representative parameters from the clustering results, or directly select the cluster center of each clustering result. These parameters can describe the main characteristics of driving behaviors and habits in the cluster. Static parameters may include average speed, acceleration distribution, steering angle, average distance to other agents, etc.
[0043] Step A14, dividing the static parameters into driving habits based on the environmental information to obtain self-vehicle driving perception data.
[0044] It is understandable that the self-vehicle driving perception data may be based on perception data in different environments, and the self-vehicle driving data may be divided based on driving habits in different environments.
[0045] It should be noted that the driving habit classification of the static parameters based on the environmental information to obtain the self-vehicle driving perception data can be firstly integrated and analyzed by firstly, and the static parameters (such as average speed, acceleration distribution, etc.) previously extracted from the surrounding intelligent agent information are mapped with the environmental information, and the correlation between them is analyzed. Through the correlation analysis, the specific habit characteristics that the driver may show under different environmental conditions are identified, such as conservative driving, aggressive driving, strong adaptability, etc. Based on the correlation analysis results of the environmental information and the static parameters, the classification criteria of driving habits are formulated. These criteria may include driving behavior characteristics under different road types, traffic conditions and weather conditions. According to the classification criteria, the static parameters are classified and labeled, and the driving habits are divided into different categories, such as cautious, adventurous, economical, etc. The driving habit labels are integrated with the corresponding environmental information data to generate comprehensive data containing driving habits and environmental perception information, and the comprehensive data is formatted into a standard data structure to facilitate subsequent analysis, processing and application, and stored in an appropriate data storage system.
[0046] In this embodiment, by acquiring the vehicle information, surrounding intelligent agent information and environmental information and performing information fusion, it not only includes static parameters, but also integrates driving habits and environmental perception information, providing more comprehensive and accurate data support for subsequent driving behavior analysis, intelligent driving decision-making, etc.
[0047] The above are only feasible implementations of step S10 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S10.
[0048] Step S20, selecting an initial input and output parameter set of the prediction model according to the functional finite state machine conditions and a priori rules.
[0049] It can be understood that joint mutual information, or mutual information (abbreviated as MI), is a core concept in information theory, which is used to measure the amount of shared information between two random variables. Simply put, it represents the uncertainty of a random variable reduced due to the knowledge of another random variable.
[0050] It should be understood that the initial input-output parameter set can be the characteristic parameters of the input function state machine after redundancy reduction and optimization. Simply put, the initial input function state machine may have 10 parameters, but only 5 characteristic parameters are needed after redundancy reduction and optimization.
[0051] It should be noted that the characteristic parameters can be key parameters (state machine parameters) roughly selected from the parameters defined according to the functional state machine and the priori rules, such as TTC (Time to Collision), safety distance, etc. There may be parameter redundancy. The input parameters are x I , the output parameter is x O .
[0052] Step S30, performing parameter screening based on the self-vehicle driving perception data to obtain an input and output parameter set.
[0053] It is understandable that parameter screening based on self-driving perception data can be performed by state observation through a probability model, such as state observation during training based on a functional state machine model obtained through a Markov decision process.
[0054] It should be understood that the probability model is continuously trained and tested according to the initial input-output parameter set, and the initial model parameters are optimized according to the training results to obtain the input-output parameter set.
[0055] Step S40, performing probability model regression analysis based on the input and output parameter sets and the maximum likelihood estimation probability to obtain model parameters.
[0056] It is understandable that Maximum Likelihood Estimation (MLE) is a method for estimating parameters in a probability model. Simply put, the model parameters should maximize the probability of the observed data occurring.
[0057] It should be noted that performing regression analysis based on the model parameters and the maximum likelihood estimation probability may include defining a likelihood function, maximizing the likelihood function, and estimating parameters.
[0058] In the specific implementation, for data of different driving habits, the corresponding model parameter Z is obtained using maximum likelihood estimation. Based on the model parameter Z and its maximum likelihood estimation, a Gaussian mixture model is used for regression analysis, that is, the output parameter is expressed as a multimodal Gaussian distribution of the input parameter (model parameter), that is:
[0059] in, is the weight coefficient, is the maximum likelihood estimate of the parameters obtained above. It can be understood as the possibility that the current observation obeys the i-th Gaussian component, and it can be understood as the probability estimate of the current observation entering a given state, which represents the potential state and driving behavior.
[0060] In a feasible implementation, step S30 may include steps A41 to A43: Step A41, input the probability model according to the input and output parameter set to obtain initial observation values and initial actions It should be understood that the probability model can be obtained by a Markov decision process, and the input and output parameter sets are predicted based on the Markov decision process to obtain an observation value of the state, namely the initial observation value, and a driving behavior action, namely the initial action, is predicted based on the observation value.
[0061] Step A42, obtaining a potential state based on the initial observation value and the initial action.
[0062] It is understandable that, considering that the actual driver makes driving decisions during driving, they are not based on the environmental information perceived at the current moment, but on the understanding of the current perceived environment, which can be expressed as a latent state using hidden Markov processing.
[0063] It should be understood that each observation may correspond to a potential state, and the observed state at time t is x t , the potential state is expressed as y t ∈Y={1,2,3···,k}.
[0064] The observation model is expressed as p(x t |y t = k,μ k , σ k )~N(x t |μ k , σ k ), and assumes a normal distribution.
[0065] Step A43, performing model training on the latent state according to the latent state expected vector and the latent state covariance matrix to obtain model parameters.
[0066] It should be understood that based on the observation model, μ k is the expected vector and covariance matrix of the corresponding potential state.
[0067] It should be noted that based on the observation model, the joint probability distribution of the potential states can be expressed as:
[0068] Among them, A is the Markov state transfer matrix of the potential state, π is the initial probability, Z is the parameter to be learned, Z={π, A, μ, σ}.
[0069] It should be noted that the model training of the latent state is carried out according to the latent state expectation vector and the latent state covariance matrix, and the model parameters can be set according to the model structure of the hidden Markov model based on the latent state expectation vector and the covariance matrix and the observed data, such as the mean and covariance of the latent state, the transition probability or emission probability of the model, etc., and the model training is carried out based on the set parameters.
[0070] It should be emphasized that before training, a target optimization function is determined, which can be the maximum likelihood estimation or the maximum a posteriori probability, which can be limited according to the actual situation, and then the model parameters are iteratively updated using gradient descent, expectation maximization (EM) algorithm or other optimization techniques. In each iteration, the expectation and covariance of the potential state are calculated based on the current parameters, and these values are used to update the model parameters.
[0071] In this embodiment, the hidden Markov model is trained based on the pre-obtained latent state expectation vector and latent state covariance matrix, so that the hidden Markov model obtains parameters that are more in line with the prediction target. At the same time, the trained hidden Markov model can handle hidden states and learn the relationship between the observation sequence and the hidden state, thereby more accurately predicting the driving behavior of the vehicle.
[0072] The above are only feasible implementations of step S30 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S10.
[0073] Step S50, updating the probability model according to the model parameters to obtain a new probability model.
[0074] It is understandable that updating the probability model based on the model parameters can enable the probability model to better predict a driving behavior that is more in line with the current driving environment based on the current driving environment.
[0075] Step S60: Analyze the driving behavior according to the vehicle perception data based on the new probability model to obtain an analysis result.
[0076] In a feasible implementation, step S60 may include steps A61 to A63: Step A61, calculating the maximum likelihood estimate of the model parameters based on the new probability model according to the vehicle perception data.
[0077] It will be appreciated that computing maximum likelihood estimates of model parameters typically involves optimizing the likelihood function to find parameter values that maximize the likelihood function.
[0078] It should be noted that in the context of hidden Markov models, the estimation of state transition probabilities and observation probabilities is involved.
[0079] In practice, a likelihood function is defined as the probability of an observation sequence occurring given the model parameters and the observation sequence. In a hidden Markov model, the likelihood function is usually expressed as: [P(O|\lambda)], where (O) is the observation sequence and (\lambda) is the model parameter (including state transition probability and observation probability); the Baum-Welch algorithm (or forward-backward algorithm) is used to iteratively calculate the maximum likelihood estimates of the model parameters. The Baum-Welch algorithm works by maximizing the expected log-likelihood of the complete data, even though the complete data (i.e., the hidden state and the observed state) are not visible; initial values are set for the state transition probabilities and the observation probabilities. These can be random or based on some heuristic method; the state transition probabilities and the observation probabilities are iteratively updated using the Baum-Welch algorithm until a certain convergence criterion is met (such as the parameter change is less than a certain threshold); after enough iterations, the algorithm will converge to a set of parameter values, which are the maximum likelihood estimates of the model parameters.
[0080] Step A62, performing Gaussian mixture model regression analysis based on the maximum likelihood estimate to obtain a multimodal Gaussian distribution.
[0081] It is understandable that the Gaussian Mixture Model (GMM) is a probability model that assumes that all data is a mixture of a finite number of Gaussian distributions (also called normal distributions). Each Gaussian distribution represents a category or subgroup, and the entire data set is a mixture of these subgroups in different proportions.
[0082] It should be understood that the multimodal Gaussian distribution is a probability distribution model commonly used in statistics, which can describe the joint probability distribution between multiple random variables. In this model, the distribution of data is no longer unimodal, but may have multiple peaks, that is, multiple modes.
[0083] It should be noted that Gaussian mixture model regression analysis can be Gaussian Process Regression (GPR), which is a non-parametric model that uses Gaussian Process (GP) priors to perform regression analysis on data. It assumes that the regression residual and Gaussian process prior are two parts of the model and is solved according to Bayesian inference. Gaussian process is a random process in which each finite set of random variables obeys a multivariate normal distribution, that is, each of their finite linear combinations is normally distributed, which can provide uncertainty estimates for the prediction results.
[0084] Step A63, determining the analytical results of the model parameters according to the multimodal Gaussian distribution.
[0085] It can be understood that driving data is a mixture of multiple Gaussian distributions, each of which represents a driving behavior.
[0086] In this embodiment, by using a hidden Markov model to describe the driving state, the parameters in, such as the state transition probability and the observation probability, can be estimated by maximum likelihood estimation, which can help to more accurately understand the driving behavior and make predictions or analyses based on these parameters.
[0087] The above are only feasible implementations of step S60 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S60.
[0088] Step S70, comparing the analysis result with the result obtained based on the finite state machine condition to obtain a comparison result, and optimizing the autonomous driving behavior according to the comparison result.
[0089] In a feasible implementation, step S70 may include steps A71 to A74: Step A71, obtaining the actual driving behavior according to the vehicle perception data.
[0090] It can be understood that the self-vehicle perception data can be the self-vehicle driving parameters, surrounding intelligent body parameters and environmental parameters during the self-vehicle driving process over a period of time.
[0091] It should be understood that the actual driving behavior of the self-vehicle perception data in the previous second can be obtained based on the self-vehicle perception data in the next second.
[0092] It should be noted that in the process of predicting and analyzing the driving behavior of the vehicle, the parameters of the Markov decision model can be improved based on the continuous collection of the vehicle's perception data, so that the model can predict the driving action at the current moment, that is, the driving behavior, more accurately.
[0093] Step A72: Compare the actual driving behavior with the driving behavior to obtain a comparison result.
[0094] Step A73, reducing the redundancy of the vehicle perception data according to the comparison result to obtain reduced-redundancy vehicle driving perception data, thereby realizing analysis of the vehicle driving behavior.
[0095] It is understandable that redundancy reduction of the vehicle perception data can be performed by removing data that has no correlation or little correlation with the predicted vehicle driving behavior from the vehicle perception data identified by model training. This can enable more efficient data processing and achieve faster and more accurate vehicle driving behavior analysis.
[0096] It is worth noting that according to the parameters corresponding to each potential state, the driving behavior is analyzed and the redundancy of the rules of the finite state machine is reduced. For the functional conditions based on the finite state machine, for the functional state under the given data, it is compared with the results of the prediction model based on human driving data to check whether there is redundancy in the condition settings based on the finite state machine, and redundancy is reduced according to different driving habits; the actual vehicle road test verifies that for possible functional abnormalities, the problem can be tracked based on the driving behavior analysis in the previous step, and the above process is repeated to perform functional iteration.
[0097] Step A74, optimizing the autonomous driving behavior according to the reduced-redundancy self-driving perception data.
[0098] It is understandable that the driving behavior prediction result is obtained based on the original finite state machine prediction parameters, and the driving behavior prediction is performed based on the reduced-redundancy probability model. The redundant parameters of the finite state machine are made according to the difference between the two results, which can effectively reduce the recognition and reaction speed of the driving scene in the driving behavior prediction process.
[0099] In this implementation, model parameters are trained through a hidden Markov decision model, and then the model parameters are screened through joint mutual information, so as to achieve more efficient data processing and faster and more accurate self-driving behavior analysis.
[0100] The above are only feasible implementations of step S50 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S50.
[0101] This embodiment provides an automatic driving behavior optimization method, which analyzes the driving behavior of the vehicle through the perception data of the vehicle's driving process, screens parameters according to the analysis results, and simplifies the parameters related to the vehicle's driving behavior. This can reduce the parameter redundancy of the automatic driving behavior when making behavioral decisions, improve the algorithm performance, and ensure safety at the same time.
[0102] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2 , step S20, the automatic driving behavior optimization method further includes steps S21 to S23: Step S21, obtaining the number of target parameters, and obtaining a total feature parameter set according to the self-vehicle driving perception data.
[0103] It is understandable that the target parameter may be a parameter obtained by filtering the acquired driving perception data of the vehicle.
[0104] It should be understood that the parameters after filtering the self-vehicle driving perception data do not affect the accuracy of the predicted state in the hidden Markov decision model.
[0105] It should be noted that effective feature redundancy reduction of self-driving perception data can effectively reduce the amount of data processing, increase the data processing speed, and enable faster self-driving prediction.
[0106] It can be understood that the target number of parameters may be a predetermined number of parameters retained after redundancy reduction.
[0107] It should be further explained that the total feature parameter set obtained according to the self-vehicle driving perception data may be a feature set including all parameters of the self-vehicle driving perception data.
[0108] Step S22: screening the features in the total feature parameter set based on the target parameter quantity to obtain a target parameter set.
[0109] It is understandable that there may be multiple target parameter sets, and the number of parameters in each target parameter set is the same.
[0110] It should be understood that a target parameter number parameters are randomly selected from the total feature parameter set as a target parameter set, and mutual information analysis is performed on each target parameter set.
[0111] Step S23, performing joint mutual information analysis on the target parameter set and the pre-selected driving behavior parameters to obtain an input and output parameter set.
[0112] It is understandable that the preselected driving behavior parameters may be the vehicle speed, acceleration and angular velocity.
[0113] It should be noted that the mutual information analysis here can be a joint mutual information maximization analysis. Joint mutual information maximization is an optimization strategy that aims to find the best correlation or dependency between multiple variables by maximizing the joint mutual information between them. In driving behavior analysis, joint mutual information maximization can be used to determine which parameter combinations can best describe or predict specific driving behaviors.
[0114] It should be emphasized that each set of target parameter sets and pre-selected driving behavior parameters are analyzed, and the target parameters in a set of target parameter sets with the largest joint mutual information are used as the input and output parameter sets.
[0115] In a feasible implementation, step S23 may include steps A231-A232: Step A231, calculating the joint mutual information between each parameter in the target parameter set and the pre-selected driving behavior.
[0116] It is understandable that calculating the mutual information (MI) between each parameter in the target parameter set and the pre-selected driving behavior is a method for quantifying the strength of the correlation between two variables. Mutual information reflects the reduction in uncertainty of one variable when the other variable is known.
[0117] It should be understood that in driving behavior analysis, mutual information can help us understand which parameters are most relevant to specific driving behaviors (such as aggressive driving, conservative driving, etc.).
[0118] In a specific implementation, a target parameter set and a pre-selected driving behavior are determined (this can be a pre-classified driving behavior category based on certain criteria (such as acceleration threshold, braking frequency, etc.); for each parameter in the target parameter set and the pre-selected driving behavior, their joint probability distribution (P(X, Y)) is calculated, where (X) is the parameter and (Y) is the driving behavior. At the same time, the marginal probability distributions (P(X)) and (P(Y)) of the parameters and driving behaviors are calculated. The mutual information (I(X; Y)) can be calculated by the following formula: I(X;Y)=x∈X∑y∈Y∑P(x,y)log(P(x)P(y)P(x,y)) Where (x) and (y) represent the values of parameters and driving behavior respectively.
[0119] It should be noted that the larger the calculated mutual information value is, the stronger the correlation between the parameter and the pre-selected driving behavior is, and the parameters can be sorted according to the mutual information value to identify the parameters most relevant to the driving behavior.
[0120] It is worth emphasizing that the calculation of mutual information may be affected by the amount of data and the discretization method. Therefore, when performing the calculation, we should ensure that we use enough data and choose a suitable discretization strategy; if the target parameters and driving behaviors are discrete in themselves, or can be directly represented as discrete values in some way (for example, using classification labels), their mutual information can be directly calculated.
[0121] Step A232: When the mutual information meets the preset feature requirements, the parameters in the target parameter set whose mutual information meets the requirements are used as the input and output parameter sets.
[0122] In a specific implementation, feature parameter selection can use a joint mutual information maximization algorithm to select prediction feature parameters. The specific process is as follows: Assume that the total feature parameter set is G, the selected feature set is S=φ, and the number of selected feature parameters is m; the mutual information between the selected feature parameters and the output parameters can be calculated by referring to the following formula:
[0123] Based on the mutual information calculation formula, choose I (x I ;x O ) The largest characteristic parameter x IC , and update G; Among them, G=G\{x IC},S={x IC}.
[0124] Repeat the calculation of the mutual information value of each parameter in the target parameter set until the preset feature requirements are met. The preset feature requirements can be referred to the following formula:
[0125] Update G based on the above formula, G=G\{x IC},S=S∪{x IC}.
[0126] The set S is the prediction feature parameter.
[0127] In this implementation, maximizing the joint mutual information can be used as an effective feature selection method to help us screen out the feature subset that is most relevant to the target driving behavior and find the most predictive parameter combination.
[0128] The above are only feasible implementations of step S23 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S23.
[0129] This embodiment provides an automatic driving behavior optimization method, which calculates the mutual information value between each parameter in each target parameter set and the preset feature, and considers the relationship between multiple parameters (or features) and the target variable, thereby making full use of the information of multi-dimensional features. This helps to find those feature combinations that may seem unrelated individually, but have a significant impact on the target variable when combined, so as to select parameters with a higher degree of connection with the preset features for self-driving vehicle behavior prediction, which can help improve the accuracy of the prediction model and enable us to more accurately identify and analyze driving behavior.
[0130] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the autonomous driving behavior optimization method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.
[0131] This application also provides an automatic driving behavior optimization device, please refer to Figure 3 , the automatic driving behavior optimization device comprises: The parameter acquisition module 10 is used to acquire the driving perception data of the vehicle; The parameter screening module 20 is used to select the initial input and output parameter set of the prediction model according to the functional finite state machine conditions and the prior rules; The parameter screening module 20 is also used to screen parameters based on the self-vehicle driving perception data to obtain an input and output parameter set; The parameter screening module 20 is also used to perform probability model regression analysis based on the input and output parameter sets and the maximum likelihood estimation probability to obtain model parameters; A driving behavior optimization module 30, used to update the probability model according to the model parameters to obtain a new probability model; The driving behavior optimization module 30 is further used to analyze the driving behavior according to the vehicle perception data based on the new probability model to obtain the analysis result; The driving behavior optimization module 30 is further used to compare the analysis result with the result obtained based on the finite state machine condition to obtain a comparison result, and optimize the automatic driving behavior according to the comparison result.
[0132] The autonomous driving behavior optimization device provided by the present application adopts the autonomous driving behavior optimization method in the above embodiment, which can solve the technical problem that the current decision-making system based on rule definition needs to be constantly updated and maintained, and the rules in the system are also redundant, which affects the performance of autonomous driving. Compared with the prior art, the beneficial effects of the autonomous driving behavior optimization device provided by the present application are the same as the beneficial effects of the autonomous driving behavior optimization method provided by the above embodiment, and the other technical features in the autonomous driving behavior optimization device are the same as the features disclosed in the above embodiment method, which will not be repeated here.
[0133] The present application provides an autonomous driving behavior optimization device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the autonomous driving behavior optimization method in the above-mentioned embodiment one.
[0134] Reference below Figure 4, which shows a schematic diagram of the structure of an automatic driving behavior optimization device suitable for implementing the embodiment of the present application. The automatic driving behavior optimization device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The autonomous driving behavior optimization device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0135] like Figure 4 As shown, the automatic driving behavior optimization device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the xxx device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the autonomous driving behavior optimization device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an autonomous driving behavior optimization device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.
[0136] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.
[0137] The autonomous driving behavior optimization device provided by this application adopts the autonomous driving behavior optimization method in the above embodiment, which can solve the technical problem of autonomous driving behavior optimization. Compared with the prior art, the beneficial effects of the autonomous driving behavior optimization device provided by this application are the same as the beneficial effects of the autonomous driving behavior optimization method provided by the above embodiment, and the other technical features in the autonomous driving behavior optimization device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.
[0138] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0139] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0140] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the autonomous driving behavior optimization method in the above-mentioned embodiment.
[0141] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.
[0142] The above-mentioned computer-readable storage medium may be included in the autonomous driving behavior optimization device; or it may exist independently without being assembled into the autonomous driving behavior optimization device.
[0143] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the autonomous driving behavior optimization device, the autonomous driving behavior optimization device: obtains the self-vehicle driving perception data; performs parameter screening on the self-vehicle driving perception data according to the maximization of joint mutual information to obtain predicted characteristic parameters; performs preset model state observation according to the predicted characteristic parameters to obtain model parameters; performs regression analysis based on the model parameters and maximum likelihood estimation probability to obtain driving behavior; analyzes the driving behavior based on the self-vehicle perception data to obtain an analysis result, thereby realizing the analysis of the self-vehicle driving behavior.
[0144] Computer program code for performing the operations of the present application may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0145] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0146] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.
[0147] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned autonomous driving behavior optimization method, and can solve the technical problem that the current decision-making system based on rule definition needs to be constantly updated and maintained, and the rules in the system are also redundant, which affects the performance of autonomous driving. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the autonomous driving behavior optimization method provided by the above-mentioned embodiment, and will not be repeated here.
[0148] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned autonomous driving behavior optimization method.
[0149] The computer program product provided by this application can solve the technical problem that the current decision-making system based on rule definition needs to be constantly updated and maintained, and the rules in the system are also redundant, which affects the performance of autonomous driving. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the autonomous driving behavior optimization method provided by the above embodiment, which will not be repeated here.
[0150] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.
Claims
1. A method for optimizing autonomous driving behavior, characterized in that: The automatic driving behavior optimization method comprises: Acquire self-driving perception data; Selecting the initial input and output parameter set of the prediction model according to the functional finite state machine conditions and prior rules; Parameter screening is performed based on the self-driving perception data to obtain the input and output parameter sets; Based on the input and output parameter sets and the maximum likelihood estimation probability, a probability model regression analysis is performed to obtain the model parameters; The probability model is updated according to the model parameters to obtain a new probability model; Analyze driving behavior based on the self-vehicle perception data based on the new probability model to obtain the analysis results; The analysis result is compared with the result obtained based on the finite state machine condition to obtain a comparison result, and the autonomous driving behavior is optimized according to the comparison result.
2. The automatic driving behavior optimization method according to claim 1, characterized in that: The parameter screening based on the self-vehicle driving perception data to obtain the input and output parameter sets includes: Obtaining the number of target parameters, and obtaining a total feature parameter set according to the self-vehicle driving perception data; Screening the features in the total feature parameter set based on the target parameter quantity to obtain a target parameter set; A joint mutual information analysis is performed on the target parameter set and the pre-selected driving behavior parameters to obtain an input and output parameter set.
3. The automatic driving behavior optimization method according to claim 2, characterized in that: The joint mutual information analysis of the target parameter set and the pre-selected driving behavior parameters is performed to obtain an input and output parameter set, including: Calculating joint mutual information between each parameter in the target parameter set and the preselected driving behavior; When the mutual information meets the preset feature requirements, the parameters in the target parameter set whose mutual information meets the requirements are used as the input and output parameter sets.
4. The automatic driving behavior optimization method according to claim 1, characterized in that: The probability model regression analysis is performed based on the input and output parameter sets and the maximum likelihood estimation probability to obtain model parameters, including: Input the probability model according to the input and output parameter set to obtain initial observation values and initial actions; Obtaining a potential state based on the initial observation value and the initial action; The latent state is model trained according to the latent state expectation vector and the latent state covariance matrix to obtain model parameters.
5. The automatic driving behavior optimization method according to claim 1, characterized in that: The driving behavior is analyzed based on the new probability model according to the vehicle perception data to obtain the analysis results, including: Calculating maximum likelihood estimates of model parameters based on the new probability model according to the ego-vehicle perception data; Performing Gaussian mixture model regression analysis based on the maximum likelihood estimate to obtain a multimodal Gaussian distribution; The analytical results of the model parameters are determined according to the multimodal Gaussian distribution.
6. The automatic driving behavior optimization method according to claim 1, characterized in that: The step of comparing the analysis result with the result obtained based on the finite state machine condition to obtain a comparison result, and optimizing the autonomous driving behavior according to the comparison result includes: Obtaining a real driving behavior according to the self-vehicle perception data; Comparing the actual driving behavior with the driving behavior to obtain a comparison result; According to the comparison result, the self-vehicle perception data is reduced in redundancy to obtain the self-vehicle driving perception data after redundancy reduction; The automatic driving behavior is optimized according to the reduced-redundancy self-driving perception data.
7. The automatic driving behavior optimization method according to any one of claims 1 to 6, characterized in that: The obtaining of the self-vehicle driving perception data includes: Obtain driving information of the vehicle, surrounding intelligent agents and environment information; Slicing the self-vehicle driving information to obtain driving behavior and driving behavior habit parameters; Performing cluster analysis on the driving behavior and the driving behavior habit based on the surrounding intelligent agent information to obtain static parameters; The static parameters are divided into driving habits based on the environmental information to obtain self-vehicle driving perception data.
8. An automatic driving behavior optimization device, characterized in that: The automatic driving behavior optimization device comprises: A parameter acquisition module is used to obtain the driving perception data of the vehicle; A parameter screening module is used to select the initial input and output parameter set of the prediction model according to the functional finite state machine conditions and prior rules; The parameter screening module is also used to perform parameter screening based on the self-vehicle driving perception data to obtain an input and output parameter set; The parameter screening module is also used to perform probability model regression analysis based on the input and output parameter sets and the maximum likelihood estimation probability to obtain model parameters; A driving behavior optimization module, used for updating the probability model according to the model parameters to obtain a new probability model; The driving behavior optimization module is further used to analyze the driving behavior according to the vehicle perception data based on the new probability model to obtain the analysis result; The driving behavior optimization module is also used to compare the analysis result with the result obtained based on the finite state machine condition to obtain a comparison result, and optimize the automatic driving behavior according to the comparison result.
9. An automatic driving behavior optimization device, characterized in that: The device includes: a memory, a processor, and an autonomous driving behavior optimization program stored in the memory and executable on the processor, wherein the autonomous driving behavior optimization program is configured to implement the autonomous driving behavior optimization method as described in any one of claims 1 to 7.
10. A storage medium, characterized in that: The storage medium stores an autonomous driving behavior optimization program, and when the autonomous driving behavior optimization program is executed by the processor, the autonomous driving behavior optimization method according to any one of claims 1 to 7 is implemented.
Citation Information
Cited By
Driving interaction intention fragment extraction and processing method and device and storage medium
CN120354290A
