Quantitative analysis method and system for evaluating residual risk of planning algorithm

Abstract

The invention discloses a quantitative analysis method and system for evaluating the residual risk of a planning algorithm. The method comprises the steps that an automatic driving vehicle multi-mode data monitoring and collecting platform uploads original data to a vehicle-machine interaction control interface and forms a standard data set; in the vehicle-machine interaction control interface, a user inputs logic scene screening conditions, extracts data fragments of related business scenes from a data set and generalizes logic scene parameters to generate a test case meeting the mileage verification requirement. In the process, the system evaluates the occurrence frequency and the exposure degree of the scene in the real world through statistical analysis and probability density function fitting, and therefore it is guaranteed that the generated test case meets the occurrence probability requirement of the real world. According to the invention, a verification result for safety evaluation can be provided, and verification of a planning algorithm is supported. Through strict evaluation of various scenes, the system provides powerful support for the safety and reliability of the automatic driving technology.

Description

Technical Field

[0001] The present invention relates to the field of quantitative analysis of residual risks of vehicle automatic driving modes, and in particular to a quantitative analysis method and system for evaluating the residual risks of planning algorithms. Background Art

[0002] With the continuous development of autonomous driving technology, the safety and reliability of vehicles in complex traffic environments have become key issues that need to be addressed. Autonomous driving systems need to be tested extensively in diverse and complex traffic scenarios to ensure the effectiveness and safety of their planning algorithms under different road and traffic conditions. However, existing testing methods mainly rely on real road tests and partial simulation tests, which are not only costly and time-consuming, but also difficult to cover all possible traffic conditions and cannot fully evaluate the potential risks of autonomous driving systems, especially the residual risks related to the intended functional safety (SOT IF).

[0003] Currently, traffic scene simulation and data analysis tools mainly focus on two aspects: one is to create simulation scenes manually or semi-automatically, and the other is to perform simple statistical analysis on existing driving data. However, these two methods have obvious shortcomings in practical applications:

[0004] Manual or semi-automatic methods to create simulation scenarios: Manual methods to create simulation scenarios require manual configuration of various scenario parameters, which is time-consuming and easily affected by subjective factors, making it difficult to quickly generate representative and diverse traffic scenarios. Although the semi-automatic method introduces a certain degree of automation, it still relies heavily on preset templates and cannot generate complex and changing traffic scenarios in real time. In particular, for residual risks related to SOTIF, existing tools have difficulty generating scenarios that contain potential dangers and uncertainties, and cannot fully verify the safety of autonomous driving systems in abnormal or extreme situations.

[0005] Simple statistical analysis of driving data: Although existing data analysis tools can perform basic statistical analysis on driving data, such as the mean and distribution characteristics of the data, they lack the ability to deeply model and analyze the correlation between the data and actual driving scenarios. This limitation makes the generated simulation scenarios insufficient in diversity and representativeness, making it difficult to fully evaluate the residual risks of autonomous driving systems. The document "A Review of Technology for Generating Safety-Critical Scenario for Autonomous Driving" discusses existing methods for generating scenarios for autonomous driving, including data-driven methods and knowledge-driven methods. It points out that the current methods lack correlation with the actual traffic environment when generating scenarios, and cannot fully reflect the complex risk factors in real driving scenarios.

[0006] To solve the above problems, this application proposes a quantitative analysis method and system for evaluating the residual risk of planning algorithms. Summary of the invention

[0007] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a quantitative analysis method and system for evaluating the residual risk of a planning algorithm.

[0008] An embodiment of the present invention provides a quantitative analysis method for evaluating residual risk of a planning algorithm, which includes:

[0009] Step S100, a plurality of data acquisition sensors collect environmental data, vehicle status data, and motion data of the vehicle automatic driving mode at each unit time, and store and send the collected data to the vehicle-machine interactive control interface;

[0010] Step S200, the vehicle-machine interaction interface pre-processes the uploaded vehicle-side data to form a standard data set, and the user inputs the logical scenario screening conditions through the vehicle-machine interaction interface to extract data segments of relevant business scenarios from the data set;

[0011] Step S300: In the vehicle-machine interaction interface, the value distribution of the selected parameters is counted from the extracted scene data, a suitable probability density function is selected to fit the value distribution of the selected parameters, and the correlation between the parameters is analyzed to determine the independent variables and the dependent variables, and the conditional probability density function of the dependent variable is given, and the exposure degree of the logical scene in the data set is output;

[0012] Step S400: In the vehicle-machine interaction interface, the user generates test cases that meet the verification mileage requirements by generalizing the logic scenario parameters. In this process, the system evaluates the occurrence frequency and exposure of the scenario in the real world through statistical analysis and probability density function fitting to ensure that the generated test cases meet the occurrence probability requirements of the real world.

[0013] Step S500: In the vehicle-machine interaction interface, sampling is performed based on the scene data and scientific statistical data results, and the generalized file of the feature scene is automatically output in combination with the real parameters. The vehicle-mounted system can automatically generate a scene file that conforms to the parameter value distribution according to the required number of generalized scenes, and run it in the simulation software to enable vehicle-side ADAS testing;

[0014] Step S600, evaluating various indicators in the simulation test, including collision detection rate, reaction time, and risk assessment parameters, to determine whether these indicators are within the preset error tolerance threshold; if the indicator is not within the preset error tolerance threshold, generating a prompt signal for the vehicle automatic driving system algorithm to be optimized, repeating the simulation test after optimization until the indicator meets the requirements, and then feeding back the information to the vehicle automatic driving algorithm unit through the edge computing gateway to update the algorithm;

[0015] Step S700, risk quantification and assessment, runs the planning algorithm of the autonomous driving system in the generated simulation scenario, monitors and records the performance and response of the system, and evaluates the performance of the system in various test scenarios and its ability to handle complex and dangerous situations through quantitative methods to complete the quantification of residual risks.

[0016] Furthermore, residual risks and vehicle software and algorithm parameters include: environmental perception errors, algorithm limitations, potential risks, network security, etc., and relevant software and algorithm parameters cover sensor accuracy, algorithm real-time, computing platform performance, redundant design and network encryption mechanism.

[0017] Furthermore, the risk assessment parameters include: sensor operating efficiency, computing platform processing speed and reliability, redundant system design, communication delay, algorithm response time and software fault detection coverage.

[0018] Furthermore, it also includes: setting up a residual risk warning model analysis library in the vehicle-computer interactive control interface in advance, and storing information on residual risk and vehicle software and algorithm parameters, allowable error range of risk assessment parameters, and residual risk warning type in the residual risk warning model analysis library.

[0019] Furthermore, if the residual risk and the vehicle software and algorithm parameters and risk assessment parameters are not within the preset error tolerance threshold, a prompt signal for the vehicle's autonomous driving mode to be optimized is generated and fed back to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, including:

[0020] The vehicle-machine interactive control interface evaluates the received residual risk and vehicle software and algorithm parameters, risk assessment parameters and residual risk warning types in the residual risk warning model analysis library;

[0021] The vehicle-machine interactive control interface sends a corresponding vehicle automatic driving mode optimization prompt signal to the edge computing gateway based on the evaluation results.

[0022] Furthermore, if the residual risk and the vehicle software and algorithm parameters and the risk assessment parameters are not within the preset error tolerance threshold, a prompt signal for the vehicle's automatic driving mode to be optimized is generated and fed back to the vehicle's automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, which also includes:

[0023] The vehicle-machine interactive control interface sends a software program run / interrupt or algorithm run / interrupt signal to the vehicle's automatic driving mode mainboard based on the evaluation results to control the running steps of the vehicle's automatic driving mode per unit time.

[0024] Furthermore, the edge computing gateway feeds back the residual risk monitoring point calibration quantitative analysis algorithm unit of the vehicle's autonomous driving mode through the edge computing gateway, including: the edge computing gateway feeds back the vehicle's autonomous driving mode optimization prompt signal to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through special coding.

[0025] The embodiment of the present invention further provides a quantitative analysis system for evaluating the residual risk of a planning algorithm, comprising a vehicle automatic driving mode control server, a plurality of vehicle automatic driving mode residual risk monitoring points, an edge computing gateway, and a vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit, wherein a vehicle-machine interactive control interface is provided on the vehicle automatic driving mode control server, and a vehicle automatic driving mode multimodal data monitoring and acquisition platform is provided on the plurality of vehicle automatic driving mode residual risk monitoring points;

[0026] The vehicle-machine interactive control interface is used to send a residual risk intelligent quantitative analysis signal to the vehicle automatic driving mode multimodal data monitoring and acquisition platform;

[0027] The vehicle automatic driving mode multimodal data monitoring and acquisition platform is used to receive the residual risk intelligent quantitative analysis signal, and multiple data acquisition sensors collect the residual risk of the vehicle automatic driving mode per unit time, the vehicle software and algorithm parameters, and the risk assessment parameters of the working state of the vehicle automatic driving mode per unit time, and store and send the collected data to the vehicle-machine interactive control interface;

[0028] The vehicle-machine interactive control interface is also used to determine whether the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are within the preset error tolerance threshold; if the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are not within the preset error tolerance threshold, a vehicle automatic driving mode optimization prompt signal is generated and fed back to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway.

[0029] Furthermore, the edge computing gateway is used to feed back the vehicle automatic driving mode optimization prompt signal to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through special coding.

[0030] Furthermore, the vehicle-machine interactive control interface is used to evaluate the received residual risks and vehicle software and algorithm parameters, risk assessment parameters and residual risk warning types in the residual risk warning model analysis library; and based on the evaluation results, send a corresponding vehicle automatic driving mode optimization prompt signal to the edge computing gateway.

[0031] Beneficial effects:

[0032] The present invention proposes a quantitative analysis method and system for evaluating the residual risk of a planning algorithm. The method realizes a comprehensive quantitative analysis of various risk factors in traffic scenarios through automated data extraction, analysis and scenario generation, especially residual risks related to expected functional safety. By extracting key logical scenario data from real traffic flow data, performing in-depth data analysis, and generating a variety of simulation scenarios, a scientific and accurate residual risk assessment and verification method is finally provided for the planning algorithm of the autonomous driving system. The method provided by the embodiment of the present invention can automatically perform inspections and can timely issue a prompt signal that the vehicle's autonomous driving mode is to be optimized, thereby improving the efficiency of quantitative analysis and reducing costs. In addition, the quantitative analysis method for evaluating the residual risk of a planning algorithm provided by the embodiment of the present invention is mainly executed on the vehicle-machine interactive control interface and the vehicle's autonomous driving mode multimodal data monitoring and acquisition platform, and has a fast response speed. The present invention can provide verification results for safety evaluation and support the verification of planning algorithms. Through rigorous evaluation of various scenarios, the system provides strong support for the safety and reliability of autonomous driving technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0034] Figure 1 A schematic flow chart of a quantitative analysis method for evaluating residual risk of a planning algorithm provided by an embodiment of the present invention;

[0035] Figure 2 A structural diagram of a quantitative analysis system for evaluating the residual risk of a planning algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other. The present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0037] See also Figure 1 , a flowchart of a quantitative analysis method for evaluating residual risk of a planning algorithm provided by an embodiment of the present invention, the method comprising:

[0038] Step S100, a plurality of data acquisition sensors collect environmental data, vehicle status data, and motion data of the vehicle automatic driving mode at each unit time, and store and send the collected data to the vehicle-machine interactive control interface;

[0039] Step S200, the vehicle-machine interaction interface pre-processes the uploaded vehicle-side data to form a standard data set, and the user inputs the logical scenario screening conditions through the vehicle-machine interaction interface to extract data segments of relevant business scenarios from the data set;

[0040] Step S300: In the vehicle-machine interaction interface, the value distribution of the selected parameters is counted from the extracted scene data, a suitable probability density function is selected to fit the value distribution of the selected parameters, and the correlation between the parameters is analyzed to determine the independent variables and the dependent variables, and the conditional probability density function of the dependent variable is given, and the exposure degree of the logical scene in the data set is output;

[0041] Step S400: In the vehicle-machine interaction interface, the user generates test cases that meet the verification mileage requirements by generalizing the logic scenario parameters. In this process, the system evaluates the occurrence frequency and exposure of the scenario in the real world through statistical analysis and probability density function fitting to ensure that the generated test cases meet the occurrence probability requirements of the real world.

[0042] Step S500: In the vehicle-machine interaction interface, sampling is performed based on the scene data and scientific statistical data results, and the generalized file of the feature scene is automatically output in combination with the real parameters. The vehicle-mounted system can automatically generate a scene file that conforms to the parameter value distribution according to the required number of generalized scenes, and run it in the simulation software to enable vehicle-side ADAS testing;

[0043] Step S600, evaluating various indicators in the simulation test, including collision detection rate, reaction time, and risk assessment parameters, to determine whether these indicators are within the preset error tolerance threshold; if the indicator is not within the preset error tolerance threshold, generating a prompt signal for the vehicle automatic driving system algorithm to be optimized, repeating the simulation test after optimization until the indicator meets the requirements, and then feeding back the information to the vehicle automatic driving algorithm unit through the edge computing gateway to update the algorithm;

[0044] Step S700, risk quantification and assessment, runs the planning algorithm of the autonomous driving system in the generated simulation scenario, monitors and records the performance and response of the system, and evaluates the performance of the system in various test scenarios and its ability to handle complex and dangerous situations through quantitative methods to complete the quantification of residual risks.

[0045] In the embodiment of the present invention, the vehicle-machine interactive control interface can send a residual risk intelligent quantitative analysis signal to the vehicle automatic driving mode multimodal data monitoring and acquisition platform, and the vehicle automatic driving mode multimodal data monitoring and acquisition platform can control the acquisition of residual risk and vehicle software and algorithm parameters, risk assessment parameters according to the residual risk intelligent quantitative analysis signal and feedback to the vehicle-machine interactive control interface, so that the vehicle-machine interactive control interface can determine whether the residual risk and vehicle software and algorithm parameters, risk assessment parameters are within the error allowable threshold. If not, it automatically generates a vehicle automatic driving mode to be optimized prompt signal, and feeds it back to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, so that the management personnel can timely understand the residual risk warning information of the vehicle automatic driving mode per unit time, and timely process the residual risk warning to prevent the damage of the vehicle automatic driving mode per unit time. The method provided in the embodiment of the present invention can automatically perform inspections, and can timely send out a vehicle automatic driving mode to be optimized prompt signal, improve the efficiency of quantitative analysis, and reduce costs. In addition, the quantitative analysis method for evaluating the residual risk of the planning algorithm provided in the embodiment of the present invention is mainly executed on the vehicle-machine interactive control interface and the vehicle automatic driving mode multimodal data monitoring and acquisition platform, and has a fast response speed.

[0046] Specifically, the vehicle-machine interactive control interface can send the residual risk intelligent quantitative analysis signal to the vehicle's automatic driving mode multimodal data monitoring platform in different ways, such as sending the residual risk intelligent quantitative analysis signal in a timed manner, that is, sending the residual risk intelligent quantitative analysis signal once every same period of time, which can be the device default duration or a custom duration.

[0047] In addition, the residual risk intelligent quantitative analysis signal can be sent irregularly, that is, the residual risk intelligent quantitative analysis signal is sent every different period of time. In this way, it can be manually started by the management personnel, that is, the management personnel can evaluate whether it is necessary to send the residual risk intelligent quantitative analysis signal based on their own experience or actual working status. It is also possible to divide each cycle into several different time periods in advance, and set a time interval for different time periods, so that in a time period, the residual risk intelligent quantitative analysis signal is sent regularly according to the corresponding time interval. This method is actually a scheduled inspection, but the time interval is different in different time periods. The advantage of setting different time periods is that the detection frequency can be set in a targeted manner.

[0048] The multiple data acquisition sensors are started, and according to the type of each sensor in the sensor group, the residual risk of the vehicle's automatic driving mode per unit time, vehicle software and algorithm parameters, and risk assessment parameters of the vehicle's automatic driving mode working status per unit time are collected.

[0049] Among them, residual risks and vehicle software and algorithm parameters include: environmental perception errors, algorithm limitations, potential risks, network security, etc., and related software and algorithm parameters cover sensor accuracy, algorithm real-time performance, computing platform performance, redundancy design and network encryption mechanism. The risk assessment parameters include: sensor efficiency, computing platform processing speed and reliability, redundant system design, communication delay, algorithm response time, and software fault detection coverage.

[0050] The above data can be collected by corresponding sensors. The combination of sensors can be collectively referred to as a sensor group. A sensor group can be set for each unit time of the vehicle's automatic driving mode to collect the residual risk of the vehicle's automatic driving mode per unit time, the vehicle software and algorithm parameters, and the risk assessment parameters of the working state of the vehicle's automatic driving mode per unit time, thereby realizing the remote warning function of the vehicle's automatic driving mode per unit time.

[0051] The above-mentioned residual risk and vehicle software and algorithm parameters and risk assessment parameters have different functions. For example, the residual risk and vehicle software and algorithm parameters can directly reflect the current state of the vehicle's automatic driving mode per unit time, and can determine whether the vehicle's automatic driving mode per unit time has a residual risk warning. The risk assessment parameters of the working state of the vehicle's automatic driving mode per unit time can directly reflect the current working state of the vehicle's automatic driving mode per unit time, such as whether the working state of the vehicle's automatic driving mode per unit time has undergone a major change, or whether the working state has deteriorated, etc. In this way, it can be expected whether the vehicle's automatic driving mode per unit time will have a residual risk warning, so that it can be handled in a timely manner.

[0052] The vehicle's autonomous driving mode multimodal data monitoring and acquisition platform can store the data collected by the sensor group and eventually send it to the vehicle-machine interactive control interface.

[0053] After receiving the residual risk, vehicle software and algorithm parameters, and risk assessment parameters, the vehicle-machine interactive control interface can determine whether the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are within a preset error tolerance threshold.

[0054] If the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are not within the preset error tolerance threshold, that is, the residual risk, vehicle software and algorithm parameters are not within the preset error tolerance threshold, then it is necessary to generate a vehicle automatic driving mode optimization prompt signal and feed it back to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway.

[0055] The vehicle automatic driving mode residual risk monitoring point calibration and quantitative analysis algorithm unit can be the vehicle automatic driving mode residual risk monitoring point calibration and quantitative analysis algorithm unit of the maintenance personnel, so that the maintenance personnel can promptly handle the residual risk warning that is occurring or is about to occur in the vehicle automatic driving mode per unit time. The vehicle automatic driving mode residual risk monitoring point calibration and quantitative analysis algorithm unit can also be the vehicle automatic driving mode residual risk monitoring point calibration and quantitative analysis algorithm unit of the management personnel, so that the management personnel can arrange personnel to handle it according to the area to which the vehicle automatic driving mode belongs per unit time.

[0056] Calibration and quantitative analysis algorithms for residual risk monitoring points of vehicle automated driving modes are often used to ensure that the geometric characteristics of the vehicle automated driving mode or physical prototype meet the design requirements during the manufacturing process. The main goal of such algorithms is to calibrate the measurement data, correct errors and perform accurate residual risk assessment.

[0057] In order to better analyze the residual risk warning of the vehicle's automatic driving mode per unit time, the embodiment of the present invention preferably simultaneously collects the residual risk of the vehicle's automatic driving mode per unit time and the risk assessment parameters of the vehicle's software and algorithm parameters and the working state of the vehicle's automatic driving mode per unit time, and analyzes the residual risk warning based on these two types of information. When the residual risk and the vehicle's software and algorithm parameters and the risk assessment parameters are all within the preset error tolerance threshold, a prompt signal for the vehicle's automatic driving mode to be optimized is not generated. When any of the residual risk and the vehicle's software and algorithm parameters and the risk assessment parameters are not within the preset error tolerance threshold, a prompt signal for the vehicle's automatic driving mode to be optimized is generated. This ensures that the residual risk warning is discovered in a timely manner and that the residual risk warning is processed.

[0058] In one embodiment, the quantitative analysis method for evaluating the residual risk of a planning algorithm further includes:

[0059] The residual risk warning model analysis library is set in advance in the vehicle-machine interactive control interface, and the residual risk and vehicle software and algorithm parameters, the error tolerance range of risk assessment parameters, and the residual risk warning type are stored in the residual risk warning model database.

[0060] In this embodiment, the residual risk and the allowable error range of the vehicle software and algorithm parameters, and the risk assessment parameters can be stored in advance in the residual risk warning model analysis library. In this way, the vehicle-machine interactive control interface can call up the corresponding allowable error range at any time and perform an evaluation when performing residual risk warning analysis. At the same time, the information on the residual risk warning type can also be stored in the residual risk warning model analysis library. In this way, the type of residual risk warning can be determined, which is convenient for maintenance personnel to perform residual risk warning processing.

[0061] In one embodiment, if the residual risk and the vehicle software and algorithm parameters and the risk assessment parameters are not within the preset error tolerance threshold, a vehicle automatic driving mode optimization prompt signal is generated and fed back to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, including:

[0062] The vehicle-machine interactive control interface evaluates the received residual risk and vehicle software and algorithm parameters, risk assessment parameters and residual risk warning types in the residual risk warning model analysis library;

[0063] The vehicle-machine interactive control interface sends a corresponding vehicle automatic driving mode optimization prompt signal to the edge computing gateway based on the evaluation results.

[0064] After determining that the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are not within the preset error tolerance range, the residual risk, vehicle software and algorithm parameters, risk assessment parameters, and residual risk warning types in the residual risk warning model analysis library can be evaluated. Then, the vehicle automatic driving mode optimization prompt signal is sent to the edge computing gateway, and the edge computing gateway is used to calibrate the quantitative analysis algorithm unit of the residual risk monitoring point of the vehicle automatic driving mode. In this way, not only can the residual risk warning of the vehicle automatic driving mode per unit time be obtained, but also the residual risk warning of the vehicle automatic driving mode per unit time can be obtained, so as to facilitate targeted processing.

[0065] In one embodiment, the feedback to the residual risk monitoring point calibration quantitative analysis algorithm unit of the vehicle autonomous driving mode through the edge computing gateway includes:

[0066] The edge computing gateway feeds back the vehicle automatic driving mode optimization prompt signal to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through a special coding method.

[0067] The edge computing gateway can feed back a prompt signal for optimization of the vehicle's autonomous driving mode to the residual risk monitoring point calibration quantitative analysis algorithm unit of the vehicle's autonomous driving mode by email or a signal pop-up window. Of course, other methods can also be used for early warning, such as WeChat, mini-programs, dedicated apps, etc.

[0068] In one embodiment, if the residual risk and the vehicle software and algorithm parameters and the risk assessment parameters are not within the preset error tolerance threshold, a vehicle automatic driving mode optimization prompt signal is generated and fed back to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, and further includes:

[0069] The vehicle-machine interactive control interface sends a software program run / interrupt or algorithm run / interrupt signal to the vehicle's automatic driving mode mainboard based on the evaluation results to control the running steps of the vehicle's automatic driving mode per unit time.

[0070] In this embodiment, when it is determined that the residual risk and the vehicle software and algorithm parameters and risk assessment parameters are not within the preset error tolerance threshold and a vehicle automatic driving mode optimization prompt signal is generated and sent to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit, the maintenance personnel can handle it in time. In the embodiment of the present invention, the vehicle-machine interactive control interface can also send a software program running / interrupt signal or an algorithm running / interrupt signal to the vehicle automatic driving mode mainboard to control the running steps of the vehicle automatic driving mode per unit time. For example, when it is detected that the vehicle software running failure of the vehicle automatic driving mode working state per unit time, a software program running / interrupt signal can be sent to the relay in the vehicle automatic driving mode mainboard, or an algorithm running / interrupt signal can be sent to the relay in the vehicle automatic driving mode mainboard to transmit power to the air conditioner.

[0071] In other embodiments, a neural network model may also be used to predict the collected data, so as to obtain early warning information of impending residual risks, which is helpful in avoiding risks such as burnout of the vehicle's automatic driving mode per unit time.

[0072] Generally speaking, if a certain state of the vehicle automatic driving mode per unit time continues to change or certain parameters in the working state continue to deviate from normal values ​​during use, then the risk of damage to the vehicle automatic driving mode per unit time is relatively high. Therefore, the embodiment of the present invention can collect historical residual risk warning data, and the historical residual risk warning data includes the type of residual risk warning and the residual risk before the residual risk warning occurs, and the vehicle software and algorithm parameters and risk assessment parameters.

[0073] Then, the above historical residual risk warning data is preprocessed, and the preprocessing content may include normalization, etc., so that each type of data is classified into the same dimension, and then the historical residual risk warning data is divided into a sample set and a test set. Then, the neural network model is used to train the sample set of the historical residual risk warning data, and then the trained neural network model is tested using the test set, and finally the preset accuracy requirements are met.

[0074] Finally, the collected data can be input into the neural network model, and then the neural network model is used for prediction to obtain the output value, which can be the probability and type of residual risk warning, so that the maintenance personnel can quickly process according to the output result, avoid causing actual impact, and solve the problem in advance. The neural network model can adopt a differential back propagation neural network, which is a multi-layer neural network with three or more layers, each layer consisting of a number of neurons. Each neuron between adjacent layers is fully connected, that is, each neuron in the left layer is connected to each neuron in the right layer, and there is no connection between the upper and lower neurons. The neural network model can be trained in a supervised learning manner. When a pair of learning patterns are provided to the network, the activation value of its neurons will propagate from the input layer through each hidden layer to the output layer, and each neuron in the output layer outputs a network response corresponding to the input pattern. Then, according to the principle of reducing the error between the desired output and the actual output, the connection weights are corrected layer by layer from the output layer through each hidden layer and finally back to the input layer. The embodiment of the present invention adopts the above-mentioned prediction method to improve the accuracy, which is particularly suitable for the field of residual risk warning prediction of vehicle automatic driving mode per unit time in the embodiment of the present invention.

[0075] See also Figure 2 , an embodiment of the present invention further provides a quantitative analysis system for evaluating the residual risk of a planning algorithm, which includes a vehicle automatic driving mode control server, a plurality of vehicle automatic driving mode residual risk monitoring points, an edge computing gateway, and a vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit, the vehicle automatic driving mode control server is provided with a vehicle-machine interactive control interface, and the plurality of vehicle automatic driving mode residual risk monitoring points are all provided with a vehicle automatic driving mode multimodal data monitoring and acquisition platform;

[0076] The vehicle-machine interactive control interface is used to send a residual risk intelligent quantitative analysis signal to the vehicle automatic driving mode multimodal data monitoring and acquisition platform;

[0077] The vehicle automatic driving mode multimodal data monitoring and acquisition platform is used to receive the residual risk intelligent quantitative analysis signal, and multiple data acquisition sensors collect the residual risk of the vehicle automatic driving mode per unit time, the vehicle software and algorithm parameters, and the risk assessment parameters of the working state of the vehicle automatic driving mode per unit time, and store and send the collected data to the vehicle-machine interactive control interface;

[0078] The vehicle-machine interactive control interface is also used to determine whether the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are within the preset error tolerance threshold; if the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are not within the preset error tolerance threshold, a vehicle automatic driving mode optimization prompt signal is generated and fed back to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway.

[0079] In one embodiment, the edge computing gateway is used to feed back the vehicle automatic driving mode optimization prompt signal to the vehicle automatic driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through a special encoding method.

[0080] In one embodiment, the vehicle-machine interactive control interface is used to evaluate the received residual risks and vehicle software and algorithm parameters, risk assessment parameters and residual risk warning types in the residual risk warning model analysis library; and based on the evaluation results, send a corresponding vehicle automatic driving mode optimization prompt signal to the edge computing gateway.

[0081] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection threshold of the claims of the present invention.

Claims

1. A quantitative analysis method for evaluating the residual risk of a planning algorithm, characterized in that: The method includes: Step S100, a plurality of data acquisition sensors collect environmental data, vehicle status data, and motion data of the vehicle automatic driving mode at each unit time, and store and send the collected data to the vehicle-machine interactive control interface; Step S200, the vehicle-machine interaction interface pre-processes the uploaded vehicle-side data to form a standard data set, and the user inputs the logical scenario screening conditions through the vehicle-machine interaction interface to extract data segments of relevant business scenarios from the data set; Step S300: In the vehicle-machine interaction interface, the value distribution of the selected parameters is counted from the extracted scene data, a suitable probability density function is selected to fit the value distribution of the selected parameters, and the correlation between the parameters is analyzed to determine the independent variables and the dependent variables, and the conditional probability density function of the dependent variable is given, and the exposure degree of the logical scene in the data set is output; Step S400: In the vehicle-machine interaction interface, the user generates test cases that meet the verification mileage requirements by generalizing the logic scenario parameters. In this process, the system evaluates the occurrence frequency and exposure of the scenario in the real world through statistical analysis and probability density function fitting to ensure that the generated test cases meet the occurrence probability requirements of the real world. Step S500: In the vehicle-machine interaction interface, sampling is performed based on the scene data and scientific statistical data results, and the generalized file of the feature scene is automatically output in combination with the real parameters. The vehicle-mounted system can automatically generate a scene file that conforms to the parameter value distribution according to the required number of generalized scenes, and run it in the simulation software to enable vehicle-side ADAS testing; Step S600, evaluating various indicators in the simulation test, including collision detection rate, reaction time, and risk assessment parameters, to determine whether these indicators are within the preset error tolerance threshold; if the indicator is not within the preset error tolerance threshold, generating a prompt signal for the vehicle automatic driving system algorithm to be optimized, repeating the simulation test after optimization until the indicator meets the requirements, and then feeding back the information to the vehicle automatic driving algorithm unit through the edge computing gateway to update the algorithm; Step S700, risk quantification and assessment, runs the planning algorithm of the autonomous driving system in the generated simulation scenario, monitors and records the performance and response of the system, and evaluates the performance of the system in various test scenarios and its ability to handle complex and dangerous situations through quantitative methods to complete the quantification of residual risks.

2. A quantitative analysis method for evaluating residual risk of a planning algorithm according to claim 1, characterized in that: Residual risks and vehicle software and algorithm parameters include: environmental perception errors, algorithm limitations, potential risks, network security, and related software and algorithm parameters cover sensor accuracy, algorithm real-time, computing platform performance, redundant design, and network encryption mechanisms.

3. A quantitative analysis method for evaluating residual risk of a planning algorithm according to claim 1, characterized in that: The risk assessment parameters include: sensor working efficiency, computing platform processing speed and reliability, redundant system design, communication delay, algorithm response time and software fault detection coverage.

4. A quantitative analysis method for evaluating residual risk of a planning algorithm according to claim 1, characterized in that: Also includes: A residual risk warning model analysis library is set up in advance in the vehicle-machine interactive control interface, and information on residual risk, vehicle software and algorithm parameters, allowable error range of risk assessment parameters, and residual risk warning type is stored in the residual risk warning model analysis library.

5. A quantitative analysis method for evaluating residual risk of a planning algorithm according to claim 4, characterized in that: If the residual risk and vehicle software and algorithm parameters and risk assessment parameters are not within the preset error tolerance threshold, a prompt signal for the vehicle's autonomous driving mode to be optimized is generated and fed back to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, including: The vehicle-machine interactive control interface evaluates the received residual risk and vehicle software and algorithm parameters, risk assessment parameters and residual risk warning types in the residual risk warning model analysis library; Based on the evaluation results, the vehicle-machine interactive control interface sends a corresponding vehicle autonomous driving mode optimization prompt signal to the edge computing gateway.

6. A quantitative analysis method for evaluating residual risk of a planning algorithm according to claim 5, characterized in that: If the residual risk and the vehicle software and algorithm parameters and risk assessment parameters are not within the preset error tolerance threshold, a prompt signal for the vehicle's autonomous driving mode to be optimized is generated and fed back to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway, which also includes: The vehicle-computer interaction control interface sends a software program run / interrupt or algorithm run / interrupt signal to the vehicle's automatic driving mode mainboard based on the evaluation results to control the running steps of the vehicle's automatic driving mode per unit time.

7. A quantitative analysis method for evaluating residual risk of a planning algorithm according to claim 1, characterized in that: Feedback to the residual risk monitoring point of the vehicle's autonomous driving mode through the edge computing gateway to calibrate the quantitative analysis algorithm unit, including: The edge computing gateway feeds back the vehicle's autonomous driving mode optimization prompt signal to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through special coding.

8. A quantitative analysis system for evaluating the residual risk of a planning algorithm, characterized in that: It includes a vehicle automatic driving mode control server, multiple vehicle automatic driving mode residual risk monitoring points, an edge computing gateway and a calibration and quantitative analysis algorithm unit for the vehicle automatic driving mode residual risk monitoring points. The vehicle automatic driving mode control server is provided with a vehicle-machine interactive control interface, and multiple vehicle automatic driving mode residual risk monitoring points are provided with a vehicle automatic driving mode multimodal data monitoring and acquisition platform; The vehicle-machine interactive control interface is used to send a residual risk intelligent quantitative analysis signal to the vehicle automatic driving mode multimodal data monitoring and acquisition platform; The vehicle automatic driving mode multimodal data monitoring and acquisition platform is used to receive the residual risk intelligent quantitative analysis signal, and multiple data acquisition sensors collect the residual risk of the vehicle automatic driving mode per unit time, the vehicle software and algorithm parameters, and the risk assessment parameters of the working state of the vehicle automatic driving mode per unit time, and store and send the collected data to the vehicle-machine interactive control interface; The vehicle-computer interactive control interface is also used to determine whether the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are within the preset error tolerance threshold; if the residual risk, vehicle software and algorithm parameters, and risk assessment parameters are not within the preset error tolerance threshold, a prompt signal for the vehicle's autonomous driving mode to be optimized is generated and fed back to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through the edge computing gateway.

9. A quantitative analysis system for evaluating residual risk of a planning algorithm according to claim 8, characterized in that: The edge computing gateway is used to feed back the vehicle's autonomous driving mode optimization prompt signal to the vehicle's autonomous driving mode residual risk monitoring point calibration quantitative analysis algorithm unit through special coding.

10. A quantitative analysis system for evaluating residual risk of a planning algorithm according to claim 8, characterized in that: The vehicle-machine interactive control interface is used to evaluate the received residual risk and vehicle software and algorithm parameters, risk assessment parameters and residual risk warning types in the residual risk warning model analysis library; And according to the evaluation results, the corresponding vehicle autonomous driving mode optimization prompt signal is sent to the edge computing gateway.

Images