Accident cause determination method, device, storage medium and computer equipment

Abstract

The present application provides a method, apparatus, storage medium, and computer device for determining the cause of an accident. The method includes: in the event of an accident involving an autonomous vehicle, using hardware device detection data and software module result data corresponding to the accident investigation period as accident data; for each investigation moment in the accident investigation period, determining the hardware fault and software fault existing in the autonomous vehicle at that investigation moment based on the hardware device detection data and software module result data, respectively, to obtain hardware fault information and software fault information corresponding to that investigation moment; determining the critical moment of the accident from each investigation moment based on each investigation moment, the hardware fault information corresponding to each investigation moment, and the software fault information corresponding to each investigation moment; and using the critical moment of the accident to locate the time when the fault causing the accident occurred. The method of the present application can improve the efficiency of analyzing the cause of the accident.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a method, apparatus, storage medium, and computer equipment for determining the cause of an accident. Background Art

[0002] An autonomous vehicle is a vehicle controlled by an autonomous driving algorithm. When operating on real roads, an autonomous vehicle may be subject to accidents such as collisions due to algorithmic and / or hardware failures. In the event of an accident, to facilitate subsequent optimization of the autonomous driving algorithm and / or vehicle hardware configuration, it is necessary to analyze the cause of the accident to determine the specific fault or combination of faults that caused the accident. The fault or combination of faults that caused the accident is considered the causal fault.

[0003] During actual driving, autonomous vehicles may experience different faults or combinations of faults at different times. Furthermore, the moment an accident occurs can lag behind the occurrence of the underlying fault. Therefore, it is necessary to analyze the autonomous vehicle's operating data for a period of time before the accident to determine the underlying fault and subsequently complete the accident cause analysis. However, the inventors have discovered that existing technologies require a significant amount of time to perform accident cause analysis. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above technical deficiencies, especially the technical defect that the prior art requires a large amount of time to analyze the cause of the accident.

[0005] In a first aspect, an embodiment of the present application provides a method for determining the cause of an accident, the method comprising:

[0006] In the event of an accident involving an autonomous vehicle, the hardware device detection data and software module result data corresponding to the accident investigation period are used as accident data; the end time of the accident investigation period is the time of occurrence of the accident;

[0007] For each troubleshooting moment in the accident troubleshooting period, determining a hardware fault existing in the autonomous driving vehicle at the troubleshooting moment based on the hardware device detection data, thereby obtaining hardware fault information corresponding to the troubleshooting moment; and determining a software fault existing in the autonomous driving vehicle at the troubleshooting moment based on the software module result data, thereby obtaining software fault information corresponding to the troubleshooting moment;

[0008] Based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment, the critical moment of the accident is determined from each troubleshooting moment; the critical moment of the accident is used to locate the time when the cause of the fault that caused the accident occurred.

[0009] In a second aspect, an embodiment of the present application provides a device for determining the cause of an accident, the device comprising:

[0010] An accident data acquisition module is used to, in the event of an accident involving an autonomous vehicle, use hardware device detection data and software module result data corresponding to an accident investigation period as accident data; the end time of the accident investigation period is the time of occurrence of the accident;

[0011] a fault information acquisition module configured to, for each troubleshooting moment during the accident troubleshooting period, determine, based on the hardware device detection data, a hardware fault existing in the autonomous driving vehicle at that troubleshooting moment, thereby obtaining hardware fault information corresponding to that troubleshooting moment; and determine, based on the software module result data, a software fault existing in the autonomous driving vehicle at that troubleshooting moment, thereby obtaining software fault information corresponding to that troubleshooting moment;

[0012] The accident critical moment determination module is used to determine the accident critical moment from each troubleshooting moment based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment; the accident critical moment is used to locate the time when the cause of the fault that caused the accident occurred.

[0013] In a third aspect, an embodiment of the present application provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the accident cause determination method described in any of the above embodiments.

[0014] In a fourth aspect, an embodiment of the present application provides a computer device, comprising: one or more processors, and a memory;

[0015] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the accident cause determination method described in any of the above embodiments are executed.

[0016] Embodiments of the present application provide a method, apparatus, storage medium, and computer device for determining the cause of an accident. In the event of an accident involving an autonomous vehicle, the moment of the accident is used as the end time of the accident investigation period, and the hardware device detection data and software module result data corresponding to the accident investigation period are used as accident data. The hardware device detection data is used to determine the hardware faults present in the autonomous vehicle at each investigation time during the accident investigation period, thereby obtaining hardware fault information for each investigation time. Furthermore, the software module result data is used to determine the software faults present in the autonomous vehicle at each investigation time during the accident investigation period, thereby obtaining software fault information for each investigation time. Based on each investigation time during the accident investigation period, the hardware fault information corresponding to each investigation time, and the software fault information corresponding to each investigation time, the critical moment of the accident is determined from each investigation time. In this manner, the critical moment of the accident can be quickly determined in an automated manner. Because the critical moment of the accident can be used to locate the time when the cause fault occurred, when analyzing the cause of the accident based on the critical moment of the accident, the scope of investigation of the cause fault can be narrowed, thereby shortening the time required for analyzing the cause of the accident and improving analysis efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 This is a flow chart of a method for determining the cause of an accident in one embodiment;

[0019] Figure 2 A schematic diagram of a flow chart of steps for determining a critical moment of an accident in one embodiment;

[0020] Figure 3 This is a flowchart of the steps of obtaining hardware fault information in one embodiment;

[0021] Figure 4 This is a second flow chart of the steps of obtaining hardware fault information in one embodiment;

[0022] Figure 5 This is a third flowchart of the step of obtaining hardware fault information in one embodiment;

[0023] Figure 6 A schematic diagram of a cut-in scene in one embodiment;

[0024] Figure 7This is a second flow chart of a method for determining the cause of an accident in an embodiment.

[0025] Figure 8 is a schematic structural block diagram of an accident cause determination device in one embodiment;

[0026] Figure 9 Schematic diagram of the structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0028] As mentioned in the background art, the existing technology takes a lot of time to obtain the cause of the fault. The inventors found that the reason for this problem is that the existing technology uses manual methods to analyze the huge amount of autonomous driving log data, and engineers need to spend a lot of time to complete the analysis and then obtain the cause of the fault that caused the accident. At the same time, since the autonomous driving system involves multiple software and hardware modules, for example, the software modules involved include positioning algorithm models, perception algorithm models, prediction algorithm models and planning algorithm models, and the hardware modules involved include lidar, camera equipment and GPS (Global Positioning System) modules, etc., there is a complex coupling relationship between the various software and hardware modules. Therefore, it is also necessary to combine multiple engineers responsible for different modules to conduct manual analysis before the cause of the fault can be obtained, which further aggravates the problem of long time consumption.

[0029] To address the above-mentioned issues, embodiments of the present application provide a method, apparatus, storage medium, and computer device for determining the cause of an accident, which can quickly and automatically determine the critical moment of an accident. Because the critical moment of an accident can be used to locate the moment when the cause of the fault occurs, when analyzing the cause of an accident based on the critical moment of the accident, the scope of troubleshooting the cause of the fault can be narrowed, allowing for rapid location of the cause of the accident and a list of software and hardware modules corresponding to the cause of the fault to be determined, helping engineers locate and repair the problem. This can shorten the time required to analyze the cause of the accident and improve analysis efficiency.

[0030] In one embodiment, the present application provides a method for determining the cause of an accident. This method can be applied to a computer device with data processing capabilities, which can be, but is not limited to, a server or a terminal. The following embodiment uses the method applied to a cloud server as an example. Specifically, the method may include the following steps:

[0031] S110, in the event of an accident involving an autonomous driving vehicle, the hardware equipment detection data and software module result data corresponding to the accident investigation period are used as accident data; the end time of the accident investigation period is the time when the accident occurred.

[0032] The accident may be, but is not limited to, a collision accident, a non-collision traffic accident, or an accident where the vehicle misses a fork in the road. In the event of an accident involving an autonomous vehicle, considering that the occurrence of the accident may lag behind the occurrence of the cause of the fault, it is necessary to troubleshoot the faults of the autonomous vehicle for a period of time before the accident. Before the troubleshooting, an accident troubleshooting period may be determined. The accident troubleshooting period starts at a certain time before the accident and ends at the time of the accident. It is understood that the starting time of the accident troubleshooting period may be determined based on actual circumstances, and this application does not impose specific restrictions on this. In one embodiment, the starting time may be determined based on a pre-set time period and the time of the accident, thereby determining the accident troubleshooting period. In one example, the pre-set time period may be half a minute or one minute. In this example, if the time of the accident is T, the accident troubleshooting period may be [T-30s, T] or [T-60s, T].

[0033] The cloud server may use the hardware device detection data and software module result data corresponding to the accident investigation period as accident data. The hardware device detection data may be data detected or collected by the autonomous vehicle's hardware devices, and the software module result data may be intermediate data or result data generated and / or output by the autonomous vehicle's algorithm module.

[0034] During the driving process of an autonomous vehicle, each hardware device and each algorithm module in the autonomous vehicle will record its input and output data to obtain hardware device detection data and software module result data, respectively. For example, the LiDAR, camera, and GPS device will all record the data collected for each frame. The positioning algorithm model, perception algorithm model, prediction algorithm model, and planning algorithm model will all record the data output by the algorithm model. For example, the positioning algorithm model will use information such as the output data of the GPS device and the output data of the IMU (Inertial Measurement Unit) to obtain the autonomous vehicle's current position information. The perception algorithm model can use a neural network model to process the output data of the GPS device, the LiDAR output data, and the camera output data to obtain obstacle information around the autonomous vehicle at the current moment. This obstacle information may include obstacle type, obstacle location, and / or obstacle speed. The prediction algorithm model can predict the obstacle's trajectory to obtain a predicted obstacle trajectory. The planning algorithm model can plan the autonomous vehicle's driving route based on the predicted obstacle trajectory, the output data of the GPS device, the LiDAR output data, and the camera output data.

[0035] S120. For each troubleshooting moment in the accident troubleshooting period, determine the hardware fault existing in the autonomous driving vehicle at the troubleshooting moment based on the hardware device detection data to obtain the hardware fault information corresponding to the troubleshooting moment; and determine the software fault existing in the autonomous driving vehicle at the troubleshooting moment based on the software module result data to obtain the software fault information corresponding to the troubleshooting moment.

[0036] Specifically, the accident investigation period may include multiple investigation times. After determining the accident investigation period, the cloud server can obtain the faults present in the autonomous vehicle at each investigation time to obtain the fault information corresponding to each investigation time. It should be noted that if the autonomous vehicle has fault E1 at investigation time t1, the time of occurrence of fault E1 can be either investigation time t1 or a time before investigation time t1. If fault E1 occurs before investigation time t1, fault E1 still exists at investigation time t1.

[0037] Fault information of autonomous vehicles can be divided into hardware fault information and software fault information. Hardware fault information is used to reflect hardware device faults that exist in the autonomous vehicle at the corresponding troubleshooting time. For example, it can reflect any one or any combination of LiDAR faults, camera faults, and GPS faults in the autonomous vehicle at the corresponding troubleshooting time. Software fault information is used to reflect algorithm faults that exist in the autonomous vehicle at the corresponding troubleshooting time. For example, it can reflect any one or any combination of positioning algorithm faults, perception algorithm faults, prediction algorithm faults, and planning algorithm faults in the autonomous vehicle at the corresponding troubleshooting time.

[0038] For each troubleshooting moment, the cloud server can determine the hardware faults existing in the autonomous driving vehicle at the troubleshooting moment based on the hardware device detection data to obtain the hardware fault information corresponding to the troubleshooting moment; and the cloud server can determine the software faults existing in the autonomous driving vehicle at the troubleshooting moment based on the software module result data to obtain the software fault information corresponding to the troubleshooting moment.

[0039] S130, based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment, determine the critical moment of the accident from each troubleshooting moment; the critical moment of the accident is used to locate the time when the cause of the fault that caused the accident occurred.

[0040] Specifically, the cloud server can determine one or more target troubleshooting moments from multiple troubleshooting moments based on the hardware and software fault information at each troubleshooting moment. These one or more target troubleshooting moments are the critical moments of the accident. The critical moments of the accident can be used to locate the time when the fault that caused the accident occurred.

[0041] In an embodiment of the present application, if an autonomous vehicle is involved in an accident, the time of the accident is used as the end time of the accident investigation period, and the hardware device detection data and software module result data corresponding to the accident investigation period are used as the accident data. The hardware device detection data is used to determine the hardware faults present in the autonomous vehicle at each investigation time during the accident investigation period, thereby obtaining hardware fault information for each investigation time. Furthermore, the software module result data is used to determine the software faults present in the autonomous vehicle at each investigation time during the accident investigation period, thereby obtaining software fault information for each investigation time. Based on each investigation time during the accident investigation period, the hardware fault information corresponding to each investigation time, and the software fault information corresponding to each investigation time, the critical moment of the accident is determined from each investigation time. In this way, the critical moment of the accident can be quickly determined in an automated manner. Because the critical moment of the accident can be used to locate the time when the cause of the fault occurred, when analyzing the cause of the accident based on the critical moment of the accident, the scope of the cause of the fault can be narrowed, thereby shortening the time required for analyzing the cause of the accident and improving analysis efficiency.

[0042] In one embodiment, Figure 2 As shown, based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment, the steps of determining the critical moment of the accident from each troubleshooting moment include:

[0043] S210: Obtain the accident contribution weight corresponding to each fault. Faults include software faults and hardware faults. The accident contribution weight corresponding to each fault reflects the probability of the fault causing the accident. In one embodiment, the accident contribution weight corresponding to each fault can be determined based on the severity and confidence level of the fault.

[0044] S220 , calculating the accident inducing index corresponding to each troubleshooting moment according to the accident contribution weight corresponding to each fault, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment.

[0045] Specifically, for each fault information corresponding to the troubleshooting moment, the cloud server performs the following steps to obtain the accident induction index corresponding to each troubleshooting moment: according to the accident contribution weight corresponding to each fault, the hardware fault information corresponding to the troubleshooting moment, and the software fault information corresponding to the troubleshooting moment, the accident induction index corresponding to the troubleshooting moment is calculated. The accident induction index can be used to evaluate the probability of an accident being caused by the fault existing at the troubleshooting moment.

[0046] In one embodiment, when calculating the accident susceptibility index corresponding to each troubleshooting time, the cloud server may add the accident contribution weights corresponding to each hardware fault and each software fault present in the autonomous vehicle at that time to obtain the accident susceptibility index corresponding to that time. For example, at troubleshooting time t2, if the autonomous vehicle has faults E1, E2, and E3, the cloud server may add the accident contribution weights corresponding to fault E1, fault E2, and fault E3. The result of this addition is the accident susceptibility index corresponding to troubleshooting time t2.

[0047] S230: sorting the investigation moments based on the numerical values ​​of the accident inducing indexes, and determining the accident critical moment from the sorted investigation moments.

[0048] Specifically, the cloud server can sort the accident induction indices by numerical value and determine a target accident induction index from the sorted indices. In one example, the cloud server can use the highest value among the accident induction indices as the target accident induction index and the investigation moment corresponding to this target accident induction index as the critical moment of the accident. This can further narrow the scope of investigation for the cause of the fault, further shorten the time spent analyzing the cause of the accident, and improve analysis efficiency.

[0049] In an embodiment of the present application, the accident induced index corresponding to each troubleshooting moment is calculated based on the accident contribution weight corresponding to each fault and the fault information corresponding to each troubleshooting moment, so as to evaluate the probability of an accident being induced by the fault existing at the troubleshooting moment through the accident induced index. After obtaining the accident induced index corresponding to each troubleshooting moment, the present application can determine the critical moment of the accident from each troubleshooting moment based on each accident induced index, thereby being able to quickly determine the critical moment of the accident in an automated manner. Since the critical moment of the accident can be used to locate the time when the cause of the fault occurs, when analyzing the cause of the accident based on the critical moment of the accident, the scope of investigation of the cause of the fault can be narrowed, thereby further shortening the time spent on analyzing the cause of the accident and improving the efficiency of the analysis.

[0050] In one embodiment, the step of determining the critical moment of the accident from each troubleshooting moment based on each troubleshooting moment, hardware fault information corresponding to each troubleshooting moment, and software fault information corresponding to each troubleshooting moment includes:

[0051] Obtaining the critical moment determination model;

[0052] Generate fault sequence information based on each troubleshooting moment, hardware fault information corresponding to each troubleshooting moment, and software fault information corresponding to each troubleshooting moment;

[0053] The fault sequence information is input into the critical moment determination model to obtain the accident critical moment output by the critical moment determination model.

[0054] Specifically, autonomous driving systems involve multiple modules with complex coupling relationships between them. Multiple fault combinations may occur at any given moment. Furthermore, a single accident is often caused by multiple faults. Therefore, automated methods can be used to accurately and quickly identify the critical moments associated with each fault, further shortening the time required to analyze the cause of the accident and improving analysis efficiency.

[0055] In an embodiment of the present application, an algorithmic model can be used to determine the critical moment of an accident from multiple troubleshooting moments. Specifically, the cloud server can obtain a critical moment determination model and generate fault sequence information based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment. In one embodiment, the fault sequence information can be a fault sequence list, such as that shown in Table 1.

[0056] Table 1 Fault sequence list

[0057]

[0058] The cloud server inputs the fault timing information into the critical moment determination model, so that the critical moment of the accident can be determined from multiple investigation moments through the critical moment determination model, and the critical moment of the accident output by the critical moment determination model can be obtained.

[0059] In one embodiment, the step of obtaining the critical moment determination model includes:

[0060] Obtaining an initial algorithm model, the initial algorithm model being configured with accident contribution weights corresponding to each fault, the initial algorithm model being configured to receive multiple moments and fault information corresponding to each moment, and calculating an accident inducing index corresponding to each moment based on the configured accident contribution weights and the received fault information corresponding to each moment, and outputting the moment corresponding to the maximum accident inducing index among the accident inducing indexes;

[0061] Obtaining a training set, the training set including multiple groups of training data, each group of training data including multiple training moments, fault information corresponding to each training moment, and key training moments manually marked from the multiple training moments;

[0062] For each set of training data, multiple training moments included in the set of training data and fault information corresponding to each training moment are input into the initial algorithm model, and an output moment of the initial algorithm model is obtained; based on the error between the output moment and the key training moment included in the set of training data, the accident contribution weight corresponding to each fault in the initial algorithm model is adjusted until the initial algorithm model outputs the key training moment included in the set of training data;

[0063] The initial algorithm model trained with each set of training data is used as the key moment determination model.

[0064] In an embodiment of the present application, the accident contribution weight corresponding to each fault can be obtained by model training. The cloud server can obtain a training set, which includes multiple groups of training data. Each group of training data includes multiple training moments, fault information corresponding to each training moment, and a key training moment. The key training moment is manually selected from the multiple training moments in the group. The cloud server can input each group of training data into the initial algorithm model. For each group of training data, the initial algorithm model calculates the accident induction index corresponding to each training moment in the group of training data based on the currently configured accident contribution weights, and outputs the training moment corresponding to the maximum accident induction index in the group of training data as the key moment of the group of training data. The cloud server can compare the key moment output by the initial algorithm model with the key training moment of the group of training data to obtain the error between the two, and then adjust the accident contribution weight corresponding to each fault in the initial algorithm model based on the error, until the key moment output by the initial algorithm model is the manually marked key training moment.

[0065] In one example, the ReLU function can be used as the loss function during model training, and the stochastic gradient descent method can be used to adjust the accident contribution weights corresponding to each fault. In another example, if the fault types include E1, E2, and E3, the accident contribution weights corresponding to each fault type are W1, W2, and W3, respectively. F1n, F2n, and F3n are used to reflect whether faults E1, E2, and E3 occur at time Tn, respectively. For example, if fault E1 occurs at time Tn, F1n takes the value of 1, and if fault E1 does not occur at time Tn, F1n takes the value of 0. Then, the accident induced index Sn at time Tn = F1n*W1+F2n*W2+F3n*W3.

[0066] After training the initial algorithm model using multiple sets of training data, the cloud server can obtain the trained initial algorithm model, which is the critical moment determination model. The cloud server can use this critical moment determination model to automatically infer the critical moment of the accident.

[0067] In an embodiment of the present application, the accident contribution weight corresponding to each fault is obtained by model training, and the critical moment of the accident is obtained by the trained model, so that the accident contribution weight corresponding to each fault can be accurately and quickly obtained through an automated method.

[0068] In one embodiment, the autonomous vehicle may be equipped with a camera device for capturing images of the surroundings of the autonomous vehicle. When the autonomous vehicle is equipped with a camera device, the cloud server may determine whether the camera device has a fault at each inspection moment based on the hardware device detection data, and obtain hardware fault information accordingly. Specifically, Figure 3 As shown, for each troubleshooting moment, the cloud server can perform the following steps:

[0069] S310: Extract a first image set captured by the camera device within a fault determination period from hardware device detection data.

[0070] Since the end time of the fault judgment period is the troubleshooting time, the first image set is the image set captured by the camera device some time before the troubleshooting time.

[0071] S320: Calculate the camera output frequency corresponding to the screening moment according to the first image set.

[0072] It is understood that the present application may calculate the camera output frequency corresponding to the troubleshooting moment using any method known in the art, and this application does not impose any specific limitation thereto. In one embodiment, the cloud server may calculate the camera output frequency corresponding to the troubleshooting moment based on the total number of images in the first image set and the duration of the fault judgment period. For example, the ratio of the total number of images in the first image set to the duration of the fault judgment period may be used as the camera output frequency corresponding to the troubleshooting moment.

[0073] S330: Determine whether the autonomous driving vehicle has a camera device failure at the time of the inspection based on the camera output frequency, and obtain a first judgment result.

[0074] S340, obtaining hardware fault information corresponding to the inspection time according to the first judgment result, and the hardware fault information can reflect whether there is a camera equipment failure in the autonomous driving vehicle at the inspection time.

[0075] In one embodiment, the cloud server may compare the camera output frequency corresponding to the check time with a pre-set first frequency threshold, and based on the comparison result, determine whether the autonomous vehicle has a camera device failure at the check time. For example, if the camera output frequency is less than the first frequency threshold, it is determined that the camera device is suspended at the check time, i.e., the autonomous vehicle has a camera device failure at the check time.

[0076] In this embodiment, the first image set corresponding to each inspection moment is extracted from the hardware device detection data, and the camera output frequency corresponding to the inspection moment is calculated based on each first image set. Based on the camera output frequency corresponding to each inspection moment, the cloud server can determine whether the autonomous vehicle has a camera device failure at that inspection moment and thereby obtain hardware failure information corresponding to that inspection moment. This allows the cloud server to automatically determine and identify camera device failures, thereby reducing the level of manual intervention in the accident analysis process and further improving analysis efficiency.

[0077] In one embodiment, the step of calculating the camera output frequency corresponding to the screening moment based on the first image set includes:

[0078] removing the black screen image from the first image set to obtain a second image set;

[0079] The total number of images in the second image set is obtained, and the ratio of the total number of images to the duration of the fault judgment period is used as the camera output frequency corresponding to the troubleshooting moment.

[0080] Specifically, when a camera device fails, there is a situation where the camera device can still output images at the set frequency, but the output images contain one or more black screen images that cannot correctly reflect the environment around the autonomous driving vehicle, making it impossible for the software module to perform positioning, perception, trajectory prediction and planning based on the images output by the camera device.

[0081] In order to accurately determine whether the camera equipment has malfunctioned, in this embodiment, for each troubleshooting moment, after obtaining the first image set corresponding to the troubleshooting moment, the cloud server can identify and remove the black screen image in the first image set to obtain a second image set. The second image set can include all other images in the first image set except the black screen image. For example, at a certain troubleshooting moment, the first image set includes three images: (T0, Image0), (T1, Image1), and (T2, Image2). After deduplication of the black screen image, the second image set obtained is: (T0, Image0), (T2, Image2).

[0082] After obtaining the second image set corresponding to the troubleshooting moment, the cloud server can obtain the total number of images in the second image set, and use the ratio of the total number of images in the second image set to the length of the fault judgment period as the camera output frequency corresponding to the troubleshooting moment.

[0083] This eliminates the interference of black screen images and allows accurate determination of camera equipment failure. This allows hardware fault information to more accurately reflect the hardware fault present in the autonomous vehicle at the time of the investigation. Furthermore, the accident susceptibility index calculated based on hardware fault information can more accurately reflect the probability of an accident caused by the fault present at the time of the investigation. This allows for more precise identification of the moment of occurrence of the underlying fault at the critical moment of the accident, further improving analysis efficiency.

[0084] In one embodiment, the autonomous vehicle may be equipped with a radar that is configured to send detection data packets based on radar detection results, and the hardware device detection data includes detection data packets sent by the radar. In one embodiment, the radar may send detection data packets to the autonomous driving system on the autonomous vehicle via a network UDP (User Datagram Protocol). When the autonomous vehicle is equipped with a radar, the cloud server can determine whether the radar is faulty at each inspection moment based on the hardware device detection data and obtain hardware fault information based on this.

[0085] Specifically, if Figure 4 As shown, for each troubleshooting moment, the cloud server can perform the following steps:

[0086] At step S410, the reception time of each detection data packet is obtained, and each detection data packet whose reception time is before the check time is used as the first target data packet. In other words, the cloud server may use the detection data packet received by the autonomous driving system before the check time as the first target data packet.

[0087] At step S420, the two first target data packets with the smallest absolute difference between the reception time and the check time are used as the second target data packets. In this way, the cloud server uses the two detection data packets received by the autonomous driving system at the latest before the check time as the second target data packets.

[0088] S430: Based on the difference between the reception times of the two second target data packets, determine whether the autonomous driving vehicle has a radar failure at the time of the inspection, and obtain a second judgment result.

[0089] In one embodiment, if the time interval between the reception of two detection data packets is too long, a radar malfunction may be determined. For example, if one second target data packet is received at time T1 and another second target data packet is received at time T2, the cloud server may calculate the difference between the reception times of the two second target data packets, i.e., (T1-T2), and determine whether the difference is greater than a first preset threshold to obtain a second determination result.

[0090] S440: Obtain hardware fault information corresponding to the inspection time based on the second judgment result. The hardware fault information can reflect whether the autonomous driving vehicle has a radar fault at the inspection time.

[0091] In this embodiment, two second target data packets corresponding to each inspection time are extracted from the hardware device detection data. The difference between the reception times of the two second target data packets is used to determine whether the autonomous vehicle has a radar fault at the inspection time, thereby obtaining the hardware fault information corresponding to the inspection time. In this way, the cloud server can automatically determine and identify radar faults, thereby reducing the level of manual intervention in the accident analysis process and further improving analysis efficiency.

[0092] In one embodiment, the step of determining whether the autonomous driving vehicle has a radar fault at the time of the inspection based on the difference between the reception times of the two second target data packets and obtaining a second determination result includes:

[0093] When the difference between the receiving times of the two second target data packets is greater than a first preset threshold, respectively calculating the sending and receiving transmission durations of the two second target data packets;

[0094] Comparing the transmission time of each of the two second target data packets with a second preset threshold value to obtain a comparison result;

[0095] Based on the comparison result, it is determined whether there is a radar failure in the autonomous driving vehicle at the time of inspection, and a second judgment result is obtained.

[0096] Specifically, a radar malfunction or excessive radar transmission delay can cause the difference in the reception times of the two second target data packets to exceed a threshold. Therefore, to eliminate the impact of transmission delay, when the difference in the reception times of the two second target data packets exceeds a first preset threshold, the cloud server can obtain the transmission and reception times ΔT1 and ΔT2 of the two second target data packets, respectively. The transmission and reception time is the time it takes from the radar sending the detection data packet to the onboard autonomous driving system receiving the detection data packet.

[0097] After obtaining the transmission and reception times ΔT1 and ΔT2, the cloud server may compare the transmission and reception time ΔT1 with a second preset threshold, and compare the transmission and reception time ΔT2 with a second preset threshold to obtain a comparison result. This comparison result may indicate whether the transmission delay of any second target data packet is excessively high, allowing the cloud server to determine whether the autonomous vehicle has a radar fault based on the comparison result, thereby obtaining a second determination result.

[0098] This eliminates the impact of excessive transmission delay on radar fault diagnosis, allowing hardware fault information to more accurately reflect the hardware fault present in the autonomous vehicle at the time of the troubleshooting. Furthermore, the accident propensity index calculated based on hardware fault information more accurately reflects the probability of an accident caused by the fault present at the time of the troubleshooting. This allows for more accurate identification of the cause of the fault at the critical moment of the accident, further improving analysis efficiency.

[0099] In one embodiment, the autonomous vehicle may be equipped with a GPS device that can be used for GPS positioning. When the autonomous vehicle is equipped with a GPS device, the cloud server can determine whether the GPS device has a fault at each troubleshooting moment based on the hardware device detection data, and obtain hardware fault information accordingly. Specifically, Figure 5 As shown, for each troubleshooting moment, the cloud server can perform the following steps:

[0100] S510, determining a fault judgment period, where the end time of the fault judgment period is the troubleshooting time.

[0101] S520 , extracting a GPS log segment from the hardware device detection data, where the GPS log segment is used to record the positioning results output by the GPS device within the fault determination period and the output time of each positioning result.

[0102] Since the end time of the fault judgment period is the troubleshooting time, the cloud server can extract the positioning results output by the GPS device within a period of time before the troubleshooting period from the hardware device detection data, as well as the time when the GPS outputs the positioning results (that is, the output time of the positioning results).

[0103] S530 , calculating, based on the GPS log segments, a first difference between every two adjacent output times on the time axis, and a second difference between the latest output time and the check time.

[0104] Specifically, when the GPS device fails to output a positioning result for a period of time, it can be determined that the GPS device of the autonomous vehicle has failed. Therefore, to achieve automatic judgment and identification of GPS device failures, this application can calculate the first difference between every two adjacent output times on the time axis within the fault judgment period corresponding to the troubleshooting time, as well as the second difference between the latest output time and the troubleshooting time, so as to determine whether the GPS device has failed to output a positioning result for a period of time based on each first difference and second difference, and further determine whether the GPS device has failed.

[0105] For example, the GPS log segment corresponding to a certain troubleshooting time may include the following information: (Positioning result A, output time Ta), (Positioning result B, output time Tb), (Positioning result C, output time Tc), Ta < Tb < Tc. For this GPS log segment, the cloud server can calculate the first difference between Tb and Ta, the first difference between Tc and Tb, and the second difference between the troubleshooting time and Tc.

[0106] S540, based on each first difference and second difference, determine whether there is a GPS device failure in the autonomous vehicle at this troubleshooting time, and obtain a third judgment result.

[0107] In one embodiment, the cloud server can respectively compare each first difference with a third preset threshold, and compare the second difference with the third preset threshold. When any first difference is greater than the third preset threshold, or the second difference is greater than the third preset threshold, it can be determined that there is a GPS device failure in the autonomous vehicle at this troubleshooting time. When each first difference is less than or equal to the third preset threshold, and the second difference is also less than or equal to the third preset threshold, it can be determined that there is no GPS device failure in the autonomous vehicle at this troubleshooting time.

[0108] S550, obtain the hardware failure information corresponding to this troubleshooting time according to the third judgment result, and the hardware failure information can reflect whether there is a GPS device failure in the autonomous vehicle at this troubleshooting time.

[0109] In this embodiment, it is possible to automatically determine whether there is a failure in the GPS device of the autonomous vehicle at each troubleshooting time according to the time when the GPS device outputs a positioning result during the fault judgment period, and thus obtain the hardware failure information corresponding to this troubleshooting time. In this way, the cloud server can achieve automatic judgment and identification of GPS device failures, and further reduce the degree of manual participation in the accident analysis process to further improve the analysis efficiency.

[0110] In one embodiment, the step of determining the software failure existing in the autonomous vehicle at this troubleshooting time according to the software module result data to obtain the software failure information corresponding to this troubleshooting time includes:

[0111] Extract process operation logs from software module result data.

[0112] Determine the process running by the autonomous driving vehicle at the time of the inspection based on the process running log; based on the process running by the autonomous driving vehicle at the time of the inspection, determine whether the autonomous driving vehicle has a process crash failure at the time of the inspection, and obtain a fourth judgment result; obtain the software fault information corresponding to the time of the inspection based on the fourth judgment result.

[0113] The process log records the processes running in the autonomous vehicle at each inspection time, thereby revealing the operating status of each software module at each inspection time. For example, the process log can be used to obtain the process operation status of the positioning algorithm model, perception algorithm model, prediction algorithm model, and planning algorithm model at each inspection time.

[0114] For each troubleshooting moment, the cloud server can perform the following steps: extract the process running by the autonomous driving vehicle at the troubleshooting moment from the process running log, and judge whether the software module of the autonomous driving vehicle has a process crash failure at the troubleshooting moment based on this, so as to obtain the software fault information corresponding to the troubleshooting moment according to the judgment result.

[0115] In one embodiment, if the autonomous vehicle does not have a positioning algorithm model process at the time of the check, it can be determined that the autonomous vehicle has a positioning model crash at the time of the check. Similarly, if the autonomous vehicle does not have a perception algorithm model process at the time of the check, it can be determined that the autonomous vehicle has a perception model crash at the time of the check; if the autonomous vehicle does not have a prediction algorithm model process at the time of the check, it can be determined that the autonomous vehicle has a prediction model crash at the time of the check; if the autonomous vehicle does not have a planning algorithm model process at the time of the check, it can be determined that the autonomous vehicle has a planning model crash at the time of the check.

[0116] In this embodiment, the cloud server can automatically determine and identify process crash failures based on process operation logs, thereby reducing the degree of manual participation in the accident analysis process to further improve analysis efficiency.

[0117] In addition to process crashes, software modules running on autonomous vehicles can also experience other types of failures. For example, a prediction algorithm model can also experience a prediction error. The prediction algorithm model is used to predict the behavior and trajectory of obstacles, thereby pre-determining the relative position between the autonomous vehicle and the obstacle. This allows the planning algorithm model to perform path planning based on this information, thereby reducing the occurrence of accidents. When the obstacle trajectory predicted by the prediction algorithm model does not match the actual obstacle trajectory, the prediction algorithm model has failed.

[0118] like Figure 6 As shown in the figure, the relative position relationship between the autonomous driving vehicle (i.e., the main vehicle) and the obstacle can be divided into 8 relative positions, namely: left front (i.e., Figure 6 1), before (i.e. Figure 6 2), right front (i.e. Figure 6 3), left (i.e. Figure 6 4 in), right (i.e. Figure 6 6), left rear (i.e. Figure 6 7 in), after (ie Figure 6 8) and right rear (i.e. Figure 6 9). In a cut-in scenario, the obstacle will move from the left / left front / left rear / right front / right / right rear to the front / rear of the autonomous vehicle. For example, a cut-in scenario might have the obstacle on the left at T0 and T1, on the left front at T2 and T3, and in front at T4.

[0119] In one embodiment, to enable the cloud server to automatically determine whether the autonomous vehicle has a predicted fault error at each troubleshooting moment, when the autonomous vehicle is running a predictive algorithm model, the cloud server may perform the following steps for each troubleshooting moment:

[0120] Determine the fault judgment period and extract the obstacle prediction trajectory output by the prediction algorithm model during the fault judgment period from the software module result data; the end time of the fault judgment period is the time of the troubleshooting;

[0121] Determine whether the prediction algorithm model has a prediction error fault at the time of troubleshooting based on the predicted obstacle trajectory and the actual trajectory of the corresponding obstacle, and obtain a fifth judgment result;

[0122] The software fault information corresponding to the troubleshooting moment is obtained based on the fifth judgment result.

[0123] In one embodiment, when the predicted trajectory of any obstacle does not match the actual trajectory of that obstacle, the prediction algorithm model may be determined to have a prediction error fault at the time of the troubleshooting. In another embodiment, the cloud server may count the number and / or duration of discrepancies between the predicted and actual obstacle trajectories and, based on this, determine whether the prediction algorithm model has a prediction error fault at the time of the troubleshooting.

[0124] In one embodiment, the cloud server can also combine the relative distance between the obstacle and the autonomous vehicle to determine whether the prediction algorithm model has a prediction error fault at the time of the inspection. Alternatively, if the autonomous vehicle has a prediction error fault at a certain inspection time, the accident contribution weight corresponding to the prediction error fault can be determined based on the relative distance between the obstacle whose predicted trajectory does not match the actual trajectory and the autonomous vehicle. For example, when the relative distance is small, the prediction error fault is more likely to cause an accident. In this case, the accident contribution weight corresponding to the prediction error fault can be determined as a larger value; conversely, when the relative distance is large, the accident contribution weight corresponding to the prediction error fault can be determined as a smaller value.

[0125] In this embodiment, the cloud server can realize automatic judgment and identification of predicted erroneous faults, thereby reducing the degree of manual participation in the accident analysis process to further improve analysis efficiency.

[0126] Positioning algorithm models may also experience positioning failures. In one embodiment, to automatically identify positioning failures, when the software module result data includes a positioning error log output by the positioning algorithm model, the cloud server can determine the error time from the positioning error log. Based on the error time, the cloud server can determine whether the autonomous vehicle experienced a positioning failure at the time of the investigation, and thereby obtain the software fault information corresponding to the investigation time.

[0127] For the perception algorithm model and the planning algorithm model, frame loss failure may also occur. In one embodiment, in order to automatically determine and identify frame loss failures, the cloud server can extract the obstacle information output by the perception algorithm model during the fault judgment period, and the planning data output by the planning algorithm model during the fault judgment period from the software module result data. The cloud server can calculate the perception output frequency corresponding to the troubleshooting moment based on the obstacle information, and based on the perception output frequency corresponding to the troubleshooting moment, determine whether the perception algorithm model has a frame loss failure at the troubleshooting moment, and thereby obtain the software fault information corresponding to the troubleshooting moment. Similarly, the cloud server can calculate the planning output frequency corresponding to the troubleshooting moment based on the planning data, and thereby determine whether the planning algorithm model has a frame loss failure at the troubleshooting moment, and thereby obtain the software fault information corresponding to the troubleshooting moment.

[0128] In one embodiment, the method of the present application further includes outputting hardware fault information corresponding to the critical moment of the accident and software fault information corresponding to the critical moment of the accident. This allows direct access to the scope of investigation of the cause of the fault, further shortening the time spent analyzing the cause of the accident and improving analysis efficiency.

[0129] In one example, if Figure 7As shown, the accident cause determination method provided in the embodiment of the present application may include the following steps:

[0130] S610: In the event of an accident involving an autonomous driving vehicle, hardware fault information and software module result data corresponding to the accident investigation period are used as accident data.

[0131] S620: Analyze the accident data to determine possible faults of the autonomous driving vehicle at each inspection moment.

[0132] S630: Generate a fault time series list, which records each troubleshooting moment and the faults existing in the autonomous driving vehicle at each troubleshooting moment.

[0133] Specifically, based on the accident data, the cloud server can determine whether there is a camera equipment failure, lidar failure, GPS equipment failure, positioning model crash failure, positioning failure failure, perception model crash failure, perception model frame loss failure, prediction model crash failure, prediction error failure, planning model crash failure, and planning model frame loss failure at each inspection moment, and generate a fault time series list as shown in Table 1. In this way, the cloud server can automatically analyze the possible fault problems of each module and provide a list of possible fault problems.

[0134] S640: Acquire a critical moment judgment model.

[0135] Specifically, the critical moment determination model may be configured with accident contribution weights corresponding to each fault. The critical moment determination model may be a model obtained by training an initial algorithm model using multiple sets of training data.

[0136] Among them, when training the initial algorithm model, for each group of training data: the multiple training moments included in the group of training data and the fault information corresponding to each training moment are input into the initial algorithm model, so that the initial algorithm model calculates the accident inducing index corresponding to each training moment according to the fault information corresponding to each training moment and the accident contribution weights corresponding to each configured fault, and screens out the maximum accident inducing index from the accident inducing indexes corresponding to each training moment, and outputs the training moment corresponding to the maximum accident inducing index; based on the error between the training moment corresponding to the maximum accident inducing index and the key training moment manually marked and included in the group of training data, adjust the accident contribution weights corresponding to each fault in the initial algorithm model until the initial algorithm model outputs the key training moment included in the group of training data.

[0137] Furthermore, the ReLU function can be used as the loss function during model training, and the stochastic gradient descent method can be used to adjust the accident contribution weights corresponding to each fault.

[0138] S650: Input the fault time sequence list into the critical moment judgment model, and obtain the critical moment of the accident output by the critical moment judgment model.

[0139] In this example, the initial algorithm model is trained using labeled data to develop a critical moment judgment model. This allows for automatic, accurate, and rapid determination of the accident contribution weights corresponding to each fault. Furthermore, the critical moment judgment model can be used to quickly locate the cause of an autonomous vehicle accident.

[0140] The following describes an accident cause determination device provided in an embodiment of the present application. The accident cause determination device described below and the accident cause determination method described above can refer to each other.

[0141] In one embodiment, the present application provides an accident cause determination device 700, such as Figure 8 As shown, the device includes an accident data acquisition module 710, a fault information acquisition module 720 and an accident critical moment determination module 730.

[0142] The accident data acquisition module 710 is configured to, when an accident occurs with the autonomous driving vehicle, use the hardware device detection data and software module result data corresponding to the accident investigation period as accident data; the end time of the accident investigation period is the time of occurrence of the accident;

[0143] The fault information acquisition module 720 is configured to, for each troubleshooting moment in the accident troubleshooting period, determine, based on the hardware device detection data, a hardware fault existing in the autonomous driving vehicle at that troubleshooting moment, thereby obtaining hardware fault information corresponding to that troubleshooting moment; and determine, based on the software module result data, a software fault existing in the autonomous driving vehicle at that troubleshooting moment, thereby obtaining software fault information corresponding to that troubleshooting moment;

[0144] The accident critical moment determination module 730 is used to determine the accident critical moment from each troubleshooting moment based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment; the accident critical moment is used to locate the time when the cause of the fault that caused the accident occurred.

[0145] In one embodiment, the accident critical moment determination module 730 includes an accident contribution weight acquisition unit, an accident induction index acquisition unit, and a first accident critical moment acquisition unit.

[0146] The accident contribution weight acquisition unit is used to obtain the accident contribution weight corresponding to each fault;

[0147] The accident induction index acquisition unit is used to calculate the accident induction index corresponding to each troubleshooting moment according to the accident contribution weight corresponding to each fault, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment;

[0148] The accident critical moment acquisition unit is used to sort the investigation moments based on the numerical values ​​of the accident inducing indexes, and determine the accident critical moment from the sorted investigation moments.

[0149] In one embodiment, the accident critical moment determination module 730 includes a model acquisition unit, a time sequence information acquisition unit, and a second accident critical moment acquisition unit.

[0150] The model acquisition unit is used to acquire the key moment determination model;

[0151] The timing information acquisition unit is used to generate fault timing information based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment;

[0152] The second accident critical moment acquisition unit is used to input the fault time sequence information into the critical moment determination model to obtain the accident critical moment output by the critical moment determination model.

[0153] In one embodiment, the model acquisition unit includes an initial algorithm model acquisition unit, a training set acquisition unit, a training unit, and a critical moment determination model acquisition unit. The initial algorithm model acquisition unit is used to acquire an initial algorithm model configured with accident contribution weights corresponding to each fault, and to receive multiple moments and fault information corresponding to each moment, and to calculate the accident induction index corresponding to each moment based on the configured accident contribution weights and the received fault information corresponding to each moment, and to output the moment corresponding to the maximum accident induction index among the accident induction indices.

[0154] The training set acquisition unit is used to acquire a training set, which includes multiple groups of training data. Each group of training data includes multiple training moments, fault information corresponding to each training moment, and key training moments manually marked from the multiple training moments.

[0155] The training unit is used to input, for each group of training data, multiple training moments included in the group of training data and the fault information corresponding to each training moment into the initial algorithm model, and obtain the output moment of the initial algorithm model; according to the error between the output moment and the key training moment included in the group of training data, adjust the accident contribution weight corresponding to each fault in the initial algorithm model until the initial algorithm model outputs the key training moment included in the group of training data.

[0156] The critical moment determination model acquisition unit is used to use the initial algorithm model trained by each group of training data as the critical moment determination model.

[0157] In one embodiment, the autonomous driving vehicle is provided with a camera device. The fault information acquisition module 720 includes a first image set acquisition unit, a camera output frequency calculation unit, a first judgment unit and a hardware fault information acquisition unit. The first image set acquisition unit is used to extract the first image set acquired by the camera device during the fault judgment period from the hardware device detection data, and the end time of the fault judgment period is the troubleshooting time. The camera output frequency calculation unit is used to calculate the camera output frequency corresponding to the troubleshooting time based on the first image set. The first judgment unit is used to judge whether the autonomous driving vehicle has a camera device fault at the troubleshooting time based on the camera output frequency, and obtain a first judgment result. The hardware fault information acquisition unit is used to obtain the hardware fault information corresponding to the troubleshooting time based on the first judgment result.

[0158] In one embodiment, the camera output frequency calculation unit includes a black screen image removal unit and a first ratio calculation unit. The black screen image removal unit is configured to remove black screen images from the first image set to obtain a second image set. The first ratio calculation unit is configured to obtain the total number of images in the second image set and use the ratio of the total number of images to the duration of the fault determination period as the camera output frequency corresponding to the troubleshooting moment.

[0159] In one embodiment, the autonomous driving vehicle is provided with a radar, and the hardware device detection data includes a detection data packet sent by the radar. The fault information acquisition module 720 includes a first target data packet determination unit, a second target data packet determination unit, a second judgment unit, and a hardware fault information acquisition unit. Among them, the first target data packet determination unit is used to obtain the receiving time of each detection data packet, and each detection data packet whose receiving time is before the troubleshooting time is used as the first target data packet. The second target data packet determination unit is used to use the two first target data packets whose absolute value of the difference between the receiving time and the troubleshooting time is the smallest as the second target data packets. The second judgment unit is used to judge whether the autonomous driving vehicle has a radar fault at the troubleshooting time based on the difference between the receiving times of the two second target data packets, and obtain a second judgment result. The hardware fault information acquisition unit is used to obtain the hardware fault information corresponding to the troubleshooting time based on the second judgment result.

[0160] In one embodiment, the second judgment unit includes a transmission and reception duration calculation unit, a comparison unit, and a second judgment result acquisition unit. The transmission and reception duration calculation unit is configured to calculate the transmission and reception duration of each of the two second target data packets when the difference between the reception times of the two second target data packets is greater than a first preset threshold. The comparison unit is configured to compare the transmission and reception duration of each of the two second target data packets with the second preset threshold to obtain a comparison result. The second judgment result acquisition unit is configured to determine whether there is a radar fault in the autonomous driving vehicle at the time of the inspection based on the comparison result, and obtain the second judgment result.

[0161] In one embodiment, the autonomous vehicle is equipped with a GPS device. The fault information acquisition module 720 includes a judgment period determination unit, a GPS log extraction unit, a difference calculation unit, a third judgment unit, and a hardware fault information acquisition unit. The judgment period determination unit is used to determine a fault judgment period, with the end time of the fault judgment period being the troubleshooting time. The GPS log extraction unit is used to extract GPS log segments from the hardware device detection data; the GPS log segments are used to record the positioning results output by the GPS device during the fault judgment period and the output time of each positioning result. The difference calculation unit is used to calculate, based on the GPS log segments, a first difference between each two adjacent output times on the time axis, and a second difference between the latest output time and the troubleshooting time. The third judgment unit is used to determine whether the autonomous vehicle has a GPS device fault at the troubleshooting time based on each of the first and second differences, and obtain a third judgment result. The hardware fault information acquisition unit is used to obtain the hardware fault information corresponding to the troubleshooting time based on the third judgment result.

[0162] In one embodiment, the fault information acquisition module 720 includes a process running log extraction unit, a running process determination unit, a fourth judgment unit, and a software fault information acquisition unit. Among them, the running log extraction unit is used to extract the process running log from the software module result data. The running process determination unit is used to determine the process running by the autonomous driving vehicle at the time of the troubleshooting based on the process running log. The fourth judgment unit is used to determine whether the autonomous driving vehicle has a process crash fault at the time of the troubleshooting based on the process running by the autonomous driving vehicle at the time of the troubleshooting, and obtain a fourth judgment result. The software fault information acquisition unit is used to obtain the software fault information corresponding to the troubleshooting time according to the fourth judgment result.

[0163] In one embodiment, the autonomous driving vehicle runs a prediction algorithm model. The fault information acquisition module 720 includes a prediction trajectory acquisition unit, a fifth judgment unit, and a software fault information acquisition unit. The prediction trajectory acquisition unit is used to determine the fault judgment period, and extract the obstacle prediction trajectory output by the prediction algorithm model during the fault judgment period from the software module result data; wherein the end time of the fault judgment period is the troubleshooting time. The fifth judgment unit is used to judge whether the prediction algorithm model has a prediction error fault at the troubleshooting time based on the obstacle prediction trajectory and the actual trajectory of the corresponding obstacle, and obtain a fifth judgment result. The software fault information acquisition unit is used to obtain the software fault information corresponding to the troubleshooting time based on the fifth judgment result.

[0164] In one embodiment, the device 700 further includes a fault information output module, which is configured to output hardware fault information corresponding to the critical moment of the accident and software fault information corresponding to the critical moment of the accident.

[0165] In one embodiment, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the accident cause determination method as described in any of the above embodiments.

[0166] In one embodiment, the present application also provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the accident cause determination method as described in any of the above embodiments.

[0167] Schematically, as Figure 9 As shown, Figure 9 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 900 can be provided as a server. Figure 9 Computer device 900 includes a processing component 902, which further includes one or more processors, and memory resources represented by memory 901 for storing instructions executable by processing component 902, such as application programs. The application programs stored in memory 901 may include one or more modules, each corresponding to a set of instructions. Furthermore, processing component 902 is configured to execute the instructions to perform the accident cause determination method according to any of the above-described embodiments.

[0168] The computer device 900 may further include a power supply component 903 configured to perform power management of the computer device 900, a wired or wireless network interface 904 configured to connect the computer device 900 to a network, and an input / output (I / O) interface 905. The computer device 900 may operate based on an operating system stored in the memory 901, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0169] Those skilled in the art will understand that the internal structure of the computer device shown in the present application is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0170] Finally, it should be noted that, in this article, relational terms such as first and second are merely used to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. Without further restriction, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element. Herein, "one," "said," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. A plurality refers to at least two, such as 2, 3, 5, or 8. "And / or" includes any and all combinations of the relevant listed items.

[0171] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0172] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining the cause of an accident, characterized in that: The method comprises: In the event of an accident involving an autonomous vehicle, the hardware device detection data and software module result data corresponding to the accident investigation period are used as accident data; the end time of the accident investigation period is the time of occurrence of the accident, and the autonomous vehicle operates a prediction algorithm model; For each troubleshooting moment in the accident troubleshooting period, determining a hardware fault existing in the autonomous driving vehicle at the troubleshooting moment based on the hardware device detection data, thereby obtaining hardware fault information corresponding to the troubleshooting moment; and determining a software fault existing in the autonomous driving vehicle at the troubleshooting moment based on the software module result data, thereby obtaining software fault information corresponding to the troubleshooting moment; Based on each troubleshooting moment, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment, a critical moment of the accident is determined from each troubleshooting moment; the critical moment of the accident is used to locate the time when the cause of the fault that caused the accident occurred; The step of determining the software fault existing in the autonomous driving vehicle at the time of the troubleshooting based on the software module result data to obtain software fault information corresponding to the time of the troubleshooting includes: Extracting a process running log from the software module result data; the process running log records the process running by the autonomous driving vehicle at each inspection moment; Determine the process of the autonomous driving vehicle running at the time of the inspection according to the process running log; Based on the process running by the autonomous driving vehicle at the time of the check, determining whether the autonomous driving vehicle has a process crash fault at the time of the check, and obtaining a fourth determination result; Obtaining software fault information corresponding to the troubleshooting moment according to the fourth judgment result; The step of determining the software fault existing in the autonomous driving vehicle at the time of the troubleshooting based on the software module result data to obtain software fault information corresponding to the time of the troubleshooting further includes: Determining a fault judgment period, and extracting the obstacle prediction trajectory output by the prediction algorithm model within the fault judgment period from the software module result data; wherein the end time of the fault judgment period is the troubleshooting time; determining, based on the predicted obstacle trajectory and the actual trajectory of the corresponding obstacle, whether the prediction algorithm model has a prediction error fault at the time of troubleshooting, and obtaining a fifth judgment result; The software fault information corresponding to the troubleshooting moment is obtained based on the fifth judgment result.

2. The method for determining the cause of an accident according to claim 1, wherein: Based on each troubleshooting moment, hardware fault information corresponding to each troubleshooting moment, and software fault information corresponding to each troubleshooting moment, the steps of determining the critical moment of the accident from each troubleshooting moment include: Obtain the accident contribution weight corresponding to each fault; Calculate the accident inducibility index corresponding to each troubleshooting moment based on the accident contribution weight corresponding to each fault, the hardware fault information corresponding to each troubleshooting moment, and the software fault information corresponding to each troubleshooting moment; Based on the numerical values ​​of the accident inducing indexes, the investigation moments are sorted, and the critical accident moment is determined from the sorted investigation moments.

3. The method for determining the cause of an accident according to claim 1, wherein: Based on each troubleshooting moment, hardware fault information corresponding to each troubleshooting moment, and software fault information corresponding to each troubleshooting moment, the steps of determining the critical moment of the accident from each troubleshooting moment include: Obtaining the critical moment determination model; Generate fault sequence information based on each troubleshooting moment, hardware fault information corresponding to each troubleshooting moment, and software fault information corresponding to each troubleshooting moment; The fault sequence information is input into the critical moment determination model to obtain the accident critical moment output by the critical moment determination model.

4. The method for determining the cause of an accident according to claim 3, characterized in that: The steps to obtain a critical moment determination model include: Obtaining an initial algorithm model, wherein the initial algorithm model is configured with accident contribution weights corresponding to each fault, the initial algorithm model is used to receive multiple moments and fault information corresponding to each moment, and calculate the accident inducing index corresponding to each moment based on the configured accident contribution weights and the received fault information corresponding to each moment, and output the moment corresponding to the maximum accident inducing index among the accident inducing indexes; Obtaining a training set, the training set including multiple groups of training data, each group of training data including multiple training moments, fault information corresponding to each training moment, and key training moments manually marked from the multiple training moments; For each set of training data, multiple training moments included in the set of training data and fault information corresponding to each training moment are input into the initial algorithm model, and an output moment of the initial algorithm model is obtained; based on the error between the output moment and the key training moment included in the set of training data, the accident contribution weight corresponding to each fault in the initial algorithm model is adjusted until the initial algorithm model outputs the key training moment included in the set of training data; The initial algorithm model trained with each set of training data is used as the key moment determination model.

5. The method for determining the cause of an accident according to any one of claims 1 to 4, characterized in that: The autonomous driving vehicle is provided with a camera device; The step of determining the hardware fault existing in the autonomous driving vehicle at the time of the troubleshooting based on the hardware device detection data to obtain hardware fault information corresponding to the time of the troubleshooting includes: Extracting a first image set captured by the camera device within a fault determination period from the hardware device detection data, wherein the end time of the fault determination period is the troubleshooting time; Calculating the camera output frequency corresponding to the inspection moment according to the first image set; Determining whether the autonomous driving vehicle has a camera device failure at the time of the inspection based on the camera output frequency, and obtaining a first determination result; The hardware fault information corresponding to the troubleshooting moment is obtained according to the first judgment result.

6. The method for determining the cause of an accident according to claim 5, wherein: The step of calculating the camera output frequency corresponding to the screening moment according to the first image set includes: removing the black screen image from the first image set to obtain a second image set; The total number of images in the second image set is obtained, and the ratio of the total number of images to the duration of the fault judgment period is used as the camera output frequency corresponding to the troubleshooting moment.

7. The method for determining the cause of an accident according to any one of claims 1 to 4, characterized in that: The autonomous driving vehicle is provided with a radar, and the hardware device detection data includes a detection data packet sent by the radar; The step of determining the hardware fault existing in the autonomous driving vehicle at the time of the troubleshooting based on the hardware device detection data to obtain hardware fault information corresponding to the time of the troubleshooting includes: Obtaining the receiving time of each detection data packet, and taking each detection data packet whose receiving time is earlier than the screening time as the first target data packet; The two first target data packets with the smallest absolute value of the difference between the receiving time and the checking time are used as the second target data packets; Determining whether the autonomous driving vehicle has a radar fault at the time of the inspection based on the difference between the reception times of the two second target data packets, and obtaining a second determination result; The hardware fault information corresponding to the troubleshooting moment is obtained according to the second judgment result.

8. The method for determining the cause of an accident according to claim 7, wherein: The step of determining whether the autonomous driving vehicle has a radar fault at the time of the inspection based on the difference between the reception times of the two second target data packets, and obtaining a second determination result, includes: When the difference between the receiving times of the two second target data packets is greater than a first preset threshold, respectively calculating the sending and receiving transmission durations of the two second target data packets; Comparing the transmission and receiving durations of the two second target data packets with a second preset threshold value to obtain a comparison result; Based on the comparison result, it is determined whether there is a radar failure in the autonomous driving vehicle at the time of inspection, and the second judgment result is obtained.

9. The method for determining the cause of an accident according to any one of claims 1 to 4, characterized in that: The autonomous driving vehicle is provided with a GPS device; The step of determining the hardware fault existing in the autonomous driving vehicle at the time of the troubleshooting based on the hardware device detection data to obtain hardware fault information corresponding to the time of the troubleshooting includes: Determine a fault judgment period, the end time of the fault judgment period being the troubleshooting time; Extracting GPS log segments from the hardware device detection data; the GPS log segments are used to record the positioning results output by the GPS device during the fault judgment period and the output time of each positioning result; Calculate, based on the GPS log fragments, a first difference between every two adjacent output times on the time axis, and a second difference between the latest output time and the check time; Determining whether the GPS device of the autonomous driving vehicle has a fault at the time of the inspection based on the first difference and the second difference, and obtaining a third determination result; The hardware fault information corresponding to the troubleshooting moment is obtained according to the third judgment result.

10. The method for determining the cause of an accident according to any one of claims 1 to 4, characterized in that: The method further comprises: Output the hardware fault information corresponding to the critical moment of the accident and the software fault information corresponding to the critical moment of the accident.

11. An accident cause determination device, characterized in that: The device comprises: an accident data acquisition module, configured to, in the event of an accident involving an autonomous vehicle, use hardware device detection data and software module result data corresponding to an accident investigation period as accident data; the end time of the accident investigation period being the time of occurrence of the accident, and the autonomous vehicle operating a prediction algorithm model; a fault information acquisition module configured to, for each troubleshooting moment during the accident troubleshooting period, determine, based on the hardware device detection data, a hardware fault existing in the autonomous driving vehicle at that troubleshooting moment, thereby obtaining hardware fault information corresponding to that troubleshooting moment; and determine, based on the software module result data, a software fault existing in the autonomous driving vehicle at that troubleshooting moment, thereby obtaining software fault information corresponding to that troubleshooting moment; An accident critical moment determination module is configured to determine the accident critical moment from each troubleshooting moment based on the hardware fault information corresponding to each troubleshooting moment and the software fault information corresponding to each troubleshooting moment; the accident critical moment is used to locate the time when the cause of the fault that caused the accident occurred; Wherein, the fault information acquisition module includes: a process operation log extraction unit, configured to extract a process operation log from the software module result data; the process operation log records the process running by the autonomous driving vehicle at each inspection moment; an operating process determining unit, configured to determine the process being operated by the autonomous driving vehicle at the time of the check based on the process operating log; a fourth determining unit, configured to determine, based on the process running by the autonomous driving vehicle at the time of the check, whether the autonomous driving vehicle has a process crash fault at the time of the check, and obtain a fourth determination result; a software fault information obtaining unit, configured to obtain the software fault information corresponding to the troubleshooting moment according to the fourth judgment result; A predicted trajectory acquisition unit, configured to determine a fault judgment period and extract, from the software module result data, the obstacle predicted trajectory output by the prediction algorithm model within the fault judgment period; wherein the end time of the fault judgment period is the troubleshooting time; a fifth judgment unit, configured to judge whether the prediction algorithm model has a prediction error fault at the time of troubleshooting based on the predicted obstacle trajectory and the actual trajectory of the corresponding obstacle, and obtain a fifth judgment result; The software fault information acquisition unit is further configured to obtain the software fault information corresponding to the troubleshooting moment based on the fifth judgment result.

12. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, enable the one or more processors to perform the steps of the accident cause determination method according to any one of claims 1 to 10.

13. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, execute the steps of the accident cause determination method according to any one of claims 1 to 10.

Images