A method and system for extending autonomous driving simulation scenarios based on data injection

By adopting a simulation scenario expansion system based on data rebill in the simulation test of autonomous driving vehicles, and using rule strategies to transform and combine scene data, the problem of scenarios cannot be expanded in the existing technology is solved, and data usage efficiency and simulation value are improved , reducing the testing cost.

CN113868873BActive Publication Date: 2024-11-05CHONGQING CHANGAN AUTOMOBILE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111155776.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-30
Publication Date
2024-11-05
Estimated Expiration
2041-09-30

AI Technical Summary

Technical Problem

In the existing simulation test of autonomous driving vehicles, the data acquisition and rebilling method has problems such as high equipment customization, increased costs and inability to scale the scenario, resulting in high testing costs and low data usage efficiency.

Method used

Adopt an autonomous driving simulation scenario expansion system based on data reblocking, including sensor configuration module, data acquisition module, reblocking simulation module, fusion data module and user control module, and transform and combine the collected scene data through rules and strategies. Generate multiple different scene data to achieve expansion and modification of the scene.

Benefits of technology

It improves the efficiency and simulation value of data usage, can conduct multiple iterative testing through the same data, reduces collection and storage costs, enriches the number of scenarios, and realizes customized modification and expansion of scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113868873B_ABST
    Figure CN113868873B_ABST
Patent Text Reader

Abstract

The present invention claims protection for an automatic driving simulation scene expansion system based on data reinjection, which relates to automobile performance testing technology. Configure the sensor information required for the automatic driving of the automobile, collect all variable sets in the road test at predetermined time intervals, automatically add time series to all collected variable sets according to the collection time intervals, unify the time axis of all data in the variable set for data classification, and ensure that the data at the same time matches each other; establish a data operation strategy set, modify, expand, and split the scene according to the strategy set, fuse the data processed by the reinjection simulation module with the sensor collection information, map, and camera collection information, store different target information in a predetermined location, and form multiple different target information. The present invention expands the scene according to the needs of the test process, greatly enriches the scene content, and improves the efficiency of scene conversion.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to automobile performance testing technology, specifically, based on the scene data collected by the automobile, different data categories are added, modified, deleted, etc., and combined into new scenes to achieve the purpose of scene expansion. The scene data is back-injected to effectively enrich the automobile test scene library. Background Art

[0002] In the simulation test plan for autonomous vehicles, in order to save testing costs, some tests are replaced by simulation. Among the alternatives, the data collection and injection scheme is a better method. The existing data injection methods have the following disadvantages: the existing collection and injection test technology requires the use of special equipment in the injection process, and the degree of customized development is high. If multiple tests are required, multiple devices need to be purchased, resulting in increased testing costs and complex test process management.

[0003] Publication number: CN 111025248 A The name "A data acquisition and playback device and a data acquisition and playback system" discloses a data acquisition and playback system, in which an information processing module is connected to an acquisition interface, a host computer interface and a recharging interface, and is used to receive an acquisition interface signal through the acquisition interface in the acquisition state, and convert the acquisition interface signal into a host computer interface signal and send it to the host computer interface after converting it; in the recharging state, receive a host computer interface signal through the host computer interface, and convert the host computer interface signal into a recharging interface signal and send it to the recharging interface after converting it. The same system is used for data acquisition and recharging of the radar system, which simplifies the complexity of data acquisition and in-loop testing, and effectively reduces production costs. The scene data played back by the device is the same as the real vehicle scene, and only the same scene can be tested each time, and the scene cannot be expanded. This method requires the use of data processing circuits such as FPGA, and has certain deficiencies in convenience and ease of use.

[0004] The Chinese invention patent application with publication number CN113256976A and titled “A vehicle-road cooperative system, simulation method, on-board equipment and roadside equipment” provides a vehicle-road cooperative system, simulation method, on-board equipment and roadside equipment. According to the configuration information, data collected by the roadside equipment and data collected by the on-board equipment, the vehicle-road cooperative environment is simulated to obtain simulation results; and behavioral decisions are made based on the simulation results. The solution can solve the problem that the communication process of the traditional vehicle-road cooperative system is lengthy and noisy, and the development and verification costs are high.

[0005] The Chinese invention patent application with publication number CN112925221A and invention name “Assisted driving closed-loop test method based on data reinjection” discloses an assisted driving closed-loop test method based on data reinjection, which parses and splits the collected real vehicle data to extract scene data; the ADAS algorithm module processes, arbitrates and makes decisions based on the received data, obtains the corresponding signal from the bus system module according to the required vehicle information, and matches and connects it, and inputs the vehicle information together with the scene data into the ADAS algorithm module to form a closed-loop simulation.

[0006] The above-mentioned prior art collects data for the same segment, and the data is fixed, resulting in that the scenario cannot be modified. A segment of data can only test one scenario, and the efficiency of data use cannot be improved, resulting in data waste. Summary of the invention

[0007] The present invention is based on the existing technology of real car scene data collection, the collected scenes are inefficient, the scenes cannot be expanded, etc., resulting in data waste problems, and the collected data is processed according to certain rules and strategies. The way the present invention solves the above problems is to adopt a new data processing method that can expand the collected scenes. According to the needs, one scene can be evolved into multiple scenes, and then back-injection testing can be performed.

[0008] Specifically: The technical solution of the present invention to solve the above technical problems is to propose an autonomous driving simulation scene expansion system based on data reinjection, including: a sensor configuration module, a data acquisition module, a reinjection simulation module, a fusion data module, a target information module and a user control module. The sensor configuration module configures the sensor information required for the autonomous driving of the car; the data acquisition module collects all variable sets in the road test at a predetermined time interval; the reinjection simulation module formulates rule strategies according to the reinjection test requirements, and transforms the scene data according to the variable set. Different rule strategies correspond to different scene data, and various rule strategies can be combined to generate more scene data; the fusion data module fuses the data processed by the reinjection simulation module with the sensor collection information, map, and camera collection information; different target information is stored in a predetermined location to form multiple different target information.

[0009] Furthermore, the back-injection simulation module reads the signal sequence of the collected data, adds a time label to each signal, and obtains a time series that labels each signal. If the collected signal frequency is inconsistent with the required frequency, it is converted into the required frequency by interpolation or secondary sampling.

[0010] Furthermore, the fusion data module first unifies the coordinate system, fuses the received data with the sensor information, vehicle status information, and switch quantities, matches the perceived targets with the lane lines, and matches the lane lines with the map data. After data fusion, the targets in the original sensor information are sorted according to the rules, and the targets are divided into different areas according to the relative position relationship between the targets and the lane lines.

[0011] Furthermore, the back-injection simulation module uses time tags to synchronize signals, operates signals with time series, copies the collected single data as copies, selects the signal sequence to be operated, manually or automatically adds signal sequences, modifies corresponding values, or deletes values ​​in the signal sequence to make the sequence empty.

[0012] The reinjection simulation module classifies data according to the determined data collection format. The classified data automatically adds time series according to the collection time interval to unify the time axis of all data. When reinjecting, it ensures that the data at the same time matches each other, establishes a data operation strategy set, and modifies, expands, and splits the scenario according to the strategy set.

[0013] The present invention also proposes an automatic driving simulation scene expansion method based on data reinjection, which is characterized in that a sensor configuration module configures sensor information required for automobile automatic driving; a data acquisition module collects all variable sets in road tests at predetermined time intervals; a reinjection simulation module formulates rule strategies according to the reinjection test requirements and transforms the scene data. Different rule strategies correspond to different scene data, and various rule strategies can be combined to generate more scene data; a fusion data module fuses the data processed by the reinjection simulation module with the sensor acquisition information, map, and camera acquisition information; different target information is stored in a predetermined location to form multiple different target information.

[0014] The present invention improves the efficiency of data use and simulation value by using the scene data multiple times and using the same data for multiple rounds of iterative testing. According to the needs of the test process, the scene can be expanded to achieve scene modification. A piece of data can test multiple scenes, and the efficiency of data use is improved, which can save the cost of collection and storage for enterprises. At the same time, the scene can be customized and modified. The same collection scene can generate multiple reinjection scenes, which greatly enriches the number of scenes and improves the efficiency of scene conversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the structure of the scenario extension function module based on data reinjection;

[0016] Figure 2 Schematic diagram of the back-injection simulation module. DETAILED DESCRIPTION

[0017] The implementation of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0018] like Figure 1 The figure shows a scenario expansion system based on data injection, including: sensor configuration module, data acquisition module, injection simulation module, fusion data module and user scenario module. The sensor configuration module determines the number, type, installation location, calibration coordinate system, output information and other initial information of the acquisition vehicle sensor. The data acquisition module collects a variable set composed of relevant data in the road test, which specifically includes the detected lane lines (including lane line equations, lane line lengths, lane line colors, lane line types, etc.), target information (lateral distance, longitudinal distance, lateral speed, longitudinal speed, obstacle type movement status, etc. of the detected obstacles), vehicle status information, road sign information, speed limit information, switch quantity, etc., covering all the information that the acquisition vehicle sensor can provide. The collected relevant data is input into the injection simulation module, and a section of collected data is called an original scene.

[0019] The back injection simulation module classifies data according to the determined data acquisition variable set format. The classified variable set automatically adds time series according to the time interval of acquisition, unifies the time axis of all variable sets, and ensures that the variable sets at the same time match each other during back injection. At the same time, the back injection simulation module can filter, modify, delete and other operations on the data values ​​of the record classification, establish a specific strategy set for data operation, and modify, expand and split the scene according to the strategy set. The scene elements after expansion are consistent with those before expansion. After the scene data is processed by the back injection simulation module, it is input into the fusion data module. The fusion data module fuses the received data with the sensor information, vehicle status information, and switch quantity. The fusion method can be to match the perceived target with the lane line, and the lane line with the map data. After data fusion, the targets in the original sensor information are sorted according to certain rules, such as the target No. 1 directly in front of the vehicle, the target No. 2 in the left lane, and the target No. 3 in the right lane. The fused information is encapsulated into a user scenario library by the user scenario module for user use.

[0020] The sensor configuration module sets the sensor type, quantity, installation location, information output by each sensor, information unit, calibration coordinate system, time series and other information according to the various sensors required by the autonomous vehicle (including cameras, radars and other sensors installed at different locations on the vehicle body and collecting different information). These information together constitute the sensor information set, and the information set output by the sensor configuration module has the same properties as the information set of the back-injection simulation module.

[0021] The data acquisition module collects relevant data from road tests and samples and records the data output by sensors during the test at a predetermined frequency. The collected information includes sensor information sets, vehicle status information, switch quantities, etc., covering all information that can be provided by vehicle sensors.

[0022] Each type of information collects data according to its own inherent frequency, and the collected data is processed (such as Matlab / Simulink). Arrange the data of each target (including longitudinal speed, longitudinal distance, lateral speed, lateral distance, target category, target status, etc.) and each lane line in column order, with the first column being the acquisition time, and the subsequent data being arranged in sequence. Each column of data is called a signal.

[0023] like Figure 2 The figure shows the schematic diagram of the reinjection simulation module. The data processed by the data processing module enters the reinjection simulation module, reads the signal sequence of the collected data, and marks each signal with a time tag to obtain the time series of each signal, which becomes a signal with a time series; based on the required frequency (frequency = 1 / collection interval time), if the collected signal frequency is inconsistent with the required frequency, interpolation or secondary sampling is used to convert it into the required frequency.

[0024] Use time tags to synchronize the signals required by the injection module to solve the problem of signal disorder. Perform operations such as filtering, adding, modifying, and deleting signals with time series, formulate scenario rule strategies, copy the collected single data as a copy, select the signal sequence that needs to be operated, manually or automatically add the signal sequence, modify the corresponding value, or delete the original value in the signal sequence to make the sequence empty. You can also add noise to the selected sequence to make the sequence value more realistic.

[0025] For example, you can filter out the distance in the target information attributes, superimpose various noises on the distance attribute signal, and let the distance information fluctuate according to the noise. You can also filter out a lane line, delete the lane line, and modify the type, length, and color of the lane line. You can also filter out all the information attributes of a target and modify a certain attribute of the target, such as changing the target to an animal and a car to a truck. You can also delete all the attributes of the target, and after deletion, the target information will disappear.

[0026] According to the requirements of the back-injection test, rule strategies are formulated to transform the scenario data. Different rule strategies will produce different scenario data. Various rule strategies can be combined to generate more scenario data.

[0027] The data after the back-injection simulation is input into the fusion data module, which matches the target information sequence, such as the longitudinal distance and lateral distance of the target with the specific lane line, and matches the lane line with the map. The specific fusion method is to first unify the coordinate system and compare the lateral distance of the target with the coordinates of the lane line. If the target is directly in front of the lane and is between the two nearest lane lines, then the target is defined as target No. 1, and all information under the target is labeled 1. If the target is in front of the left of the lane and is on the left side of the left lane line, then the target is defined as target No. 2, and all information under the target is labeled 2, and so on. Finally, according to the relative position relationship between the target and the lane line, the target is divided into different areas. The fusion information module also calculates parameters such as the radius of curvature based on the lane line information to control the driving path of the car.

[0028] The user scenario module is simply an interface program that builds an information interface framework based on the backend usage requirements. Different information is placed in the specified interface position, and all interfaces in the requirements are placed in the specified position of the framework, which is encapsulated into a scenario. The entire process, from the determination of the front-end sensor solution to the user scenario module, has one-way data flow and open-loop testing.

Claims

1. An autonomous driving simulation scenario expansion system based on data injection, characterized in that: include: Sensor configuration module, data acquisition module, back-injection simulation module, fusion data module, target information module and user control module. The sensor configuration module configures the sensor information required for the car's automatic driving; the data acquisition module collects all variable sets in the road test at predetermined time intervals; the back-injection simulation module formulates rule strategies according to the back-injection test requirements, transforms the scene data according to the variable set, and different rule strategies correspond to different scene data. Various rule strategies are combined to generate more scene data. Specifically, the back-injection simulation module reads the signal sequence of the collected variable set, marks each signal with a time tag, and obtains the time series that marks each signal. If the collected variable set signal frequency If it is inconsistent with the required frequency, it is converted into the required frequency by interpolation or secondary sampling. The fusion data module fuses the variable set processed by the injection simulation module with the sensor collection information, map, and camera collection information, and divides the target into different areas according to the relative position relationship between the target and the lane line. Specifically, the fusion data module first unifies the coordinate system, fuses the received variable set with the sensor information, vehicle status information, and switch quantity, matches the perceived target with the lane line, and matches the lane line with the map data. After data fusion, the targets in the original sensor information are sorted according to the rules, and the targets are divided into different areas according to the relative position relationship between the target and the lane line.

2. The system according to claim 1, characterized in that The back-injection simulation module uses time tags to synchronize signals, operates signals with time series, copies the collected single data, selects the signal sequence to be operated, manually or automatically adds signal sequences, modifies corresponding values, or deletes values ​​in the signal sequence to make the sequence empty.

3. The system according to claim 1, characterized in that The reinjection simulation module classifies data according to the determined data collection format. The classified data automatically adds time series according to the collection time interval to unify the time axis of all data. When reinjecting, it ensures that the data at the same time matches each other, establishes a data operation strategy set, and modifies, expands, and splits the scenario according to the strategy set.

4. A method for expanding an autonomous driving simulation scenario based on data injection, characterized in that: The sensor configuration module configures the sensor information required for the car's automatic driving; the data acquisition module collects all the variable sets in the road test at a predetermined time interval; the back-injection simulation module formulates rules and strategies according to the back-injection test requirements, transforms the scene data according to the variable set, and different rules and strategies correspond to different scene data. Various rules and strategies are combined to generate more scene data. Specifically, the back-injection simulation module reads the signal sequence of the collected variable set, marks each signal with a time tag, and obtains a time series that marks each signal. If the collected signal frequency is inconsistent with the required frequency, it is converted into the required frequency by interpolation or secondary sampling; the fusion data module fuses the data processed by the back-injection simulation module with the sensor collection information, map, and camera collection information; according to the relative position relationship between the target and the lane line, the target is divided into different areas. Specifically, the fusion data module first unifies the coordinate system, fuses the received data with the sensor information, vehicle status information, and switch quantity, matches the perceived target with the lane line, and matches the lane line with the map data. After data fusion, the target in the original sensor information is sorted according to the rules, and the target is divided into different areas according to the relative position relationship between the target and the lane line.

5. The method for expanding the automatic driving simulation scene according to claim 4, characterized in that: The back-injection simulation module uses time tags to synchronize signals, operates signals with time series, copies the collected single data, selects the signal sequence to be operated, manually or automatically adds signal sequences, modifies corresponding values, or deletes values ​​in the signal sequence to make the sequence empty.

6. The method for expanding the automatic driving simulation scene according to claim 4, characterized in that: The reinjection simulation module classifies data according to the determined data collection format. The classified data automatically adds time series according to the collection time interval to unify the time axis of all data. When reinjecting, it ensures that the data at the same time matches each other, establishes a data operation strategy set, and modifies, expands, and splits the scenario according to the strategy set.