Special simulation test method for automatic driving perception algorithm

Abstract

The invention discloses a special simulation test method for an automatic driving perception algorithm, and the method comprises the steps: building a simulation vehicle sensor model and a simulation scene, and operating a simulator and outputting a simulation sensor signal in combination with the sensor model and the simulation scene; performing off-line training and on-line testing by using the simulation sensor signal, and respectively obtaining a model parameter weight for updating the automatic driving perception algorithm and an effect evaluation result of the simulation testing; optimizing a sensing algorithm by using the weight information, and adjusting the configuration and installation position of each sensor on the real vehicle based on an effect evaluation result obtained by the output of the optimized sensing algorithm; and based on the adjusted real vehicle sensor deployment, reconstructing a sensor model and carrying out simulation test again. According to the invention, a large amount of test data can be obtained at low cost, various uncontrollable real scenes can be reproduced, the configuration and the installation position of the sensor can be changed according to the targeted requirements of a sensing algorithm, and the refitting cost and the time cost of a test vehicle are greatly saved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a special simulation test method for an autonomous driving perception algorithm. Background Art

[0002] The autonomous driving perception algorithm requires a large amount of scene data for algorithm training, so as to improve the performance of the algorithm. The autonomous driving perception algorithm concerned by the present invention is greatly affected by the environment, so it is necessary to test it in various scenes to verify the algorithm.

[0003] Specifically, the simulation test of autonomous driving algorithms is mostly to test the effect of the complete autonomous driving function (including but not limited to perception, planning, control, etc.). The test effect of the overall test will be affected by multiple factors, and it needs to be split up to troubleshoot the problem. However, there are few reliable solutions for simulation verification testing of individual perception algorithms in the industry. However, the effect of a single perception algorithm will affect the operation of other modules, so it is necessary to conduct a separate verification test on the perception algorithm.

[0004] In summary, the following obstacles currently exist in the industry that prevent a single algorithm module from being tested independently:

[0005] (1) Autonomous driving perception algorithms have a large demand for scenarios and data;

[0006] (2) If the autonomous driving perception algorithm is tested on a real vehicle, not only are the scenarios limited, but the testing cost is also very high;

[0007] (3) Currently, most of the existing solutions for autonomous driving simulation testing are closed-loop tests of simulation and vehicle control, which cannot inspire practitioners to verify the effectiveness of perception algorithms solely through simulation testing. Summary of the invention

[0008] In view of the above, the present invention aims to provide a dedicated simulation test method for an autonomous driving perception algorithm to solve problems such as insufficient data and limited scenarios when training or testing the perception algorithm.

[0009] The technical solution adopted by the present invention is as follows:

[0010] The present invention provides a dedicated simulation test method for an autonomous driving perception algorithm, which includes:

[0011] Build a simulated vehicle sensor model;

[0012] Build a simulation scene;

[0013] Combine the sensor model and simulation scenario, run the simulator and output the simulated sensor signal;

[0014] Using the simulated sensor signal to perform offline training and online testing, respectively obtaining weight information for updating the autonomous driving perception algorithm and an effect evaluation result of the simulation test;

[0015] Use the weight information to optimize the perception algorithm, and adjust the configuration and installation position of each sensor on the actual vehicle based on the effect evaluation results obtained from the output of the optimized perception algorithm;

[0016] Based on the adjusted real vehicle sensor deployment, the sensor model is rebuilt and simulation tests are performed again.

[0017] In at least one possible implementation, an initial sensor model is built in advance according to the sensor installation position and hardware configuration of the actual vehicle to be tested.

[0018] In at least one possible implementation, the offline training includes:

[0019] Saving the simulated sensor signal;

[0020] Use the saved signal data to build a training and verification dataset dedicated to the perception algorithm;

[0021] Offline training of the perception algorithm is performed based on the training verification data set to obtain an optimized and updated algorithm weight file.

[0022] In at least one possible implementation, the online test includes:

[0023] transmitting the simulated sensor signal in real time;

[0024] Connecting the simulated sensor signal transmitted in real time to the autonomous driving perception algorithm for real-time detection, and obtaining the real-time detection result of the perception algorithm;

[0025] The effect of the perception algorithm simulation test is judged according to the real-time detection results.

[0026] In at least one possible implementation, the simulation test method further includes: outputting and displaying the effect of the perception algorithm simulation test in real time on the simulation test platform in an intuitive manner.

[0027] In at least one possible implementation, a reusable simulation scenario is built based on a specific test scenario of an autonomous driving perception algorithm.

[0028] Compared with the prior art, the main design concept of the present invention is to build a simulated vehicle sensor model and a simulation scene, combine the sensor model and the simulation scene, run the simulator and output the simulated sensor signal; use the simulated sensor signal for offline training and online testing, respectively obtain the model parameter weights for updating the autonomous driving perception algorithm and the effect evaluation results of the simulation test; use the weight information to optimize the perception algorithm, and adjust the configuration and installation position of each sensor on the actual vehicle based on the effect evaluation results obtained from the output of the optimized perception algorithm; based on the adjusted actual vehicle sensor deployment, rebuild the sensor model and perform simulation tests again. The present invention can obtain a large amount of test data at a low cost and the scene covers a comprehensive range, and can reproduce various uncontrollable real scenes, and thus can change the sensor configuration and installation position according to the targeted needs of the perception algorithm, etc., greatly saving the cost and time cost of modifying the test vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described below with reference to the accompanying drawings, in which:

[0030] Figure 1 A schematic diagram of a dedicated simulation test method for an autonomous driving perception algorithm provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0032] The present invention proposes an embodiment of a dedicated simulation test method for an autonomous driving perception algorithm. Specifically, Figure 1 shown, including:

[0033] Step S1, building a simulated vehicle sensor model;

[0034] Specifically, the sensor model can be built in advance according to the sensor installation position and hardware configuration of the actual vehicle being tested. The sensors here cover various types of perception devices required for autonomous driving.

[0035] Step S2, building a simulation scene;

[0036] Specifically, the simulation scenario can be built according to the specific test scenario requirements of the perception algorithm, such as but not limited to weather, obstacles, road conditions, etc.

[0037] Step S3, running the simulator to output a simulated sensor signal;

[0038] The simulated sensor signals include, but are not limited to: image data collected by a simulated camera sensor, point cloud data collected by a simulated radar sensor, position information transmitted by a simulated positioning sensor, and the like.

[0039] Step S4, using the simulated sensor signal to perform offline training and online testing, respectively obtaining weight information for updating the autonomous driving perception algorithm and an effect evaluation result of the simulation test;

[0040] Step S5: Optimize the perception algorithm using the weight information, and adjust the configuration and installation position of each sensor on the actual vehicle based on the effect evaluation result obtained from the output of the optimized perception algorithm.

[0041] Wherein, the offline training includes:

[0042] The simulated sensor signal is saved (for example, the image, point cloud, coordinates and other data of the simulated sensor transmitted after running the simulator can be saved locally, and if there is true value data of target detection, the true value data can also be saved locally);

[0043] Use the saved signal data to build a training and verification data set dedicated to the perception algorithm (in actual operation, sample sets can be produced through batch processing);

[0044] Offline training of the perception algorithm is performed based on the training verification data set to obtain an optimized and updated algorithm weight file.

[0045] Furthermore, the online test includes:

[0046] The simulated sensor signal (such as image, point cloud, coordinates and other data) can be transmitted in real time through UDP communication or TCP communication;

[0047] Connecting the simulated sensor signal transmitted in real time to the autonomous driving perception algorithm for real-time detection, and obtaining the real-time detection result of the perception algorithm;

[0048] The effect of the perception algorithm simulation test is judged according to the real-time detection results.

[0049] It can also be supplemented from this concept that the simulation test method also includes: outputting and displaying the effect of the perception algorithm simulation test in real time in an intuitive manner on the simulation test platform.

[0050] Finally, it should be emphasized that the present invention uses simulated perception information to train and optimize the actual perception algorithm, and substitutes the simulated perception information into the actual perception algorithm; on the other hand, the output of the actual perception algorithm is used as the key basis for evaluating the aforementioned simulation test process, and the subsequent adjustment and improvement of the actual vehicle perception device is based on the evaluation content mentioned here, rather than directly using the algorithm output results or simulation test results in the conventional way. Figure 1 As shown in step S6, the aforementioned simulated vehicle sensor model is rebuilt based on the sensor configuration and position adjusted by the actual vehicle, and then simulation testing and actual vehicle tuning are performed in the above manner, thereby obtaining rich test data, improving the test process and results, and ultimately making the deployment of the actual vehicle sensors relatively optimal.

[0051] In summary, the main design concept of the present invention is to build a simulated vehicle sensor model and a simulation scene, combine the sensor model and the simulation scene, run the simulator and output the simulated sensor signal; use the simulated sensor signal for offline training and online testing, respectively obtain the model parameter weights for updating the autonomous driving perception algorithm and the effect evaluation results of the simulation test; use the weight information to optimize the perception algorithm, and adjust the configuration and installation position of each sensor on the actual vehicle based on the effect evaluation results obtained from the output of the optimized perception algorithm; based on the adjusted actual vehicle sensor deployment, rebuild the sensor model and perform simulation tests again. The present invention can obtain a large amount of test data at a low cost and the scene covers a comprehensive range, and can reproduce various uncontrollable real scenes, and thus can change the sensor configuration and installation position according to the targeted needs of the perception algorithm, etc., greatly saving the cost and time cost of modifying the test vehicle.

[0052] If the expressions expressing orientation are mentioned in the embodiments of the present invention, they are relative concepts based on the embodiments. In addition, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, wherein a, b, c can be single or multiple.

[0053] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the present invention is not limited to the scope of implementation shown in the drawings, and all changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the specification and drawings, should be within the protection scope of the present invention.

Claims

1. A special simulation test method for autonomous driving perception algorithm, characterized in that: include: Build a simulated vehicle sensor model; Build a simulation scene; Combine the sensor model and simulation scenario, run the simulator and output the simulated sensor signal; Using the simulated sensor signal to perform offline training and online testing, respectively obtaining weight information for updating the autonomous driving perception algorithm and an effect evaluation result of the simulation test; Use the weight information to optimize the perception algorithm, and adjust the configuration and installation position of each sensor on the actual vehicle based on the effect evaluation results obtained from the output of the optimized perception algorithm; Based on the adjusted real vehicle sensor deployment, the sensor model is rebuilt and simulation tests are performed again.

2. The dedicated simulation test method for the autonomous driving perception algorithm according to claim 1, characterized in that: Build the initial sensor model in advance according to the sensor installation position and hardware configuration of the actual vehicle being tested.

3. The dedicated simulation test method for the autonomous driving perception algorithm according to claim 1, characterized in that: The offline training includes: Saving the simulated sensor signal; Use the saved signal data to build a training and verification dataset dedicated to the perception algorithm; Offline training of the perception algorithm is performed based on the training verification data set to obtain an optimized and updated algorithm weight file.

4. The dedicated simulation test method for the autonomous driving perception algorithm according to claim 1, characterized in that: The online test includes: transmitting the simulated sensor signal in real time; Connecting the simulated sensor signal transmitted in real time to the autonomous driving perception algorithm for real-time detection, and obtaining the real-time detection result of the perception algorithm; The effect of the perception algorithm simulation test is judged according to the real-time detection results.

5. The dedicated simulation test method for the autonomous driving perception algorithm according to claim 1, characterized in that: The simulation test method also includes: outputting and displaying the effect of the perception algorithm simulation test in real time in an intuitive manner on the simulation test platform.

6. The dedicated simulation test method for the autonomous driving perception algorithm according to any one of claims 1 to 5, characterized in that: The reusable simulation scenario is built according to the preset test scenario of the autonomous driving perception algorithm.

Images