Automatic driving virtual simulation test platform and test method thereof

Abstract

The invention discloses an automatic driving virtual simulation test platform and a test method thereof, and relates to the technical field of virtual simulation driving, and the test platform comprises a virtual simulation module which is used for constructing a high-precision three-dimensional virtual scene and simulating sensor data; the automatic driving trolley is used for executing an automatic driving algorithm in a physical environment and collecting real sensor data; the road simulation system is used for simulating various road working conditions and forming a data closed loop with the virtual simulation module and the desktop-level automatic driving trolley; and the data interaction module is used for realizing real-time data transmission and synchronization among the virtual simulation module, the desktop-level automatic driving trolley and the road simulation system. According to the invention, by integrating a virtual high-fidelity scene library, a real sensor, trolley hardware and a flexibly adjustable rotating hub testing device, multi-level closed-loop verification from simulation to a real vehicle can be realized, and meanwhile, acquisition and playback of sensor data and multi-modal data fusion in a real environment are supported.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of virtual simulation driving, in particular relates to an automatic driving virtual simulation test platform and a test method thereof. BACKGROUND

[0002] With the development of intelligent driving technology, vehicles need to perform a large number of function verification and performance testing in various complex environments. However, traditional road real vehicle testing has problems such as high cost, long cycle, high safety risk, limited scene coverage, etc., and it is difficult to meet the full verification needs of intelligent driving systems under extreme working conditions and diversified road conditions.

[0003] Virtual simulation testing technology gradually becomes an important means of automatic driving algorithm verification due to its strong controllability, flexible and rich scenes, and large-scale parallel testing advantages. However, existing pure software simulation environments often lack close integration with real vehicle hardware, making it difficult to verify the real effect of software and hardware integration in a timely manner. In addition, although some desktop-level car platforms have certain real vehicle verification capabilities, they lack road condition simulation and cannot cover complex working conditions such as various speeds, slopes, and road resistance.

[0004] A turntable test platform is a test device commonly used for dynamic performance and component durability, with advantages such as simulating different road profiles and reproducing vehicle driving resistance characteristics. However, existing turntable platforms are often designed for whole vehicle or powertrain testing, which are large in size and high in cost, and are not suitable for rapid verification on small desktop-level automatic driving cars in the teaching or research and development stage.

[0005] Therefore, how to organically integrate a high-fidelity virtual simulation environment, a desktop-level automatic driving car that can actually run, and a flexible and adjustable road condition simulation device (turntable test platform) to form a low-cost, scalable, and repeatable closed-loop test system to better serve algorithm verification, hardware-in-the-loop testing, and teaching and research of intelligent driving has become a problem to be solved in the field. SUMMARY

[0006] The present application aims to provide an automatic driving virtual simulation test platform and a test method thereof, which realizes a low-cost, modular, high-precision, multi-scene, multi-condition, and repeatable automatic driving test platform through a closed-loop verification system combining simulation environment, desktop-level automatic driving car, and adjustable turntable test platform, solving the problems of virtual simulation and real vehicle verification fragmentation, insufficient road condition simulation of desktop-level test platforms, high cost and large size of traditional turntable platforms, and poor multi-source data playback capability in the prior art.

[0007] To solve the above technical problems, the present application is realized by the following technical scheme:

[0008] The application is an automatic driving virtual simulation test platform, which comprises:

[0009] a virtual simulation module for constructing a high-precision three-dimensional virtual scene and simulating sensor data;

[0010] an automatic driving car for executing an automatic driving algorithm in a physical environment and collecting real sensor data;

[0011] a road simulation system for simulating various road conditions and forming a data closed loop with the virtual simulation module and the desktop automatic driving car;

[0012] a data interaction module for realizing real-time data transmission and synchronization between the virtual simulation module, the desktop automatic driving car and the road simulation system.

[0013] Further, the virtual simulation module comprises:

[0014] a scene modeling unit for generating a high-precision road topology map based on RoadRunner and constructing three-dimensional scene elements including dynamic obstacles and traffic signs in combination with Blender;

[0015] a sensor simulation unit for simulating perception data of an RGB-D camera, a laser radar and a millimeter wave radar through a Carla engine and supporting spatio-temporal alignment with real sensor data;

[0016] a dynamic environment simulation unit for adjusting weather, illumination and traffic flow density in the virtual scene and constructing an extreme condition test environment.

[0017] Further, the automatic driving car comprises:

[0018] a car body structure integrating an RGB-D camera, a 16-line laser radar simulation ToF sensor and a millimeter wave radar array to realize 360° environmental perception;

[0019] a processing unit adopting an NVIDIA Jetson Nano edge computing module to run a YOLOv7 target detection algorithm and a BEV perspective conversion model with a computing power of 21TOPS;

[0020] a control unit based on an STM32 embedded controller to drive four-wheel independent motors and steering rudders through PWM signals and support physical interaction with the road simulation system.

[0021] Further, the road simulation system comprises:

[0022] a double rotary hub mechanism driven by a brushless direct current motor for fixing the desktop automatic driving car and simulating vehicle speed;

[0023] Inclination adjustment module, 0%-30% slope road inclination simulation is realized by electric push rod;

[0024] Wheel track and wheelbase adjustment mechanism, using screw rod slide rail and movable support structure, supporting wheel track ±0.1mm precision adjustment and wheelbase quick adaptation;

[0025] Data acquisition unit, integrating speed sensor, torque sensor and six-axis force sensor, real-time acquisition of vehicle operation data and feedback to virtual simulation module.

[0026] Further, the data interaction module is based on ROS2 and OPCUA protocol, which realizes:

[0027] Bidirectional injection of sensor data between virtual simulation module and autonomous vehicle (virtual data mapping to physical vehicle, real data feeding back to simulation scene);

[0028] Dynamics data synchronization between road simulation system and virtual simulation module, the road simulation system transmits physical resistance data to the virtual simulation module, and the virtual simulation module feeds back the virtual road condition parameters to the road simulation system for dynamic adjustment of working conditions.

[0029] A test method of an automatic driving virtual simulation test platform, the test method comprising the following steps:

[0030] Step S1: Constructing scene library, generating high-precision road map by RoadRunner, combining with Blender modeling dynamic obstacles, configuring virtual test scene containing extreme weather and complex traffic flow in Carla engine;

[0031] Step S2: Burning automatic driving algorithm program to Jetson Nano processing unit of desktop automatic driving car, the algorithm program integrates YOLOv7 target detection, BEV perspective conversion and path planning module based on Dijkstra algorithm;

[0032] Step S3: Linkage test, fixing the desktop automatic driving car on the road simulation system, adjusting the slope, speed and road resistance through the road simulation system to simulate flat road, slope and high speed, and injecting virtual scene data into the physical car through the data interaction middleware and feeding back the real sensor data to the virtual scene;

[0033] Step S4: Perform multi-scene test, including regulation scene (traffic rule compliance verification), comprehensive scene (intersection avoidance, following car, etc.) and extreme scene (emergency obstacle avoidance, severe weather), record the car trajectory, control command and sensor data;

[0034] Step S5: data comparison and analysis, compare the virtual scene simulation results with the physical car test data, and optimize the automatic driving algorithm;

[0035] Step S6: iterative optimization, repeat steps S3 to S5 until the algorithm meets the preset performance indicators in both virtual and physical environments.

[0036] Further, in step S1, the scene library defines the road topology based on the Open Drive format, the OpenScenario format describes the traffic event logic, and supports KITTI, Waymo real dataset driven scene reproduction.

[0037] Further, in step S3, the road simulation system controls the motor torque and roller speed through the PID adjustment module, simulates 0-150km / h acceleration and deceleration conditions and 30% slope resistance, and the response delay is ≤5ms.

[0038] Further, in step S5, the performance indicators include safety (collision rate), comfort (acceleration fluctuation), task completion efficiency (path deviation), and energy efficiency (energy consumption simulation value), and the digital twin scene visualized by Unreal Engine can evaluate the results.

[0039] The present application has the following advantages:

[0040] 1. The present application can simulate the test process by adjusting the slope, speed and road resistance of the hub platform and the virtual multi-scene library, so that the actual driving conditions can be truly restored and repeated, ensuring test consistency and accuracy.

[0041] 2. The present application can reduce the cost of equipment purchase and maintenance by designing a desktop size combined with a modular hardware architecture, and can be quickly assembled, disassembled, and supports flexible multi-scene switching, improving the efficiency of research and development and teaching verification.

[0042] 3. The present application can realize the joint verification of virtual simulation scene and real sensor car, and realize the multi-level verification of perception, planning, control and other algorithms by combining hub simulation, so as to ensure the portability and real car landing performance of the algorithm.

[0043] 4. The present application can trace the behavior under extreme conditions by playing back and reproducing the data collected by the actual sensor in the simulation environment, which helps to find potential defects and quickly iterate.

[0044] 5. The present application can adapt to the teaching and scientific research needs of universities and research institutes for automatic driving test, has the characteristics of low threshold, high expansibility and high safety, and can cover a variety of complex traffic environment tests, facilitating personnel training and experimental research.

[0045] Of course, implementing any product of the application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed for the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0047] Figure 1 Fig. 1 is a structural schematic diagram of an automatic driving virtual simulation test platform according to the present application;

[0048] Figure 2 Fig. 2 is a flowchart of a test method of an automatic driving virtual simulation test platform according to the present application;

[0049] Figure 3 Fig. 3 is a schematic diagram of a road simulation system-tumbler model according to the present application;

[0050] Figure 4 Fig. 4 is a schematic diagram of basic operation principles of a frame according to the present application;

[0051] Figure 5 Fig. 5 is a schematic diagram of a simulation system and a tumbler frame according to the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0053] Please refer to Figure 1 Fig. 1 shows that the present application is an automatic driving virtual simulation test platform, and the test platform comprises:

[0054] a virtual simulation module, configured to build a high-precision three-dimensional virtual scene and simulate sensor data;

[0055] an automatic driving car, configured to execute an automatic driving algorithm in a physical environment and collect real sensor data;

[0056] a road simulation system, configured to simulate various road conditions and form a data closed loop with the virtual simulation module and the desktop automatic driving car;

[0057] A data interaction module is configured to realize real-time data transmission and synchronization between the virtual simulation module, the desktop automatic driving vehicle, and the road simulation system.

[0058] The virtual simulation module comprises:

[0059] A scene modeling unit is configured to generate a high-precision road topology map based on RoadRunner and construct three-dimensional scene elements including dynamic obstacles and traffic signs by using Blender;

[0060] A sensor simulation unit is configured to simulate perception data of an RGB-D camera, a laser radar, and a millimeter wave radar by using a Carla engine and support spatio-temporal alignment with real sensor data;

[0061] A dynamic environment simulation unit is configured to adjust weather, illumination, and traffic flow density in a virtual scene and construct an extreme working condition test environment.

[0062] The automatic driving vehicle comprises:

[0063] A vehicle body structure is integrated with an RGB-D camera, a 16-line laser radar simulation ToF sensor, and a millimeter wave radar array to realize 360° environmental perception;

[0064] A processing unit is configured to use an NVIDIA Jetson Nano edge computing module, run a YOLOv7 target detection algorithm and a BEV perspective conversion model, and has a computing power of 21 TOPS;

[0065] A control unit is configured to use an STM32 embedded controller, drive four-wheel independent motors and a steering engine through a PWM signal, and support physical interaction with the road simulation system.

[0066] The road simulation system comprises:

[0067] A double-gear hub mechanism is driven by a brushless direct current motor and is configured to fix the desktop automatic driving vehicle and simulate vehicle speed;

[0068] An inclination adjustment module is configured to realize road inclination simulation of 0-30% slope through an electric push rod;

[0069] A wheel track and wheelbase adjustment mechanism is configured to use a lead screw sliding rail and a movable support structure, support wheel track adjustment with an accuracy of ±0.1 mm, and support quick adaptation of wheelbase;

[0070] A data acquisition unit is integrated with a rotating speed sensor, a torque sensor, and a six-axis force sensor, is configured to collect vehicle operation data in real time, and feeds back the vehicle operation data to the virtual simulation module.

[0071] The data interaction module is based on ROS2 and OPCUA protocols and is configured to realize:

[0072] The sensor data between the virtual simulation module and the autonomous driving car is bidirectionally injected (virtual data is mapped to the physical car, and real data is fed back to the simulation scene);

[0073] The dynamics data of the road simulation system and the virtual simulation module are synchronized, the physical resistance data is transmitted from the road simulation system to the virtual simulation module, and the virtual road condition parameters are fed back to the road simulation system to dynamically adjust the working condition.

[0074] Please refer to Figure 2 The present application is a kind of automatic driving virtual simulation test platform test method, test method includes the following steps:

[0075] Step S1: Construct a scene library, generate a high-precision road map through RoadRunner, combine dynamic obstacles modeled by Blender, and configure a virtual test scene containing extreme weather and complex traffic flow in the Carla engine;

[0076] Step S2: Burn the autonomous driving algorithm program to the Jetson Nano processing unit of the desktop autonomous driving car, and the algorithm program integrates the YOLOv7 target detection, BEV perspective conversion and path planning module based on Dijkstra algorithm;

[0077] Step S3: Linkage test, fix the desktop autonomous driving car on the road simulation system, adjust the slope, speed and road resistance through the road simulation system, simulate flat road, slope, high speed and other working conditions, and inject virtual scene data into the physical car through the data interaction middleware and feed back the real sensor data to the virtual scene;

[0078] Step S4: Perform multi-scene testing, including regulatory scenarios (traffic rule compliance verification), comprehensive scenarios (intersection avoidance, following, etc.), and extreme scenarios (emergency obstacle avoidance, severe weather), record the car motion trajectory, control command and sensor data;

[0079] Step S5: Data comparison and analysis, compare the virtual scene simulation results with the physical car test data, and optimize the autonomous driving algorithm;

[0080] Step S6: Iterative optimization, repeat steps S3 to S5 until the algorithm meets the preset performance indicators in both virtual and physical environments.

[0081] In step S1, the scene library defines the road topology based on the Open Drive format, the Open Scenario format describes the traffic event logic, and supports KITTI and Waymo real data set driven scene reproduction.

[0082] In step S3, the road simulation system controls the motor torque and the roller speed through the PID adjustment module to simulate 0-150 km / h acceleration and deceleration conditions and 30% slope resistance, with a response delay of ≤5 ms.

[0083] In step S5, the performance indicators include safety (collision rate), comfort (acceleration fluctuation), task completion efficiency (path deviation), and energy efficiency (energy consumption simulation value), and the evaluation results are visualized through the digital twin scene rendered by Unreal Engine.

[0084] One specific application of this embodiment is:

[0085] 1. Road simulation system:

[0086] 1) Turntable test platform

[0087] Please refer to Figures 3-4 The core of the road simulation system is the turntable test platform, and the core is the roller device. The drive wheels of the vehicle or the measured components are in contact with the roller, which is driven by the motor or the drive wheels of the vehicle to rotate, simulating different road conditions (such as flat road, slope, high speed, etc.). The platform is equipped with various sensors (such as speed sensor, torque sensor, force sensor, etc.), which can collect real-time data such as speed, torque, and force. After data processing system analysis, the performance of the measured object is evaluated. By adjusting the motor torque to simulate the climbing resistance, and by changing the roller speed to simulate different vehicle speeds, the running state of the components or vehicles under various conditions can be comprehensively detected.

[0088] The platform base is stably supported by 3D printing materials and has reserved interfaces to connect the upper computer; the inclined platform is driven by a motor and an electric push rod to simulate road slope; the wheel track adjustment mechanism is a lead screw rail structure, which accurately adjusts the wheel track and has a locking device to ensure stability; the wheelbase adjustment mechanism is a movable support structure that supports quick adjustment of the wheelbase and is equipped with a scale for easy and accurate setting; the control system includes motor drivers, sensors, control panels, etc., responsible for motion control and data acquisition and communicates with the upper computer to realize automatic testing; the double-turntable mechanism is driven by a motor, which uses a double-turntable fixed trolley to measure speed and relies on a rudder to steer.

[0089] 2) Module integration

[0090] Inclined platform electric push rod: micro electric push rod 24V, push and pull 30N, 24mm / s stroke 100mm, hole pitch 138mm, ensuring that the slope can be quickly adjusted while being supported Wheel track adjustment mechanism: precision lead screw motor is selected, matched with closed-loop control system, ensuring ±0.1mm adjustment accuracy. Cylinder drive motor: brushless DC motor, speed range 0-5000RPM, supporting high-speed response in dynamic scenarios.

[0091] Modular design with SolidWorks, mechanical components into independent modules (like base, tilt platform, etc.), easy to quickly assemble and maintain; in terms of heat dissipation and protection, the motor is equipped with heat dissipation fins and IP54 protective shell to ensure long-term high-load operation stability; dynamic balance calibration uses SolidWorks simulation to analyze the moment of inertia of rotating parts to optimize bearing layout and reduce the impact of vibration on test accuracy; perception module integrates cameras, lidar and other sensors to simulate data and test the environmental understanding and recognition capabilities of perception algorithms; decision module receives dynamic traffic scenarios from simulation platform to test path planning and traffic rule compliance capabilities; control module is tested in a closed loop to verify vehicle control behavior such as acceleration, steering, braking, etc. in dynamic environments.

[0092] 2. Scene library simulation modeling:

[0093] In this embodiment, the simulation construction of the scene library mainly relies on the CARLA and LGSVL two open-source autonomous driving simulation platforms, which have high-fidelity environment modeling capabilities and flexible API extension interfaces, and can simulate cameras, lidars, millimeter wave radars, GPS, IMU and other sensors, providing strong support for algorithm testing and verification of autonomous driving systems. Compared with traditional static simulation tools, these platforms not only support intelligent behavior modeling of dynamic traffic participants (such as pedestrians, vehicles, bicycles, etc.), but also simulate different weather, lighting and complex road conditions to enhance the adaptability of the system in complex environments. At the same time, this embodiment also combines commercial simulation tools such as Prescan, which has the advantage of more accurate physical simulation capabilities and more complete vehicle dynamics models, suitable for high-precision control algorithm testing and system-level performance evaluation.

[0094] To realize data synchronization and module linkage between various tools, an intermediate communication bridge is built through ROS (Robot Operating System), such as carla_ros_bridge, which enables deep integration of perception, control and planning modules in the CARLA and ROS ecosystems. The construction of the simulation scene adopts the internationally recognized OpenDrive, OpenCRG and OpenScenario standards: OpenDrive defines the topological structure and geometric information of the road network, OpenCRG describes the three-dimensional texture and micro properties of the road surface, and OpenScenario supports abstract modeling of traffic events and behavior logic. Through natural language description of functional scenarios (such as "avoiding pedestrians on urban roads in rainy weather"), the logical scene parameters are quantified, such as the number of lanes, driving speed, signal light status, and pedestrian appearance frequency, and finally the specific scene is constructed for the simulation platform to load and run.

[0095] 3. Autonomous driving car:

[0096] 1) Use the RTRC 4S car developed by Lezhihang (Chongqing) Technology Co., Ltd. to deploy the vehicle end of the test platform. This car can support multi-sensor fusion, real-time control and high-precision positioning functions. The car is equipped with a Velodyne 16-line laser radar, a multi-view surround camera (supports HDR and dynamic exposure adjustment), a millimeter wave radar and an ultrasonic radar array, which can realize 360° environmental perception coverage. Through the NVIDIA Jetson Xavier NX edge computing unit, multi-source data real-time processing is realized, with a computing power of 21 TOPS. The car body adopts a four-wheel independent drive and a steer-by-wire architecture, supports CAN bus protocol and ROS 2 communication framework seamless docking, and can flexibly adapt to the development needs of L2-L4 level automatic driving algorithms. Its open interface design allows rapid expansion of sensors (V2X communication modules) and access to Carla simulation tools to form a "virtual-real combined" test closed loop, significantly improving algorithm iteration efficiency and real vehicle deployment reliability.

[0097] 2) Automatic driving test method

[0098] After the system is developed, the automatic driving test will be carried out according to the following steps:

[0099] S1, build a scene library: First, create multiple test scenes in the simulation platform. Generate high-precision road maps through RoadRunner and use Blender to model scene elements to build a highly realistic virtual environment. These scenes will simulate various driving conditions, including urban roads, highways, adverse weather and other complex situations.

[0100] S2, burn automatic driving program: Burn the developed automatic driving algorithm program into the test car to ensure that the program can process data from various sensors and make real-time decisions.

[0101] S3, test platform linkage: Connect the automatic driving car with the turntable test platform, which simulates different road environments and working conditions. The turntable test platform provides real dynamic feedback by adjusting speed, slope, road conditions and other variables to help verify the performance of the automatic driving car in different environments.

[0102] S4, execute multi-scene test: Set up regulatory scenarios according to real traffic regulations and driving rules to verify the accuracy and compliance of the automatic driving system in complying with traffic regulations. At the same time, common situations in daily driving are simulated as comprehensive scenarios, such as intersections, turns, merging, following, etc., to test the decision-making and control capabilities of the system in different driving scenarios. Considering the special environmental factors, extreme driving situations such as emergency obstacle avoidance, sudden obstacles and emergency braking are simulated to test the system's response ability and safety in emergency situations.

[0103] S5, Data Recording and Analysis: During the test process, all sensor data of the autonomous driving system (such as cameras, radars, IMUs, etc.) will be collected and recorded in real time. By comparing the sensor data with the real situation feedback by the car, detailed analysis is conducted to ensure that each test link meets the expectations and provides data support for subsequent optimization.

[0104] S6, Optimization and Iteration: After each round of testing, the system will be optimized and adjusted according to the test data and feedback results. Through multiple iterations, the system can adapt to changing road environments and continuously improve the safety, robustness and driving experience of the autonomous driving system.

[0105] For detailed design, please refer to Figure 5 .

[0106] 3) Simulation Environment and Modeling

[0107] Through Carla virtual platform simulation, dynamic adjustment of weather (rain and snow intensity, sun angle), traffic flow density and pedestrian behavior can be achieved, and diversified test scenarios covering urban roads, highways, extreme weather (sandstorm, heavy fog) can be constructed. In the simulation of multiple scenarios for testing, we combine the intersection map generated by RoadRunner with the failure vehicle modeled by Blender to simulate the emergency obstacle avoidance algorithm verification under accident scenarios; we also use the spatio-temporal alignment results of LiDAR point cloud and camera data to train multi-modal fusion perception model. This integration of tools from map modeling, scene building to sensor simulation provides high-fidelity and efficient virtual environment support for closed-loop testing of autonomous driving algorithms, extreme working condition coverage and large-scale parallel verification.

[0108] 4) Perception Algorithm and Data Processing

[0109] In terms of object detection, YOLOv7, as a current efficient single-stage detection algorithm, plays an important role in autonomous driving scenarios with its optimized ELAN module network architecture, which can accurately detect vehicles, pedestrians, traffic signs, etc. with real-time performance of 30-160 FPS. Its lightweight version (YOLOv7-tiny) is suitable for vehicle edge computing platforms, and through multi-scale feature fusion (FPN+PAN) and attention mechanism (CBAM), it significantly improves the robustness of small target detection and occlusion scenarios. At the same time, BEV technology generates a bird's eye view through inverse perspective transformation, multi-view fusion BEVFormer model, effectively eliminating image perspective distortion, combined with corner detection technology, it can accurately locate the starting / ending point of lane lines and traffic sign boundaries, providing geometric consistency guarantee for lane topology reconstruction, dynamic drivable area modeling and local high-precision map construction. To further improve the perception ability in complex traffic scenarios, visual, millimeter wave radar, laser radar and IMU data can be aligned and complemented across modalities, combined with SORT / DeepSORT algorithm to realize dynamic target tracking and trajectory prediction. This multi-dimensional technology collaboration not only significantly enhances the accuracy and system robustness of target detection, but also provides a full-stack solution for the safety and reliability of autonomous driving systems in extreme weather, dense traffic and other complex scenarios through vehicle-end, road-end and cloud-end integrated simulation testing.

[0110] 5) Autonomous driving main functions

[0111] In the entire vehicle-related system, the planning and decision-making module plays a crucial role. This module is written in Python language, and in terms of route planning, Dijkstra algorithm is used for local planning and can also complete global planning. Decision-making is constructed through state machines or behavior trees. Vehicle control covers two main aspects: longitudinal control and lateral control. Longitudinal control mainly involves throttle and brake operation, while lateral control focuses on steering angle. These control functions can be verified in the Carla simulation environment. In actual real vehicle application scenarios, the vehicle control module will be connected with STM32 to achieve effective control of the real vehicle. When interacting with vehicle data, Jetson Nano and other embedded platforms undertake local real-time computing tasks, while the corresponding perception algorithms also run on the platform. The host computer and remote server have special purposes, they can be used for large-scale data recording work, and provide data support for subsequent training, which helps to improve the performance of the entire system in different scenarios.

[0112] 6) Hardware end integration

[0113] STM32 is used to drive interface and data collection with real sensors (camera, radar, IMU); also used for sending vehicle chassis control signals (through PWM / CAN, etc.); data / command interaction with Jetson Nano (serial, SPI, I2C or Ethernet). Get vehicle status and image transmission data through SW (Software Wireless) and TCP communication protocol; share positioning or environmental information with the vehicle system to expand the range of environmental perception.

[0114] 7) System testing and iteration

[0115] Verify functionality and stability on the simulation platform first, then transplant to the real environment for integration testing. Then record multi-source sensor data in each test for data recording and playback, which can be played back in Carla or ROS environment for offline analysis and model optimization. Keep manual remote control or emergency braking channel on the basis of intelligent system to ensure controllability and safety during testing.

[0116] In the description of the present specification, the description of the terms "one embodiment", "example", "specific example" and the like means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are contained in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0117] The preferred embodiments of the application disclosed above are only used to help explain the application. The preferred embodiments do not describe all the details and limit the application to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the present specification. The present specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. An autonomous driving virtual simulation test platform, characterized in that: The test platform includes: Virtual simulation module, used to build high-precision three-dimensional virtual scenes and simulate sensor data; An autonomous driving car, used to execute autonomous driving algorithms in a physical environment and collect real sensor data; A road simulation system, which is used to simulate various road conditions and form a data closed loop with the virtual simulation module and the desktop-level autonomous driving car; The data interaction module is used to realize real-time data transmission and synchronization between the virtual simulation module, the desktop-level autonomous driving car and the road simulation system.

2. The autonomous driving virtual simulation test platform according to claim 1, characterized in that: The virtual simulation module includes: The scene modeling unit generates high-precision road topology maps based on RoadRunner and combines it with Blender to build three-dimensional scene elements including dynamic obstacles and traffic signs. The sensor simulation unit simulates the perception data of RGB-D cameras, lidar, and millimeter-wave radar through the Carla engine, and supports spatiotemporal alignment with real sensor data; The dynamic environment simulation unit can adjust the weather, lighting and traffic flow density in the virtual scene to create an extreme working condition test environment.

3. The autonomous driving virtual simulation test platform according to claim 1, characterized in that: The autonomous driving car includes: The vehicle structure integrates an RGB-D camera, a lidar simulated ToF sensor, and a millimeter-wave radar array to achieve 360-degree environmental perception; The processing unit uses the NVIDIA Jetson Nano edge computing module to run the YOLOv7 target detection algorithm and the BEV perspective conversion model; The control unit, based on an STM32 embedded controller, drives the four-wheel independent motors and servo steering through PWM signals, and supports physical interaction with the road simulation system.

4. The autonomous driving virtual simulation test platform according to claim 1, characterized in that: The road simulation system comprises: A dual-hub mechanism, driven by a brushless DC motor, is used to secure the desktop autonomous vehicle and simulate its speed. The tilt adjustment module simulates the slope of the road surface through an electric push rod; The track and wheelbase adjustment mechanism uses a screw guide rail and a movable bracket structure to support track adjustment and quick adaptation of the wheelbase; The data acquisition unit integrates a speed sensor, a torque sensor, and a six-axis force sensor to collect vehicle operation data in real time and feed it back to the virtual simulation module.

5. The autonomous driving virtual simulation test platform according to claim 1, characterized in that: The data interaction module is based on ROS2 and OPCUA protocols to achieve: Bidirectional injection of sensor data between the virtual simulation module and the autonomous driving vehicle; The road simulation system synchronizes dynamic data with the virtual simulation module. The road simulation system transmits physical resistance data to the virtual simulation module, and the virtual simulation module feeds back virtual road condition parameters to the road simulation system to dynamically adjust the working conditions.

6. A testing method for the autonomous driving virtual simulation test platform according to claim 1, characterized in that: The test method comprises the following steps: Step S1: Build a scenario library, generate high-precision road maps using RoadRunner, combine it with Blender to model dynamic obstacles, and configure virtual test scenarios containing extreme weather and complex traffic flows in the Carla engine; Step S2: Burning the autonomous driving algorithm program to the Jetson Nano processing unit of the desktop-level autonomous driving car, the algorithm program integrates YOLOv7 target detection, BEV perspective conversion, and path planning modules based on the Dijkstra algorithm; Step S3: Linkage test: The desktop-level autonomous driving car is fixed to the road simulation system. The road simulation system adjusts the slope, speed, and road resistance to simulate flat roads, ramps, and highways. At the same time, the virtual scene data is injected into the physical car through the data interaction middleware, and the real sensor data is fed back to the virtual scene. Step S4: Execute multi-scenario testing, including regulatory scenarios, comprehensive scenarios, and extreme scenarios, and record the vehicle's motion trajectory, control commands, and sensor data; Step S5: Data comparison and analysis: compare the virtual scene simulation results with the physical car measured data to optimize the autonomous driving algorithm; Step S6: Iterative optimization, repeating steps S3 to S5 until the algorithm meets the preset performance indicators in both the virtual and physical environments.

7. The testing method of the autonomous driving virtual simulation test platform according to claim 6, characterized in that: In step S1, the scenario library defines the road topology based on the Open Drive format and describes the traffic event logic in the Open Scenario format, and supports the reproduction of scenarios driven by KITTI and Waymo real data sets.

8. The testing method of the autonomous driving virtual simulation test platform according to claim 6, characterized in that: In step S3, the road simulation system controls the motor torque and the roller speed through the PID adjustment module to simulate the acceleration and deceleration conditions and the slope resistance.

9. The testing method of an autonomous driving virtual simulation test platform according to claim 6, characterized in that: In step S5, the performance indicators include safety, comfort, task completion efficiency and energy efficiency, and the evaluation results are visualized through the digital twin scene rendered by Unreal Engine.

Images