An automatic driving simulation test method integrating multiple platforms in a complex traffic environment
By integrating multiple simulation platforms, constructing vehicle dynamics and sensor models, and building complex traffic environments, the comprehensiveness and efficiency of autonomous driving testing methods under complex traffic conditions were solved, achieving efficient autonomous driving simulation testing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2022-12-16
- Publication Date
- 2026-05-19
AI Technical Summary
Existing autonomous driving testing methods lack comprehensiveness and efficiency under complex traffic conditions, have high field testing costs, and are insufficient for simulation software testing, thus failing to fully support further research on autonomous driving technology.
By integrating multiple simulation platforms, including Carsim, Prescan, and Simulink, vehicle dynamics models, sensor models, and control systems are constructed, complex traffic environments are built, and decision-making and planning models are established through dynamic programming and quadratic programming, forming a simulation testing method for autonomous driving under specific environments.
It improves the comprehensiveness and efficiency of autonomous driving simulation testing, can dynamically and realistically simulate complex traffic scenarios, supports joint simulation testing across multiple platforms, has greater adaptability, and improves the efficiency of road network construction and the accuracy of simulation testing.
Smart Images
Figure CN115859470B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an autonomous driving simulation testing method, belonging to the field of traffic simulation technology. Background Technology
[0002] With the continuous development of urbanization in my country, a large influx of people has entered cities, leading to a significant increase in both the urban population and the number of motor vehicles owned by urban residents. The ever-increasing demand for transportation has exacerbated the strain on transportation resources, making traffic congestion a major problem in most cities, especially large ones. Autonomous driving has significant positive implications for improving traffic order and efficiency. In recent years, the rapid development of technologies such as vehicle-to-everything (V2X), artificial intelligence, and intelligent vehicles has provided excellent conditions for the development of autonomous driving, which has now become an important research direction for intelligent mobility and intelligent connectivity development both domestically and internationally.
[0003] Testing autonomous driving is essential. While current autonomous driving technologies are continuously improving in automation, they still haven't reached the level required for real-world road use, necessitating extensive testing. Existing field testing methods have limitations in terms of cost, efficiency, and comprehensiveness; therefore, simulation software testing is a crucial supplement. Computer simulation technology can dynamically and realistically simulate traffic phenomena such as vehicles, road environments, pedestrians, and traffic flow, including changes in vehicle-related parameters, spatiotemporal variations in traffic flow, and in-depth analysis of the characteristics of various traffic participants, providing support for further research. Summary of the Invention
[0004] The technical problem to be solved by this invention is to provide an autonomous driving simulation test method that integrates multiple platforms under complex traffic conditions. By integrating the advantages of different platforms in the field of traffic simulation and adding the built environment and the corresponding algorithms, different complex traffic scenarios are built to form an autonomous driving simulation test method under specific environments.
[0005] To solve the above technical problems, the technical solution adopted by the present invention is as follows:
[0006] This invention covers the entire process of autonomous driving perception, decision-making, planning, and control, integrating the advantages of multiple platforms. It proposes an autonomous driving simulation testing method integrating multiple platforms in complex traffic environments, including the following steps:
[0007] S1. First, construct the vehicle dynamics model, then build the vehicle model in Carsim, and write the vehicle's lateral and longitudinal control algorithms to complete the control module of the autonomous vehicle.
[0008] S2. Extract the required road network from Openstreemap, import it into Prescan for road network topology adjustment and redundant road deletion, and build the basic road network on this basis.
[0009] S3. Add a vehicle perception model to Prescan and build complex traffic environment scenarios;
[0010] S4. Establish decision-making and planning models adapted to complex traffic environments based on dynamic programming and quadratic programming;
[0011] S5 integrates Carsim, Prescan, and Simulink to establish a joint simulation platform, where Carsim provides vehicle dynamics models, Prescan provides sensor models and diverse simulation scenarios, and Simulink provides simulation control systems.
[0012] S6. Test the joint simulation platform built in step S5, as well as the road network, algorithm, and model added in steps S1-S4.
[0013] Furthermore, the autonomous driving simulation testing method integrating multiple platforms in complex traffic environments proposed in this invention includes the following specific steps in step S1:
[0014] S101. Construct a vehicle dynamics model based on the two-degree-of-freedom vehicle dynamics equations. The specific derivation formula is as follows:
[0015]
[0016] Where m is the mass of the vehicle, a and b are the distances from the front and rear wheels to the center of mass, respectively, and v x and v y These represent the velocities in the two directions in the vehicle's coordinate system, and δ is the front wheel steering angle. Angular velocity, C is the angular acceleration. αf and C αr It is the lateral stiffness of the front and rear wheels of the vehicle, and I is the moment of inertia of the vehicle.
[0017] S102. A linear quadratic regulator is used for vehicle lateral control. Under the constraints of the vehicle dynamics model, the vehicle body's v is controlled by selecting an appropriate front wheel steering angle δ. y and angular velocity Control it to adapt it to the planned position, speed, and heading angle information;
[0018] S103. A PID controller is used for longitudinal vehicle control. By debugging and adopting appropriate proportional, integral and derivative parameters, the speed and acceleration can reach the planned desired speed and acceleration.
[0019] Furthermore, the autonomous driving simulation testing method integrating multiple platforms in complex traffic environments proposed in this invention includes the following specific steps in step S2:
[0020] S201. Perform road network topology adjustments and delete redundant roads, including adjusting the road network that does not match the actual situation and deleting redundant roads that are no longer in use.
[0021] S202. Update the basic road network, including: road network topology verification, redundant intersection adjustment, and subnetwork deletion;
[0022] S203. Based on the adjustment of the road network, update the road network data according to the actual road conditions.
[0023] Furthermore, the autonomous driving simulation testing method integrating multiple platforms under complex traffic environments proposed in this invention includes the construction of complex traffic environment scenarios in step S3. These complex traffic environments include adverse weather conditions, existing harsh road conditions, and multi-mode traffic operation conditions. Adverse weather conditions refer to weather environments that affect the road's adhesion coefficient and the vehicle's perception capabilities, including rain, fog, ice, and snow.
[0024] Furthermore, in the autonomous driving simulation testing method integrating multiple platforms under complex traffic environments proposed in this invention, step S4 involves establishing a decision-making and planning model adapted to complex traffic environments based on dynamic programming and quadratic programming, as detailed below:
[0025] S401. Based on dynamic programming and quadratic programming, establish a decision and planning model. When an autonomous vehicle senses an obstacle in front, the vehicle needs to make an optimal decision based on the magnitude of the cost function, choosing to avoid the obstacle on the left or the right.
[0026] The cost function of the established dynamic programming is:
[0027] C1=∑l i 2 +∑h[(s i -s) 2 +(l i -l) 2 ]+k1∑f'(s i ) 2 +k2∑f"s i ) 2 +k3∑f”'(s i ) 2
[0028] Where l is the distance from the point to the reference line of the planned path, l i Let s be the distance from the i-th point to the reference line of the planned path, and s be the coordinates of the forward direction. iLet l = f(s) be the coordinates of the i-th point in the direction of travel. The distance to the reference line can be calculated from the coordinates of the point. f(s) is expressed as a fifth-order polynomial, f'(s) i ), f" i ), f”'(s i ) represent the first, second, and third derivatives of the distance from the i-th point to the reference line, respectively, and k1, k2, and k3 are influence coefficients. This cost function includes the influence of the reference line distance, the obstacle distance, and the smoothness.
[0029] S402. The state transition equation for dynamic programming is shown below. Search for the shortest cost path from the starting point P0 to the j-th level:
[0030] C1(P0,P ij =min[C1(P0,P 1j-1 )+C1(P 1j-1 ,P ij ),C1(P0,P 2j-1 )+C1(P 2j-1 ,P ij ),..,C1(P0,P i-1j-1 )+C1(P i-1j-1 ,P ij )]
[0031] Among them, P ij To represent distinct points after spatial discretization;
[0032] S403. After dynamic programming decision-making, a convex space is created, and quadratic programming is performed on this space. The cost function is:
[0033]
[0034] Where k4, k5, k6, and k7 are the influence coefficients of the reference line, the first derivative, the second derivative, and the third derivative, respectively. i Let l = f(s) be the distance from each point to the reference line of the planned path, and l' be the distance between the upper and lower bounds of the convex space. i 、l” i and They are respectively the lth i First, second, and nth derivatives, l” i+1 It is the second derivative of the distance from the (i+1)th point to the reference line.
[0035] The present invention adopts the above technical solution and has the following technical effects compared with the prior art:
[0036] This invention presents a multi-platform co-simulation testing method that integrates the advantages of various simulation tools, including Carsim's vehicle dynamics model, Prescan's sensor and scene model, and Simulink's control system. This results in a more widely applicable and adaptable autonomous driving simulation testing method. Furthermore, in scene construction, this invention improves efficiency by downloading, importing, and adjusting road networks from OpenStreetMap. Additionally, it builds complex traffic environments and various sensor models, including those for severe weather and multimodal traffic, within Prescan for simulation testing. Attached Figure Description
[0037] Figure 1 This is a technical flowchart of an autonomous driving simulation testing method integrating multiple platforms under complex traffic conditions, according to the present invention.
[0038] Figure 2 Configure vehicle parameters in Carsim.
[0039] Figure 3 Setting up scenarios for the Prescan platform.
[0040] Figure 4 This is a simulation of a snowfall scenario.
[0041] Figure 5 Simulation test results for pedestrian avoidance by autonomous vehicles.
[0042] Figure 6 This is a schematic diagram of the planned route based on dynamic programming and quadratic programming. Detailed Implementation
[0043] The implementation process of this invention will be further described in detail below with reference to examples and accompanying drawings. Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless defined as herein.
[0044] like Figure 1 As shown, this invention proposes an autonomous driving simulation testing method integrating multiple platforms in complex traffic environments. The specific steps of this method are as follows:
[0045] I. Constructing the vehicle model
[0046] (1) First, a vehicle dynamics model is constructed based on the two-degree-of-freedom vehicle dynamics equation. Then, a vehicle model is built in Carsim, and the vehicle lateral control and longitudinal control algorithms are written in. Finally, the control module of the autonomous vehicle is completed in Simulink.
[0047] The derived formula for the established vehicle dynamics model is as follows:
[0048]
[0049] Where m is the mass of the vehicle, a and b are the distances from the front and rear wheels to the center of mass, respectively, and v x and v y These represent the velocities in the two directions in the vehicle's coordinate system, and δ is the front wheel steering angle. Angular velocity, C is the angular acceleration. αf and C αr I is the lateral stiffness of the front and rear wheels of the vehicle, and I is the moment of inertia of the vehicle.
[0050] (2) Connect the Audi A8 Sedan dynamics interface in Prescan to Carsim, and then set more detailed vehicle dynamics parameters in Carsim, such as... Figure 2 As shown, it includes vehicle wheelbase, center of gravity position, sprung mass, etc.
[0051] (3) In Carsim, select Simulink in Models to connect it with Matlab / Simulink. In the Simulink model, select the required inputs and outputs. The inputs include front wheel angle, throttle, brake, etc., and the outputs include position, speed, acceleration, etc.
[0052] II. Vehicle Control
[0053] According to automotive theory, in vehicle lateral control, the driver's control of the steering wheel causes a change in the front wheel steering angle, thereby controlling the vehicle's heading angle. In this embodiment, vehicle lateral control mainly employs LQR (Linear Quadratic Regular) control. Under the constraints of the vehicle dynamics model, by selecting an appropriate front wheel steering angle δ, the vehicle's v... y and angular velocity Control the system to make it as close as possible to the planned speed, acceleration, and heading angle.
[0054] In vehicle longitudinal control, when the driver presses the accelerator, the engine power changes, and the engine speed and torque also change, thus changing the vehicle's speed and acceleration, causing the vehicle to accelerate. Conversely, when the driver presses the brake, the braking pressure increases, the brake disc friction coefficient increases, and the resistance increases, causing the vehicle to decelerate. Different engines have different engine torque-speed curves. In this embodiment, the longitudinal control of the vehicle uses a PID controller (Proportion Integration Differential). By adjusting and using appropriate proportional, integral, and derivative parameters, the speed and acceleration can quickly and stably reach the planned desired speed and acceleration.
[0055] III. Road Network Construction
[0056] First, the required road network is extracted from OpenStreetMap and imported into Prescan for topology adjustment and redundant road removal. Based on this, the basic road network is built.
[0057] Road network topology adjustments and redundant road removal include adjusting road networks that do not accurately reflect reality and removing redundant roads that are no longer in use. Because OpenStreetMap's accuracy cannot fully reach the lane level, appropriate adjustments to some lanes are necessary. For example... Figure 3 As shown, a section of road within a campus is imported from OpenStreetMap and adjusted.
[0058] Then, the basic road network is updated. The main steps include checking the road network topology, adjusting redundant intersections, and deleting subnetworks. Finally, based on the adjustments to the road network, the road network data is updated according to the actual road conditions.
[0059] IV. Construction of Complex Traffic Scenarios
[0060] In Prescan, vehicle perception models such as LiDAR sensors are added, and scenarios are built for complex traffic environments that significantly impact autonomous driving. These complex environments include severe weather conditions that greatly affect vehicle operation, as well as harsh road conditions under special circumstances and complex road conditions in multi-modal traffic. Therefore, severe weather conditions such as rain, fog, ice, and snow are added, as these weather conditions affect the road adhesion coefficient and the vehicle's perception capabilities. The constructed 3D environment includes… Figure 4 As shown.
[0061] V. Decision-making and planning models for complex environments
[0062] A decision-making and planning model adapted to complex traffic environments is established based on dynamic programming. The decision-making and planning algorithm is built upon dynamic programming and quadratic programming.
[0063] (1) When an autonomous vehicle senses an obstacle ahead, it needs to make a decision based on a cost function, deciding whether to avoid the obstacle on the left or right. The cost function for dynamic programming is:
[0064] C1=∑l i 2 +∑h[(s i -s) 2 +(l i -l) 2 ]+k1∑f'(s i ) 2 +k2∑f"s i ) 2 +k3∑f”'(s i ) 2
[0065] Where l is the distance from the point to the reference line of the planned path, l i Let s be the distance from the i-th point to the reference line of the planned path, and s be the coordinates of the forward direction. i Let l = f(s) be the coordinates of the i-th point in the direction of travel. The distance to the reference line can be calculated from the coordinates of the point. f(s) is expressed as a fifth-order polynomial, f'(s) i ), f" i ), f”'(s i ) represent the first, second, and third derivatives of the distance from the i-th point to the reference line, respectively, and k1, k2, and k3 are influence coefficients. This cost function includes the influence of the reference line distance, the obstacle distance, and the smoothness.
[0066] (2) Establish a planning model adapted to complex traffic environments based on quadratic programming. The state transition equation of dynamic programming is shown below, which searches for the shortest cost path from the starting point P0 to the j-th layer.
[0067] C1(P0,P ij =min[C1(P0,P 1j-1 )+C1(P 1j-1 ,P ij ),C1(P0,P 2j-1 )+C1(P 2j-1 ,P ij ),..,C1(P0,P i-1j-1 )+C1(P i-1j-1 ,P ij )]
[0068] Where Pij It represents the distinct points after spatial discretization.
[0069] (3) After dynamic programming decision-making, a convex space is opened up, and quadratic programming is performed on this basis. The cost function is:
[0070]
[0071] Where k4, k5, k6, and k7 are the influence coefficients of the reference line, the first derivative, the second derivative, and the third derivative, respectively. l = f(s), and it lies between the upper and lower bounds of the convex space. i 、l” i and They are respectively the lth i First, second, and nth derivatives, l” i+1 It is the second derivative of the distance from the (i+1)th point to the reference line.
[0072] VI. Co-simulation integrating multiple platforms
[0073] A joint simulation platform is established by integrating Carsim, Prescan, and Simulink. Carsim can provide more accurate vehicle dynamics models compared to other tools, Prescan can provide sensor models and diverse simulation scenarios, and Simulink has a relatively convenient and mature simulation control system.
[0074] VII. Simulation Platform Testing
[0075] Finally, the platform, along with the added road network, algorithms, and models, were tested. The simulation results showed 3D simulation effects as follows. Figure 5 As shown, autonomous vehicles can avoid pedestrians and obstacles on the road. Figure 6 This is a schematic diagram of a route based on dynamic programming and quadratic programming.
[0076] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.
Claims
1. A simulation testing method for autonomous driving integrating multiple platforms in complex traffic environments, characterized in that, The method includes the following steps: S1. First, construct the vehicle dynamics model, then build the vehicle model in Carsim and write the vehicle's lateral and longitudinal control algorithms, and complete the control module of the autonomous vehicle in Simulink. S2. Extract the required road network from Openstreemap, import it into Prescan for road network topology adjustment and redundant road deletion, and build the basic road network on this basis. S3. Add a vehicle perception model to Prescan and build complex traffic environment scenarios; S4. Establish a decision-making and planning model adapted to complex traffic environments based on dynamic programming and quadratic programming; details are as follows: S401. Based on dynamic programming and quadratic programming, establish a decision and planning model. When an autonomous vehicle senses an obstacle in front, the vehicle needs to make an optimal decision based on the magnitude of the cost function, choosing to avoid the obstacle on the left or the right. The cost function of the established dynamic programming is: ， in, The distance from the point to the reference line of the planned path. For the first The distance from each point to the reference line of the planned path The coordinates are for the direction of travel. For the first The coordinates of the direction of movement of each point. The distance to the reference line can be calculated from the coordinates of the point. Represented using a fifth-degree polynomial, , , They represent the first The first, second, and third derivatives of the distance from each point to the reference line. , , As the influence coefficient, this cost function includes the influence of reference line distance, obstacle distance, and smoothness. S402, The state transition equation for dynamic programming is shown below. The search starts from the starting point. To the Shortest cost path of the layer: ； in, To represent distinct points after spatial discretization; S403. After dynamic programming decision-making, a convex space is created, and quadratic programming is performed on this space. The cost function is: ， in, , , , These are the influence coefficients of the reference line, the first derivative, the second derivative, and the third derivative, respectively. The distance from each point to the reference line of the planned path. Between the upper and lower bounds of convex space, , and The first First-order, second-order and First derivative, For the first The second derivative of the distance from each point to the reference line; S5 integrates Carsim, Prescan, and Simulink to establish a joint simulation platform, where Carsim provides vehicle dynamics models, Prescan provides sensor models and diverse simulation scenarios, and Simulink provides simulation control systems. S6. Test the joint simulation platform built in step S5, as well as the road network, algorithm, and model added in steps S1-S4.
2. The autonomous driving simulation testing method integrating multiple platforms in complex traffic environments according to claim 1, characterized in that, Step S1 is as follows: S101. Construct a vehicle dynamics model based on the two-degree-of-freedom vehicle dynamics equations. The specific derivation formula is as follows: ； Where m is the mass of the vehicle. and These are the distances from the front and rear wheels of the vehicle to its center of gravity, respectively. and These represent the velocities in two directions within the vehicle's coordinate system. For the front wheel steering angle, Angular velocity, Angular acceleration, and It refers to the lateral stiffness of the front and rear wheels of the vehicle. It is the vehicle's moment of inertia; S102. A linear quadratic regulator is used for vehicle lateral control. Under the constraints of the vehicle dynamics model, a suitable front wheel steering angle is selected. For the car body and angular velocity Control it to adapt it to the planned position, speed, and heading angle information; S103. A PID controller is used for longitudinal vehicle control. By debugging and adopting appropriate proportional, integral and derivative parameters, the speed and acceleration can reach the planned desired speed and acceleration.
3. The autonomous driving simulation testing method integrating multiple platforms in complex traffic environments according to claim 1 or 2, characterized in that, The step S1 involves building a vehicle model in Carsim, including setting vehicle dynamics parameters in Carsim, such as wheelbase, center of gravity position, and sprung mass.
4. The autonomous driving simulation testing method integrating multiple platforms in complex traffic environments according to claim 1 or 2, characterized in that, In step S1, controlling the autonomous vehicle in Simulink includes selecting the required inputs and outputs in the Simulink model. The inputs include front wheel steering angle, throttle, and brake, and the outputs include position, speed, and acceleration.
5. The autonomous driving simulation testing method integrating multiple platforms in complex traffic environments according to claim 1, characterized in that, Step S2 is as follows: S201. Perform road network topology adjustments and delete redundant roads, including adjusting the road network that does not match the actual situation and deleting redundant roads that are no longer in use. S202. Update the basic road network, including: road network topology verification, redundant intersection adjustment, and subnetwork deletion; S203. Based on the adjustment of the road network, update the road network data according to the actual road conditions.
6. The autonomous driving simulation testing method integrating multiple platforms in complex traffic environments according to claim 1, characterized in that, In step S3, a complex traffic environment scenario is constructed, which includes severe weather conditions, existing harsh road conditions, and multi-mode traffic operation conditions.
7. The autonomous driving simulation testing method integrating multiple platforms in complex traffic environments according to claim 6, characterized in that, The severe weather conditions refer to weather environments that affect the road's adhesion coefficient and the vehicle's perception capabilities, including rain, fog, ice, and snow.