Lightweight generation method for human-like interaction behavior of test background vehicle in unprotected left-turn scene

Abstract

The invention discloses a test background vehicle-like man interaction behavior lightweight generation method in an unprotected left-turn scene, and the method comprises the steps: 1, setting two closed loops as a route 1 and a route 2, and enabling the route 1 and the route 2 to have crossed road sections; 2, a test background vehicle is arranged on the route 1, and an automatic driving main vehicle is arranged on the route 2; 3, simulation is carried out, the background vehicle is tested before simulation is started, and the automatic driving main vehicle is still; after simulation is started, each test background vehicle is controlled in an IDM following mode when not driving to an intersection area, and is switched to a human-like interaction mode when driving to the intersection area, so that the performance reliability of an automatic driving main vehicle algorithm is tested; 4, after passing through the intersection area, the test background vehicle is switched to the initial IDM vehicle following mode for the second time; and 5, repeating the steps 3 and 4, and circularly and continuously monitoring the performance reliability of the automatic driving main vehicle algorithm. According to the invention, the safety and credibility of the automatic driving algorithm can be tested.

Description

Technical Field

[0001] The present invention relates to the field of autonomous vehicle testing methods, and specifically to a lightweight generation method for the human-like interaction behavior of a test background vehicle in an unprotected left-turn scenario. Background Art

[0002] In recent years, the development of the autonomous driving field has been changing with each passing day. As one of the three major pillars of autonomous driving test deployment, scenario-based virtual simulation has become a major research hotspot due to its significant advantages such as high efficiency and safety. In autonomous driving simulation tests, it is necessary to test the reliability of autonomous vehicle algorithms, which puts higher requirements on the background vehicle test system. Therefore, the method for the human-like interaction behavior of the test background vehicle under dangerous working conditions is of great significance. In previous autonomous vehicle tests, traditional microscopic traffic models (such as IDM, improved ACC) were generally used to generate background vehicle trajectories. These models rarely considered driver behavior characteristics, and the test authenticity was limited; or high-precision models of human-like interaction (such as LSTM and GAN models based on deep learning) were adopted, which helped to simulate real driving behaviors, but had high requirements for CPU / GPU energy consumption and computing power, and poor economy and model interpretability. Summary of the Invention

[0003] The present invention provides a lightweight generation method for the human-like interaction behavior of a test background vehicle in an unprotected left-turn scenario to solve the problems existing in the existing background vehicle test system.

[0004] To achieve the above object, the technical solution adopted by the present invention is as follows: A lightweight generation method for the human-like interaction behavior of a test background vehicle in an unprotected left-turn scenario includes the following steps: Step 1: Set the simulation working condition as an unprotected left-turn scenario on an urban road. There are two closed loops in the scenario, which are set as Route 1 and Route 2 respectively, and Route 1 and Route 2 have a shared intersection section. Step 2: Set a number of test background vehicles on Route 1. Each test background vehicle is arranged in sequence along the driving direction. The distance between adjacent test background vehicles on Route 1 is equal, and each test background vehicle can circulate on Route 1. Set an autonomous driving host vehicle on Route 2. The autonomous driving host vehicle can drive on Route 2 and turn left at the intersection, so there is a conflict between the test background vehicle and the left-turning autonomous driving host vehicle in the intersection area. Step 3: Conduct the simulation. When simulating, each test background vehicle has two different control modes, namely the IDM following mode and the human-like interaction mode. Before the simulation starts, make each test background vehicle and the autonomous driving host vehicle stand still on their respective routes. After the simulation starts, let each test background vehicle circulate on Route 1, and let the autonomous driving host vehicle drive on Route 2 and turn left at the intersection; when the test background vehicle has not traveled to the intersection area, use the IDM following mode to control the leading vehicle of each test background vehicle to travel at a constant speed with the IDM free flow speed, and the remaining test background vehicles follow along the driving direction; when any test background vehicle travels to the intersection area, switch to the human-like interaction mode, so that the test background vehicle traveling to the intersection area randomly performs lane-cutting / yielding / game-playing behaviors to interfere with the autonomous driving host vehicle, thereby testing the reliability of the autonomous driving host vehicle algorithm performance; Step 4. After passing through the intersection area, switch the test background vehicle to the initial IDM following mode for the second time, and let the test background vehicle continue to drive on Route 1; Step 5. After completing a test cycle according to Steps 3 and 4, let the test background vehicle continue to drive on Route 1; after completing a test cycle according to Steps 3 and 4, the autonomous driving host vehicle continues to drive on Route 2. Until the test background vehicle reaches the intersection area next time, then repeat Steps 3 and 4, and continuously monitor the reliability of the autonomous driving host vehicle algorithm performance in this cycle.

[0005] Further, the scenario described in Step 1 is built by Prescan.

[0006] Further, in Steps 2-5, each test background vehicle is equipped with a TIS sensor, thereby realizing vehicle safety following.

[0007] Further, in Steps 3 and 4, the test background vehicle mode switching uses the switch module in Simulink, takes the coordinates of the test background vehicle in the global coordinate system as the input, realizes the switching of different control modes, and then outputs two signals of throttle and brake to the test background vehicle dynamics model for longitudinal speed control.

[0008] Compared with the prior art, the advantages of the present invention are: Based on the autonomous driving simulation test, the present invention realizes a lightweight generation method of human-like interaction of test background vehicles in an unprotected left-turn scenario through the joint construction of Prescan and Simulink. The algorithm is flexible and simple, can realize human-like decision-making, makes the driving interaction behavior more real and accurate, and has the advantages of light weight, a small algorithm model, appropriate required computing power, and high accuracy. It can be used as an important means for the reliability test of autonomous driving vehicles, improve the intelligence level of test background vehicles, and test the safety and credibility of autonomous driving algorithms. Brief Description of the Drawings

[0009] Figure 1 is the flowchart of the method of the embodiment of the present invention.

[0010] Figure 2 is the schematic diagram of the scenario built in the embodiment of the present invention. Specific implementation mode

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0012] As Figure 1 shown, this embodiment discloses a lightweight generation method for testing background vehicle human-like interaction behaviors in an unprotected left-turn scenario, including the following steps: Step 1: Establish a scenario.

[0013] As Figure 2 shown, Prescan is used to build a simulation working condition for an unprotected left-turn scenario on an urban road. There are two closed loops in the scenario, which are respectively set as Route 1 and Route 2, and Route 1 and Route 2 have a shared intersection section. This intersection section has a cross-shaped intersection area and a T-shaped intersection area This embodiment is applicable to the unprotected left-turn scenario at urban road intersections, and the designed vehicle speed is 30 - 50 km / h. Therefore, this example aims to detect the reliability of the main vehicle algorithm for autonomous driving in the case of unprotected left-turn. Thus, only the cross-shaped intersection area is considered as the conflict area, and the lower T-shaped intersection area is not considered as the conflict area.

[0014] Step 2: Set a number of test background vehicles on Route 1 and set the main vehicle for autonomous driving on Route 2.

[0015] In this embodiment, a total of 8 test background vehicles are set on Route 1. The 8 test background vehicles are arranged in sequence along the driving direction. The distance between adjacent test background vehicles on Route 1 is equal, all being 20 m, and each test background vehicle can circulate on Route 1. Since the simulation scenario is at an urban road intersection and the vehicle speed is low, and it is necessary to focus on short-distance perception, each test background vehicle fleet is equipped with a TIS sensor, which is conducive to realizing vehicle safe following.

[0016] The main vehicle for autonomous driving, as the vehicle to be tested, can drive on Route 2 and turn left at the intersection. Thus, as can be seen from the Figure 2 shown route, there will be a conflict between the test background vehicle and the left-turning main vehicle for autonomous driving in the cross-shaped intersection area.

[0017] Step 3: Conduct simulation.

[0018] During simulation, each test background vehicle has two different control modes, namely the IDM following mode and the human-like interaction mode.

[0019] Before the simulation starts, make each test background vehicle and the main vehicle for autonomous driving stand still on their respective routes.

[0020] After the simulation starts, each test background vehicle is made to drive in circles on Route 1, and the autonomous driving host vehicle is made to drive on Route 2 and turn left at the intersection. When the test background vehicles have not reached the intersection area, the IDM following mode is used to control the leading vehicle among the test background vehicles to drive at a constant speed at the IDM free flow speed, and the remaining test background vehicles follow in the driving direction. When any test background vehicle reaches the intersection area, it switches to the human-like interaction mode, so that the test background vehicles that reach the intersection area randomly perform lane-changing / yielding / game-playing behaviors to interfere with the autonomous driving host vehicle, thereby testing the reliability of the algorithm performance of the autonomous driving host vehicle.

[0021] In this embodiment, in the intersection area, the decision-making behaviors of the test background vehicles will interfere with the autonomous driving host vehicle. The autonomous driving host vehicle needs to judge their behaviors (lane-changing / slowing down to yield / game-playing) based on information such as the positions and speeds of the test background vehicles, and use this as input to make decisions.

[0022] Specifically, in this implementation, the specific setting process of the IDM following mode of the test background vehicles is as follows: The IDM formula is implemented using the MATLAB Function Block in Simulink. The speed of the current test background vehicle, the speed of the preceding vehicle, and the distance to the preceding vehicle are input into the IDM controller, and then the acceleration is calculated by this formula.

[0023] The information of the preceding vehicle among the test background vehicles is transmitted by the TIS sensor (in the initial state, the leading vehicle has no input of information about the preceding vehicle and drives at the IDM free flow speed). Each test background vehicle is equipped with 2 TIS sensors (short-range and long-range), and the detection ranges are set to ensure coverage of the following distance (short-range 50m, 120°, long-range 100m, 12°) and direction (both are directly in front of the vehicle).

[0024] In the IDM following mode, the safe time headway T = 1.6s, the static safety distance s 0 = 4m, and the maximum acceleration is 1.0m / s 2 .

[0025] Among the test background vehicles, each following vehicle creates an independent subsystem, including an input signal module (the distance and speed of the preceding vehicle, obtained by the TIS sensor), a signal processing module (calculation of relative speed and relative distance, calculated by the MATLAB Subtract module), an IDM controller (the MATLAB Function Block is connected to the input signal), and a vehicle dynamics model (the Integrator module integrates the acceleration and speed respectively to obtain the displacement), and an independent controller is configured for each following vehicle to achieve signal connection, thereby realizing the platoon following of the test background vehicles.

[0026] In a specific embodiment of the present invention, the specific setting process of the humanoid interaction mode of the test background vehicle is as follows: The behavior decision-making layer of the humanoid interaction mode is a finite state machine (FSM) + game model. The state transition conditions are set to three types (triggering lane-changing: when the host vehicle is far from the conflict point, the background vehicle accelerates with probability P agg ; triggering yielding: when the host vehicle approaches the conflict point and has a high speed, the test background vehicle decelerates with probability P yield ; game mode: calculating the Nash equilibrium based on the relative distance and speed of both sides, and dynamically adjusting the acceleration). The game model is implemented using MATLAB Function code. The control execution layer is based on an improved IDM model, and the acceleration correction of the game decision is added. Then, random numbers are generated through the Random Number Generator module (MATLAB Function code) to trigger different decisions.

[0027] The inputs of the intelligent controller of the test background vehicle include: inputting the position / speed of the host vehicle and its own state (position, speed), and using the finite state machine (Stateflow) game model (MATLAB Function) for decision-making. Among them, the control layer is the improved IDM longitudinal control, and the throttle / brake commands are output to the vehicle dynamics model.

[0028] In the humanoid interaction mode, the state transition logic is implemented through the Stateflow module; the conflict point distance and random triggering are responded to; the game model and IDM formula are embedded in the MATLAB Function Block; the Vehicle Dynamics uses the vehicle model provided by Prescan; and the vehicle state is synchronized through the Prescan-Simulink interface module.

[0029] In this embodiment, position triggering is adopted, that is, when the test background vehicle does not reach the intersection triggering area, the IDM model is used for control. When any test background vehicle travels to the intersection triggering area, it switches to the humanoid interaction mode for random decision-making. The specific switching process is as follows: Prescan setting: Define the triggering area in the scenario and set the trigger point coordinates. The test background vehicle uses the TIS sensor to output its absolute position, and the autonomous driving host vehicle is equipped with sensors (camera or radar) to sense the test background vehicle.

[0030] Simulink model settings: Read the vehicle position of the test background vehicle from Prescan, determine whether the test background vehicle enters the trigger area (Relational Operator module), and use the Switch module to input 1 (default mode) for mode switching, and the acceleration output by the IDM car-following model. Input 2 (trigger mode): the acceleration output by the intelligent game model. Control signal: a Boolean signal generated by position triggering, and avoid repeated triggering to ensure that the mode remains in the switched state once triggered.

[0031] Each test background vehicle processes position detection and mode switching independently to ensure that its trigger logic and control model do not interfere with each other.

[0032] Finally, conduct a joint simulation of Prescan and Simulink. Bind the Prescan and Simulink interfaces to ensure the correct mapping of vehicle and sensor signals. Then route the position signals (X, Y) of each background vehicle to the corresponding trigger judgment module, and send the acceleration command after mode switching to the vehicle dynamics module to achieve mode switching.

[0033] Step 4: After passing through the intersection area, switch the test background vehicle back to the initial IDM car-following mode for the second time, and let the test background vehicle continue to drive on Route 1.

[0034] Specifically, the overall switching logic for the second switch is: initial mode, IDM car-following model (default). First switch, enter the trigger area → human-like interaction mode. Second switch, enter the secondary trigger area → switch back to the IDM mode.

[0035] Therefore, after passing through the conflict area, use position triggering again to switch the test background vehicle back to the initial IDM car-following mode and continue to drive on Route 1, that is, define the secondary trigger area through Prescan and use the Simulink switch module for trigger judgment.

[0036] Step 5: After completing a test cycle according to Steps 3 and 4, let the test background vehicle continue to drive on Route 1; after completing a test cycle according to Steps 3 and 4, the main autonomous vehicle continues to drive on Route 2. Until the test background vehicle reaches the intersection area next time, then repeat Steps 3 and 4 to perform mode switching again, and continuously monitor the algorithm performance reliability of the main autonomous vehicle in this cycle.

[0037] The preferred embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. The embodiments described in the present invention are only descriptions of the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. Among the various specific technical features described in the above specific embodiments, they can be combined in any suitable manner without contradiction. As long as such a combination does not violate the idea of the present invention, it should also be regarded as the content disclosed in this disclosure. To avoid unnecessary repetition, the present invention will not separately describe various possible combination methods.

[0038] The present invention is not limited to the specific details in the above embodiments. Without departing from the technical concept of the present invention and under the premise of not deviating from the design idea of the present invention, various modifications and improvements made by those skilled in the art to the technical solutions of the present invention should fall within the protection scope of the present invention. The technical content claimed by the present invention has been fully recorded in the claims.

Claims

1. A lightweight generation method for the test background vehicle-human interaction behavior in an unprotected left turn scenario, characterized in that: The following steps are involved: Step 1, setting the simulation condition to an unprotected left turn scenario on an urban road, where there are two closed loops, namely Route 1 and Route 2, and Route 1 and Route 2 have a common intersection section; Step 2: Set a number of test background vehicles on route 1. The test background vehicles are arranged in sequence along the driving direction. The distances between adjacent test background vehicles on route 1 are equal, and the test background vehicles can circulate on route 1. Set the autonomous driving main vehicle on route 2. The autonomous driving main vehicle can drive on route 2 and turn left at the intersection. Then the test background vehicle and the autonomous driving main vehicle turning left will conflict in the intersection area. Step 3: Perform simulation. During the simulation, each test background vehicle has two different control modes, namely, IDM following mode and human-like interaction mode. Before the simulation begins, all test background vehicles and the autonomous driving main vehicle are kept stationary on their respective routes; After the simulation starts, each test background vehicle is made to drive in a circle on route 1, and the autonomous driving main vehicle is made to drive on route 2 and turn left at the intersection; when the test background vehicle has not driven to the intersection area, the IDM following mode is used to control the leading vehicle among each test background vehicle to drive at a uniform speed of the IDM free flow speed, and the other test background vehicles follow along the driving direction; when any test background vehicle drives to the intersection area, it is switched to the human-like interaction mode, so that the test background vehicles driving to the intersection area randomly perform lane grabbing / yielding / game behaviors, which interfere with the autonomous driving main vehicle, thereby testing the reliability of the algorithm performance of the autonomous driving main vehicle; Step 4: After passing the intersection area, the test background vehicle is switched to the initial IDM following mode for the second time, and the test background vehicle continues to travel on route 1; Step 5: After completing a test cycle according to steps 3 and 4, the test background vehicle continues to drive on route 1; after completing a test cycle according to steps 3 and 4, the autonomous driving main vehicle continues to drive on route 2. Until the test background vehicle reaches the intersection area next time, then repeat steps 3 and 4, so as to continuously monitor the reliability of the algorithm performance of the autonomous driving main vehicle in this cycle.

2. The method for lightweight generation of test background vehicle-human interaction behavior in an unprotected left turn scenario according to claim 1, characterized in that: The scene described in step 1 is built by Prescan.

3. The method for lightweight generation of test background vehicle-human interaction behavior in an unprotected left turn scenario according to claim 1, characterized in that: In steps 2-5, each test background vehicle is equipped with a TIS sensor, thereby achieving safe vehicle following.

4. The method for lightweight generation of test background vehicle-human interaction behavior in an unprotected left turn scenario according to claim 1, characterized in that: In steps 3 and 4, the test background vehicle mode switching uses the switch module in Simulink, which takes the coordinates of the test background vehicle in the global coordinate system as input to achieve switching of different control modes, and then outputs the throttle and brake signals to the test background vehicle dynamics model for longitudinal speed control.

Images