Traffic scene simulation system

Abstract

The embodiment of the invention relates to a traffic scene simulation system. The system comprises a vehicle factory algorithm service, a scene simulation service, a scene database and a simulation database, the vehicle factory algorithm service is connected with the scene simulation service based on a gRPC protocol; the vehicle factory algorithm service and the scene simulation service are realized based on a container technology, and the vehicle factory algorithm service and the scene simulation service are respectively deployed in different container groups. According to the invention, the difficulty of development and deployment can be reduced, and the flexibility, universality and maintainability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a traffic scene simulation system. Background Art

[0002] Traffic scenario simulation systems are essential tools for testing autonomous driving algorithms. Conventional simulation systems often require the algorithm and simulation environment to be compiled into the same system software. While this approach ensures a tight coupling between the algorithm and the simulation environment, it also increases the difficulty of development and deployment, and reduces the system's flexibility, versatility, and maintainability. This drawback is particularly pronounced when the tester and the algorithm provider are not the same organization. Summary of the Invention

[0003] The purpose of the present invention is to address the shortcomings of the existing technology and provide a traffic scene simulation system, which includes: a car factory algorithm service (hereinafter referred to as the algorithm side), a scene simulation service (hereinafter referred to as the simulation side), a scene database and a simulation database; the algorithm and simulation sides communicate data through the gRPC protocol; the algorithm and simulation sides are implemented based on container (Docker) technology and are respectively deployed in different container groups (Pods), and the topic publish-subscribe mechanism based on the ROS (Robot Operating System) architecture provides a data interaction channel for the internal components of the algorithm and simulation sides. After the algorithm side sends a start instruction to the simulation side, it preheats based on the preheating data packet sent by the simulation side; after the preheating is completed, it sends a step instruction to the simulation side and performs the next time single-step prediction-planning-decision-control prediction process based on the next step perception data fed back by it, and repeats this cycle until the algorithm side sends an end instruction to the simulation side to complete the entire test process. On the simulation side, the scenario unit handles start and end commands: 1) Upon receiving a start command, the scenario unit selects and loads scenario data from the scenario database, extracts the initial scenario fragment, generates a warm-up data packet, and sends it to the algorithm side. 2) Upon receiving an end command, the scenario unit generates test scenario data based on the latest simulation scenario and stores it in the simulation database. On the simulation side, the simulation unit handles step commands. For each step command, the simulation unit simulates the ego vehicle's motion state for the next time step based on the control parameters in the command (such as steering wheel angle, throttle position, brake position), the real-time environment state of the simulation scene (the motion state of other traffic participants, traffic light status), and the real-time state of the ego vehicle. During simulation, the Runge-Kutta method is used to predict the state of the ego vehicle in the simulation scene to improve data accuracy. Based on the simulation results, the ego vehicle state in the simulation scene is updated. The next step of perception data, composed of information such as the environment state, the latest ego vehicle state, and high-precision maps, is fed back to the algorithm side. Furthermore, the simulation unit updates the motion state of other traffic participants in the simulation scene after completing a specified number (M) time steps of ego vehicle state updates to ensure the overall consistency of the simulation world. This paper separates the algorithm from the simulation by introducing Docker technology, ensures tight coupling between the algorithm and the simulation environment through stepping instructions, and improves the realism, flexibility, and overall consistency of the scenario by linking the scenario unit with the simulation unit. This paper not only ensures testing efficiency, but also reduces the difficulty of development and deployment, and improves the flexibility, versatility, and maintainability of the system.

[0004] To achieve the above-mentioned purpose, an embodiment of the present invention provides a traffic scene simulation system, the system comprising: a vehicle factory algorithm service, a scene simulation service, a scene database and a simulation database;

[0005] The depot algorithm service is connected to the scenario simulation service based on the gRPC protocol; the depot algorithm service and the scenario simulation service are implemented based on container technology, and the depot algorithm service and the scenario simulation service are respectively deployed in different container groups;

[0006] The depot algorithm service is used to generate a corresponding start instruction according to the test task information input by the user and send it to the gRPC server of the scenario simulation service; and preheat the algorithm service based on the preheating data packet sent by the gRPC server and send the first step step instruction to the gRPC server at the end of the preheating;

[0007] The vehicle manufacturer algorithm service is further configured to complete a single-step vehicle control prediction based on the next-step perception data sent by the gRPC server after each step instruction is sent, and obtain the next step instruction or end instruction and send it to the gRPC server;

[0008] The scene simulation service includes a scene unit, the gRPC server, a simulation unit, a map unit, and a simulation ROS service; the scene unit is connected to the scene and the simulation database respectively; the simulation ROS service is used to provide a data interaction channel for the scene unit, the gRPC server, the simulation unit, and the map unit through the topic publish-subscribe mechanism of the ROS architecture;

[0009] The gRPC server is used to forward the start instruction to the scene unit, and send the warm-up data packet sent back by it to the vehicle factory algorithm service; it is also used to forward the step instruction to the simulation unit, and send the next step perception data corresponding to the environmental perception data and vehicle status data sent back by the simulation unit and the map perception data sent by the map unit to the vehicle factory algorithm service; it is also used to forward the end instruction to the scene unit;

[0010] The scene unit is used to load the simulation world data body and prepare preheating data according to the startup instruction and the scene database to obtain the preheating data packet and send it back to the gRPC client; and initialize the data body copy of the simulation unit and the global high-precision map of the map unit;

[0011] The scene unit is further configured to refresh the currently loaded simulation world data volume according to the data volume copy sent by the simulation unit and reset the simulation unit copy based on the refresh result;

[0012] The scene unit is further configured to, upon receiving the end instruction, call the latest copy of the data volume from the simulation unit to refresh the currently loaded simulation world data volume and generate corresponding simulation data based on the refresh result and store it in the simulation database;

[0013] The simulation unit is configured to set a continuous counter initialized to 0 when the step instruction is received for the first time; perform a single-step time-step ego-vehicle motion state simulation based on the data body copy and the step instruction received each time according to the Runge-Kutta method, and refresh the data body copy based on the simulation result; add 1 to the continuous counter each time the copy refresh is completed; and identify whether the continuous counter after adding 1 is a preset number M; if not, use the current data body copy as the latest copy; if so, clear the continuous counter, send the current data body copy to the scene unit, and then use the simulated world data body sent back by the scene unit this time as the latest copy and save it; generate the environmental perception data and the ego-vehicle state data based on the latest copy, send them to the gRPC server, and send the latest ego-vehicle label route to the map unit; the preset number M is a positive integer;

[0014] The map unit is used to generate the corresponding map perception data based on the global high-precision map and the vehicle label route sent by the simulation unit and send it to the gRPC server.

[0015] Preferably, the vehicle factory algorithm service includes a gRPC client, a prediction unit, a planning unit, a decision unit, a control unit and an algorithm ROS service; the algorithm ROS service is used to provide a data interaction channel for the gRPC client, the prediction unit, the planning unit, the decision unit and the control unit through the message topic publish-subscribe mechanism of the ROS architecture; the gRPC client is connected to the gRPC server of the scenario simulation service based on the gRPC protocol;

[0016] The scenario database is used to store multiple test scenarios; each of the test scenarios includes a unique scenario identifier and a simulation world data body of a traffic scenario; the simulation world data body includes a high-precision map of the simulation environment and a simulation object set {O i}, 1≤index i≤N O , N O The simulation environment high-precision map includes a high-precision map of all roads in the simulation environment; the simulation object set {O i}Includes multiple simulation objects O i ; The simulation object O i Include object type d i 、Object motion trajectory {si,t}, 1≤ time step t≤N T , N T is the total number of steps, and the time interval between each two adjacent time steps t is fixed to interval L; the object type d i At least including cars, motorcycles, bicycles, pedestrians, obstacles, and traffic lights; the object type d i For cars, motorcycles, bicycles, pedestrians, obstacles, s i,t The object motion state of the current simulation object at time t, the object motion state at least includes coordinates, orientation angle, velocity and acceleration; the object type d i When it is a signal light, s i,t The motion state of the traffic light of the current simulation object at time t, wherein the motion state of the traffic light at least includes coordinates and the state of the traffic light, and the state of the traffic light at least includes red light, green light, and yellow light;

[0017] The preheating data packet consists of the target position of the vehicle and the next step perception data sequence; the next step perception data sequence is composed of n next step perception data sorted in chronological order, n is a preset positive integer, n<N T The next-step perception data includes the environmental perception data, the vehicle status data and the map perception data; the environmental perception data includes a traffic participant data set and signal light data; the traffic participant data set is composed of a plurality of traffic participant data, and the traffic participant data includes object type, coordinates, heading angle, speed and acceleration; the signal light data includes coordinates and the signal light status; the vehicle status data includes coordinates, heading angle, speed and acceleration; the map perception data includes at least lane lines, lane center lines, traffic signs / markings / markings, and road edge lines in the four directions of the front, back, left and right of the road where the vehicle is traveling;

[0018] The simulation database is used to store a plurality of simulation data; each simulation data consists of a simulation timestamp, a simulation scene identifier, a test vehicle identifier and a post-test data body.

[0019] Preferably, the vehicle manufacturer algorithm service is specifically used to generate a corresponding start instruction based on the test task information input by the user and send it to the gRPC server of the scenario simulation service: the gRPC client receives the test task information input by the user, and extracts the corresponding test scenario identifier and self-vehicle identifier from the test task information, and sends the test start instruction carrying the test scenario identifier and the self-vehicle identifier to the gRPC server.

[0020] Preferably, the vehicle factory algorithm service is specifically used to preheat the algorithm service based on the preheating data packet sent by the gRPC server and send the first step instruction to the gRPC server at the end of the preheating:

[0021] The gRPC client receives the preheating data packet sent by the gRPC server, and extracts the corresponding vehicle target position and the next-step perception data sequence from the preheating data packet; and sends the vehicle target position to the planning unit; and uses each next-step perception data of the next-step perception data sequence as the corresponding current perception data; and extracts the corresponding traffic participant data set, the traffic light data, the vehicle status data and the map perception data from the current perception data; and sends the current traffic participant data set, the traffic light data and the map perception data to the prediction unit; and sends the current vehicle status data and the map perception data to the planning unit; and waits for receiving the vehicle control instruction sent by the control unit; and after receiving the vehicle control instruction, continues to process the next next-step perception data as the new current perception data until the processing of the last next-step perception data is completed, and sends the vehicle control instruction sent by the control unit for the last time as the step instruction of the first step to the gRPC server.

[0022] Preferably, the vehicle manufacturer algorithm service is specifically used to complete a single-step vehicle control prediction based on the next-step perception data sent by the gRPC server after each step instruction is sent, and obtain the next step instruction or end instruction to send to the gRPC server:

[0023] After sending each step instruction, the gRPC client waits for the next step perception data sent by the RPC server based on a preset first waiting time; and confirms whether the next step perception data can be received within the first waiting time; if it is confirmed that the next step perception data is not received within the first waiting time, the end instruction is sent to the gRPC server; if it is confirmed that the next step perception data is received within the first waiting time, the corresponding traffic participant data set, the signal light data, the vehicle status data and the map perception data are extracted from the current next step perception data, and the current traffic participant data set, the signal light data and the corresponding The map perception data is sent to the prediction unit, and the current vehicle state data and the map perception data are sent to the planning unit, and the vehicle control instruction sent by the control unit is waited for and received, and whether the vehicle control instruction can be received within the preset second waiting time is confirmed. If it is confirmed that the vehicle control instruction is not received within the second waiting time, the end instruction is sent to the gRPC server. If it is confirmed that the vehicle control instruction is received within the second waiting time, the current vehicle control instruction is sent to the gRPC server as the next step instruction; the instruction parameters of the vehicle control instruction include at least the steering wheel angle, the throttle opening and closing, and the brake opening and closing.

[0024] Preferably, the prediction unit is configured to, upon receiving each set of the traffic participant data set, the traffic light data, and the map perception data, predict the movement trajectory of each traffic participant in a future specified time period based on the traffic participant data set, the traffic light data, and the map perception data received this time, to obtain a corresponding participant trajectory; and to form a corresponding participant trajectory set from all the obtained participant trajectories and send it to the planning unit; the time length of the future specified time period is greater than the interval L;

[0025] The planning unit is configured to refresh the stored ego vehicle target position based on the current ego vehicle target position each time the ego vehicle target position is received from the gRPC client;

[0026] The planning unit is further configured to refresh the stored vehicle state data and map perception data based on the current vehicle state data and map perception data each time a set of the vehicle state data and the map perception data sent by the gRPC client is received;

[0027] The planning unit is further configured to, upon receiving each participant trajectory set sent by the prediction unit, plan a movement trajectory of the ego vehicle within the future specified time period based on the current participant trajectory set and the most recently stored ego vehicle target position, the ego vehicle state data, and the map perception data, obtain a corresponding ego vehicle trajectory, and send the plan to the decision unit;

[0028] The decision unit is configured to, upon receiving each of the vehicle trajectories sent by the planning unit, determine a driving behavior type of the vehicle based on the current trajectory of the vehicle to obtain a corresponding vehicle behavior type; and transmit the vehicle behavior type and the trajectory of the vehicle to the control unit; the vehicle behavior types at least include decelerating and going straight, accelerating and going straight, going straight at a constant speed, changing lanes to the left, changing lanes to the right, overtaking, turning around, avoiding to the left, avoiding to the right, sudden braking, and stopping;

[0029] The control unit is configured to, upon receiving a set of the vehicle behavior type and the vehicle trajectory sent by the decision unit, predict the lateral and longitudinal driving control parameters of the vehicle at the next time step based on the current vehicle behavior type and the vehicle trajectory, obtain the corresponding vehicle control instruction, and send it to the gRPC client.

[0030] Preferably, the scene unit is specifically configured to, when the simulation world data body is loaded and preheating data is prepared according to the startup instruction and the scene database to obtain the preheating data packet and send it back to the gRPC client:

[0031] Extracting and saving the corresponding test scene identifier and vehicle identifier from the startup instruction; and using the simulation environment high-precision map and the simulation world data volume of the test scene in the scene database that match the scene identifier with the test scene identifier as the corresponding current high-precision map and current data volume; and loading the current data volume;

[0032] And the simulation object set {O i} as the corresponding current object set; and in the current object set, the simulation object O whose index i matches the self-vehicle identifier i Extract it as the corresponding vehicle object; and collect the object types d in the current object collection i The simulation object O that is not a traffic light and whose index i does not match the vehicle identifier i Extract it as the corresponding participant object; and collect the object types d in the current object i The simulation object O of the signal light i Extract it as the corresponding traffic light object; and extract the object motion trajectory of the vehicle object in the current high-precision map The local high-precision map passed by is used as the corresponding local map of the vehicle; and the object motion trajectory is The last motion state of the object The coordinates of as the corresponding target position of the vehicle;

[0033] The object motion trajectories of the vehicle object, each of the participant objects, and each of the signal light objects are {s i,t}, the object or signal light motion state s where the time step t is greater than n i,t Delete; and the object motion trajectory {s i,1≤t≤n} Deleting the participant and traffic light objects that do not intersect with the map area of ​​the local map of the vehicle;

[0034] The object motion state s of each participant object at each time step t i,t and the corresponding object type d i Forming a corresponding traffic participant data; and forming a corresponding traffic participant data set by the traffic participant data of all the participant objects at each time step t;

[0035] The object motion state s of the vehicle object at each time step t is i,t Forming a corresponding vehicle state data;

[0036] and identifying the number of signal light objects; if the number of signal light objects is zero, setting an empty signal light data as the corresponding signal light data for time step t; if the number of signal light objects is 1, the coordinates of the unique signal light object at each time step t and the signal light state constitute a corresponding signal light data; if the number of signal light objects is greater than 1, then at each time step t, a signal light object located in front of the travel path of the current vehicle object coordinates and closest to the current vehicle object coordinates is used as the corresponding current single-step signal light, and the coordinates of the current single-step signal light at the current time step and the signal light state constitute a signal light data corresponding to the current time step;

[0037] Extracting multiple types of map elements from the local map of the vehicle to form a corresponding map element set; and using the map element set as the map perception data corresponding to each time step t; the multiple types of map elements at least include lane lines, lane center lines, traffic signs / markings / markings, and road edge lines;

[0038] The traffic participant data set and the traffic light data corresponding to each time step t constitute a corresponding environmental perception data; the environmental perception data, the vehicle status data and the map perception data corresponding to each time step t constitute a corresponding next-step perception data; the obtained n next-step perception data are sorted in chronological order to constitute a corresponding next-step perception data sequence; and the vehicle target position and the next-step perception data sequence constitute the corresponding preheating data packet and send it back to the gRPC client.

[0039] Preferably, the scene unit is specifically used to send the vehicle identification and the currently loaded simulation world data body to the simulation unit when initializing the data body copy of the simulation unit and the global high-precision map of the map unit; and send the vehicle local map to the map unit.

[0040] Preferably, the simulation unit is further configured to save the received simulation world data body as the latest local copy of the data body upon receiving the vehicle identification and the simulation world data body sent by the scene unit; and save the received vehicle identification as the latest local simulation vehicle identification; and based on the simulation object set {O i The simulation object O whose index i matches the simulation vehicle identifier i As the corresponding simulation vehicle object; and initialize the single-step counter C to n; and the sub-object motion trajectory of the simulation vehicle object {s i,1≤t≤n} is an observation quantity, which initializes the model parameters of the built-in vehicle dynamics model; and after the model parameter initialization is completed, waits for the reception of the first step instruction.

[0041] Preferably, the map unit is further configured to save the local map of the vehicle received this time as the latest local global high-precision map when receiving the local map of the vehicle sent by the scene unit.

[0042] Preferably, the simulation unit is specifically configured to: when performing a single-step time-step ego-vehicle motion state simulation according to the data volume copy and the stepping instruction received each time according to the Runge-Kutta method and refreshing the data volume copy based on the simulation result: i,t＝C As the corresponding starting motion state s C ; and extract the corresponding steering wheel angle, throttle opening and closing, brake opening and closing from the current stepping instruction as the corresponding starting steering wheel angle, starting throttle opening and closing, starting brake opening and closing; and according to the starting motion state s according to the Runge-Kutta method C, the initial steering wheel angle, the initial throttle opening and closing degree, and the initial brake opening and closing degree for the motion state s at the next time step t = C + 1 C+1 Make a prediction; and based on the predicted motion state s C+1 The object motion state s of the simulated vehicle object in the data volume copy i,t＝C+1 Reset; and add 1 to the single-step counter C.

[0043] Preferably, the simulation unit is specifically configured to, when the environmental perception data and the vehicle status data are generated based on the latest copy and sent to the gRPC server and the latest vehicle label route is sent to the map unit:

[0044] The simulation object set of the latest copy {O i} as the corresponding current object set; and the simulation object O whose index i matches the simulation vehicle identifier in the current object set i As the corresponding current simulation vehicle object; and the current object is concentrated in the object type d i The simulation object O that is not a signal light and is not the current simulation vehicle object i As the corresponding participant object; and the object type d of the current object i The simulation object O of the signal light i As the corresponding signal light object; and based on the single-step counter C, the object motion state s of the current simulation vehicle object i,t＝C As a corresponding vehicle state data;

[0045] The coordinates of the vehicle state data are marked as the current vehicle coordinates; the signal light object whose coordinates are located in front of the driving road of the current vehicle coordinates, closest to the current vehicle coordinates, and whose distance from the current vehicle coordinates is less than a preset first distance threshold is recorded as the current single-step signal light; and whether the current single-step signal light does not exist is confirmed, if so, an empty signal light data is set as the corresponding signal light data, otherwise the signal light motion state s of the current single-step signal light is recorded. i,t＝C As the corresponding signal light data;

[0046] and the object motion state s of each participant object i,t＝C The coordinates of the participants are taken as the corresponding coordinates of the participants; and the coordinates of the participants whose distance from the current vehicle coordinates is less than the preset second distance threshold are marked as first coordinates; and the object type d corresponding to each first coordinate is i and the object motion state s i,t＝Cforming a corresponding traffic participant data set; and forming a corresponding traffic participant data set from all the traffic participant data obtained;

[0047] The traffic participant data set and the traffic light data obtained this time are combined into a corresponding piece of environmental perception data; and the environmental perception data and the vehicle status data obtained this time are sent to the gRPC server;

[0048] And the object motion trajectory of the current simulated vehicle object {s i,C-1≤t The coordinates of the vehicle are extracted and sorted in order to form the corresponding vehicle label route and sent to the map unit.

[0049] Preferably, the map unit is specifically used for: when the corresponding map perception data is generated based on the global high-precision map and the self-vehicle label route sent by the simulation unit and sent to the gRPC server: using the local high-precision map through which the self-vehicle label route passes in the global high-precision map as the corresponding current local map; and extracting multiple types of map elements on the current local map to form a corresponding current element set; and sending the current element set as the corresponding map perception data to the gRPC server; the multiple types of map elements include at least lane lines, lane center lines, traffic signs / signs / markings, and road edge lines.

[0050] Preferably, the scene unit is specifically configured to, when refreshing the currently loaded simulation world data volume according to the data volume copy sent by the simulation unit and resetting the simulation unit copy based on the refresh result:

[0051] The simulation object set of the data body copy {O i} as the corresponding current object set; and the simulation object O whose index i matches the vehicle identifier in the current object set i As the corresponding current vehicle object; and the current object is concentrated in each of the object types d i The simulation object O of the car i As the corresponding other vehicle objects; and the M object motion states s that are most recently updated by the current self-vehicle object i,t Composed of the corresponding vehicle history trajectory; and each of the object motion trajectories of the other vehicle objects {s i,t} extract the sub-motion trajectory synchronized with the historical trajectory of the own vehicle as the corresponding historical trajectory of other vehicles; and record the historical trajectory of other vehicles that are on the same road section / traffic intersection as the historical trajectory of the own vehicle as the first historical trajectory; and record the historical trajectory of other vehicles that are not on the same road section / traffic intersection as the historical trajectory of the own vehicle as the second historical trajectory;

[0052] The built-in vehicle dynamics model predicts the object motion state s of the vehicle in the next M time steps based on the vehicle's historical trajectory. i,t Predict and use the predicted M motion states s of the object i,t Update the object motion trajectory of the current vehicle object i,t};

[0053] The built-in vehicle dynamics model predicts the object motion state s of the vehicle in the next M time steps according to each of the second historical trajectories. i,t Predict and use the predicted M motion states s of the object i,t Update the object motion trajectory of the corresponding other vehicle object {s i,t};

[0054] The built-in traffic participant behavior model analyzes the driving behavior type of the other vehicle objects corresponding to each of the first historical trajectories in the future M time steps based on the historical trajectory of the vehicle and all the first historical trajectories to obtain the corresponding other vehicle behavior type; and the built-in vehicle dynamics model analyzes the object motion state s of the vehicle in the future M time steps based on each of the other vehicle behavior types and their corresponding first historical trajectories. i,t Predict and use the predicted M motion states s of the object i,t Update the object motion trajectory of the corresponding other vehicle object {s i,t The other vehicle behavior types include slowing down and going straight, accelerating and going straight, going straight at a constant speed, changing lanes to the left, changing lanes to the right, overtaking, turning around, avoiding to the left, avoiding to the right, sudden braking, and stopping;

[0055] The updated data body copy is loaded to become the latest simulation world data body; and the updated data body copy is sent back to the simulation unit as the latest data body copy on the simulation unit.

[0056] Preferably, the scene unit is specifically configured to, when the latest copy of the data volume is called from the simulation unit to refresh the currently loaded simulation world data volume and generate corresponding simulation data based on the refresh result and store it in the simulation database:

[0057] Call the latest copy of the data body from the simulation unit as the corresponding current copy; and set the simulation object set of the current copy {O i} as the corresponding current object set; and the simulation object O whose index i matches the vehicle identifier in the current object set iAs the corresponding current vehicle object; and the current object is concentrated in each of the object types d i The simulation object O of the car i As the corresponding other vehicle objects; and the M object motion states s that are most recently updated by the current self-vehicle object i,t Composition of the corresponding historical trajectory of the vehicle;

[0058] The last time step t of the vehicle's historical trajectory is recorded as the corresponding current step t * ; and for the current single step t * Is it equal to the total number of steps N? T If yes, the current copy is used as the corresponding first copy; if not, the current object is set to each of the simulation objects O i The object motion trajectory {s i,t}'s sub-motion trajectory Deleting the current copy after deletion and using it as the corresponding first copy;

[0059] and loading the first copy into the latest simulation world data body;

[0060] And use the first copy as the corresponding post-test data body; use the test scene identifier as the corresponding simulation scene identifier, and use the self-vehicle identifier as the corresponding test vehicle identifier; and use the current time as the corresponding simulation timestamp; and the obtained simulation timestamp, simulation scene identifier, test vehicle identifier and post-test data body to form a corresponding simulation data and store it in the simulation database.

[0061] An embodiment of the present invention provides a traffic scenario simulation system. As can be seen from the above, the system includes: a vehicle manufacturer algorithm service, a scenario simulation service, a scenario database, and a simulation database. The algorithm and simulation sides communicate data via the gRPC protocol. The algorithm and simulation sides are implemented using Docker technology and deployed in separate pods. The publish-subscribe topic mechanism of the ROS architecture provides a data exchange channel between the internal components of the algorithm and simulation sides. After the algorithm sends a start command to the simulation side, it performs a preheating operation based on a preheating data packet sent by the simulation side. After the preheating is complete, it sends a step command to the simulation side and, based on the next step of the feedback, performs the prediction-planning-decision-control prediction process for the next time step. This cycle repeats until the algorithm sends a stop command to the simulation side, completing the entire testing process. The simulation side's scenario unit processes the start and end commands: 1) Upon receiving the start command, the scenario unit selects scenario data from the scenario database for loading, extracts the initial scenario segment, generates a preheating data packet, and sends it to the algorithm side. 2) Upon receiving the stop command, the scenario unit generates test scenario data based on the latest simulation scenario and stores it in the simulation database. On the simulation side, the simulation unit processes step commands. Each time a step command is received, the simulation unit simulates the vehicle's motion state for the next time step based on the control parameters in the command (such as steering wheel angle, throttle position, brake position), the real-time environmental state of the simulation scene (the motion state of other traffic participants, signal light status), and the real-time state of the ego vehicle. During the simulation, the Runge-Kutta method is used to process the prediction process to improve data accuracy. Based on the simulation results, the ego vehicle's state in the simulation scene is refreshed. The next step of perception data, composed of information such as the environmental state, the latest ego vehicle state, and high-precision maps, is fed back to the algorithm. Furthermore, the simulation unit updates the motion state of other traffic participants in the simulation scene after completing a specified number (M) of ego vehicle state refreshes to ensure the overall consistency of the simulated world. This embodiment of the present invention achieves the separation of algorithm and simulation by introducing Docker technology. Step commands ensure a tight coupling between the algorithm and the simulation environment. The linkage between the scene unit and the simulation unit improves the scene's realism, flexibility, and overall consistency. The present invention not only ensures the test efficiency, but also reduces the difficulty of development and deployment, and improves the flexibility, versatility and maintainability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A module structure diagram of a traffic scene simulation system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0063] To make the objectives, technical solutions, and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the embodiments described herein are merely some, rather than all, of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0064] The embodiment of the present invention provides a traffic scene simulation system, such as Figure 1 As shown in the module structure diagram of a traffic scene simulation system provided by an embodiment of the present invention, it mainly includes: a vehicle factory algorithm service 1, a scene simulation service 2, a scene database 3 and a simulation database 4.

[0065] Here, the car factory algorithm service 1 and the scenario simulation service 2 of the embodiment of the present invention are each implemented based on container technology and deployed in different container groups (Pods), and the containers are orchestrated and managed based on Kubernetes (K8 for short); data communication is carried out between the car factory algorithm service 1 and the scenario simulation service 2 based on the gRPC protocol; within the car factory algorithm service 1 and the scenario simulation service 2, the topic publish-subscribe mechanism based on the ROS architecture provides a data interaction channel for their internal components.

[0066] (1) Car factory algorithm service 1:

[0067] The car factory algorithm service 1 of an embodiment of the present invention includes a gRPC client 11, a prediction unit 12, a planning unit 13, a decision unit 14, a control unit 15 and an algorithm ROS service 16; the algorithm ROS service 16 is used to provide a data interaction channel for the gRPC client 11, the prediction unit 12, the planning unit 13, the decision unit 14 and the control unit 15 through the message topic publish-subscribe mechanism of the ROS architecture; the gRPC client 11 is connected to the gRPC server 22 of the scenario simulation service 2 based on the gRPC protocol.

[0068] The car factory algorithm service 1 is used to generate the corresponding startup instructions based on the test task information input by the user and send them to the gRPC server 22 of the scene simulation service 2; and preheat the algorithm service based on the preheating data packet sent by the gRPC server 22 and send the first step step instruction to the gRPC server 22 at the end of the preheating.

[0069] The vehicle manufacturer algorithm service 1 is also used to complete a single-step self-vehicle control prediction based on the next-step perception data sent by the gRPC server 22 after each step instruction is sent, and obtain the next step instruction or end instruction to be sent to the gRPC server 22.

[0070] In a specific implementation method of an embodiment of the present invention, the vehicle manufacturer algorithm service 1 is specifically used to generate a corresponding start instruction based on the test task information input by the user and send it to the gRPC server 22 of the scenario simulation service 2: the gRPC client 11 receives the test task information input by the user, and extracts the corresponding test scenario identifier and vehicle identifier from the test task information, and sends the test start instruction carrying the test scenario identifier and vehicle identifier to the gRPC server 22.

[0071] In another specific implementation of the embodiment of the present invention, the depot algorithm service 1 is specifically used to preheat the algorithm service based on the preheating data packet sent by the gRPC server 22 and send the first step step instruction to the gRPC server 22 at the end of the preheating: the gRPC client 11 receives the preheating data packet sent by the gRPC server 22, and extracts the corresponding self-vehicle target position and the next step perception data sequence from the preheating data packet; and sends the self-vehicle target position to the planning unit 13; and uses each next step perception data of the next step perception data sequence as the corresponding current perception data; and extracts the corresponding traffic parameters from the current perception data. The data set of traffic participants, signal light data, vehicle status data and map perception data are sent to the prediction unit 12; the current vehicle status data and map perception data are sent to the planning unit 13; and the vehicle control instruction sent by the control unit 15 is waited for to be received; and after receiving the vehicle control instruction, the next next-step perception data is continued to be processed as the new current perception data until the last next-step perception data is processed, and the vehicle control instruction sent by the control unit 15 for the last time is sent to the gRPC server 22 as the first step instruction.

[0072] Here, the preheating data packet of the embodiment of the present invention is composed of the target position of the vehicle and the next step perception data sequence; the next step perception data sequence is composed of n next step perception data sorted in chronological order, n is a preset positive integer, n<N T , N T As can be seen from the following text, it is the total number of steps in a single step of the scene time; the next step perception data includes environmental perception data, vehicle status data and map perception data; environmental perception data includes traffic participant data set and signal light data; traffic participant data set consists of multiple traffic participant data, and traffic participant data includes object type, coordinates, heading angle, speed and acceleration; signal light data includes coordinates and signal light status; vehicle status data includes coordinates, heading angle, speed and acceleration; map perception data includes at least lane lines, lane center lines, traffic signs / markings / markings, and road edge lines in the four directions of the front, back, left and right of the vehicle's driving road.

[0073] In another specific implementation of the embodiment of the present invention, the vehicle manufacturer algorithm service 1 is specifically configured to complete a single-step vehicle control prediction based on the next-step perception data sent by the gRPC server 22 after each step instruction is sent, and obtain the next step instruction or end instruction to send to the gRPC server 22:

[0074] After each step instruction is sent, the gRPC client 11 waits for the next step perception data sent by the RPC server based on the preset first waiting time; and confirms whether the next step perception data can be received within the first waiting time; if it is confirmed that the next step perception data is not received within the first waiting time, the end instruction is sent to the gRPC server 22; if it is confirmed that the next step perception data is received within the first waiting time, the corresponding traffic participant data set, signal light data, vehicle status data and map perception data are extracted from the current next step perception data, and the current traffic participant data set is sent to the gRPC server 22. The vehicle data set, traffic light data and map perception data are sent to the prediction unit 12, and the current vehicle status data and map perception data are sent to the planning unit 13. The vehicle control instruction sent by the control unit 15 is waited for and received, and whether the vehicle control instruction can be received within the preset second waiting time is confirmed. If it is confirmed that the vehicle control instruction is not received within the second waiting time, the end instruction is sent to the gRPC server 22. If it is confirmed that the vehicle control instruction is received within the second waiting time, the current vehicle control instruction is sent to the gRPC server 22 as the next step instruction.

[0075] Here, the first and second waiting time periods in the embodiment of the present invention are two preset time length parameters.

[0076] The prediction unit 12 of the embodiment of the present invention is used to predict the movement trajectory of each traffic participant in a specified future time period based on the traffic participant dataset, signal light data and map perception data received this time, and obtain the corresponding participant trajectory; and form a corresponding participant trajectory set composed of all the obtained participant trajectories to be sent to the planning unit 13; here, the time length of the specified future time period is greater than the interval L.

[0077] The planning unit 13 of the embodiment of the present invention is used to refresh the stored ego vehicle target position based on the current ego vehicle target position each time a ego vehicle target position sent by a gRPC client 11 is received. The planning unit 13 of the embodiment of the present invention is also used to refresh the stored ego vehicle status data and map perception data based on the current ego vehicle status data and map perception data each time a group of ego vehicle status data and map perception data sent by a gRPC client 11 is received. The planning unit 13 of the embodiment of the present invention is also used to plan the movement trajectory of the ego vehicle in the future specified time period based on the current participant trajectory set and the latest stored ego vehicle target position, ego vehicle status data and map perception data each time a participant trajectory set sent by the prediction unit 12 is received, and obtain the corresponding ego vehicle trajectory and send it to the decision unit 14.

[0078] The decision unit 14 of this embodiment of the present invention is configured to, upon receiving each vehicle trajectory from the planning unit 13, determine the vehicle's driving behavior type based on the current vehicle trajectory to obtain a corresponding vehicle behavior type, and transmit the vehicle behavior type and trajectory to the control unit 15. The vehicle behavior types in this embodiment of the present invention include at least slowing down and going straight, accelerating and going straight, going straight at a constant speed, changing lanes to the left, changing lanes to the right, overtaking, turning around, avoiding to the left, avoiding to the right, sudden braking, and stopping.

[0079] The control unit 15 of this embodiment of the present invention is configured to, upon receiving a set of vehicle behavior types and ego vehicle trajectories from the decision unit 14, predict the lateral and longitudinal driving control parameters of the ego vehicle at the next time step based on the current vehicle behavior type and trajectory, obtain corresponding ego vehicle control instructions, and send them to the gRPC client 11. The command parameters of the ego vehicle control instructions in this embodiment of the present invention include at least steering wheel angle, throttle opening and closing, and brake opening and closing.

[0080] (2) Scenario Simulation Service 2:

[0081] The scene simulation service 2 of an embodiment of the present invention includes a scene unit 21, a gRPC server 22, a simulation unit 23, a map unit 24 and a simulation ROS service 25; the scene unit 21 is connected to the scene database 3 and the simulation database 4 respectively; the simulation ROS service 25 is used to provide a data interaction channel for the scene unit 21, the gRPC server 22, the simulation unit 23 and the map unit 24 through the topic publish-subscribe mechanism of the ROS architecture.

[0082] The gRPC server 22 is used to forward the start instruction to the scene unit 21, and send the warm-up data packet sent back by it to the vehicle factory algorithm service 1; it is also used to forward the step instruction to the simulation unit 23, and send the next step perception data corresponding to the environmental perception data and vehicle status data sent back by the simulation unit 23 and the map perception data sent by the map unit 24 to the vehicle factory algorithm service 1; it is also used to forward the end instruction to the scene unit 21.

[0083] The scenario unit 21 is used to load the simulation world data volume and preheat data according to the startup instructions and the scenario database 3, and then send the resulting preheat data packet back to the gRPC client 11. It also initializes the data volume copy of the simulation unit 23 and the global high-precision map of the map unit 24. The scenario unit 21 in this embodiment of the present invention is a scenario player (ScenerioPlayer), such as CARLA, Apollo Scenario Player, OpenCDA Scenario Player, etc.

[0084] In another specific implementation of an embodiment of the present invention, the scene unit 21 is specifically used to send the vehicle identification and the currently loaded simulation world data body to the simulation unit 23 when initializing the data body copy of the simulation unit 23 and the global high-precision map of the map unit 24; and send the vehicle local map to the map unit 24.

[0085] The scene unit 21 is further configured to refresh the currently loaded simulation world data volume according to the data volume copy sent by the simulation unit 23 and reset the copy of the simulation unit 23 based on the refresh result.

[0086] The scene unit 21 is further configured to call the latest data volume copy from the simulation unit 23 to refresh the currently loaded simulation world data volume when receiving the end instruction and generate corresponding simulation data based on the refresh result and store it in the simulation database 4 .

[0087] The simulation unit 23 is used to save the received simulated world data body as the latest local data body copy when receiving the self-vehicle identification and simulated world data body sent by the scene unit 21; and save the received self-vehicle identification as the latest local simulated vehicle identification; and based on the simulation object set {O i} where index i matches the simulated vehicle ID of the simulated object O i As the corresponding simulation car object; and initialize the single-step counter C to n; and use the motion trajectory of the sub-object of the simulation car object {s i,1≤t≤n} is the observation quantity, which initializes the model parameters of the built-in vehicle dynamics model; and after the model parameters are initialized, it waits for the first step instruction to be received.

[0088] The simulation unit 23 is further configured to set a continuous counter initialized to 0 when a step instruction is received for the first time; and perform a single-step time simulation of the vehicle's motion state according to the data body copy and each received step instruction according to the Runge-Kutta method, and refresh the data body copy based on the simulation result; and add 1 to the continuous counter each time the copy refresh is completed; and identify whether the continuous counter after adding 1 is a preset number M; if not, the current data body copy is used as the latest copy; if so, the continuous counter is cleared to zero, and the current data body copy is sent to the scene unit 21, and the simulation world data body sent back by the scene unit 21 this time is used as the latest copy and saved; and generate environmental perception data and vehicle status data based on the latest copy and send them to the gRPC server 22, and send the latest vehicle label route to the map unit 24; the preset number M is a positive integer.

[0089] Here, the Runge-Kutta method is a method for solving nonlinear ordinary differential equations. Compared with the conventional Euler method, the Runge-Kutta method has higher accuracy.

[0090] The map unit 24 is further configured to save the received local map of the vehicle as the latest local global high-precision map when receiving the local map of the vehicle sent by the scene unit 21 .

[0091] The map unit 24 is used to generate corresponding map perception data based on the global high-precision map and the vehicle label route sent by the simulation unit 23 and send it to the gRPC server 22.

[0092] In another specific implementation of the embodiment of the present invention, the scene unit 21 is specifically configured to load the simulation world data body and prepare the preheating data according to the startup instruction and the scene database 3 to obtain a preheating data packet and send it back to the gRPC client 11:

[0093] Step A1: extract the corresponding test scene identifier and vehicle identifier from the startup instruction and save them; use the simulation environment high-precision map and simulation world data volume of the test scene whose scene identifier matches the test scene identifier in the scene database 3 as the corresponding current high-precision map and current data volume; and load the current data volume;

[0094] Step A2, and set the simulation object set of the current data body {O i} as the corresponding current object set; and in the current object set, the simulation object O whose index i matches the vehicle identifier i Extract it as the corresponding vehicle object; and collect the current object into each object type d i Simulation object O that is not a signal light and whose index i does not match the vehicle ID iExtract it as the corresponding participant object; and concentrate the current object on each object type d i The simulation object O of the traffic light i Extract it as the corresponding traffic light object; and the object motion trajectory of the vehicle object in the current high-precision map The local high-precision map passed by is used as the corresponding local map of the vehicle; and the object motion trajectory is The last object motion state The coordinates of are used as the corresponding target position of the vehicle;

[0095] Step A3, and the object motion trajectories of the vehicle object, each participant object and each signal light object {s i,t}, the object or signal light motion state s where the time step t is greater than n i,t Delete; and set the object motion trajectory {s i,1≤t≤n} Delete the participants and signal light objects that have no intersection with the map area of ​​the local map of the vehicle;

[0096] Here, as we know from the previous text, n is a preset positive integer, n<N T ;

[0097] Step A4, and the object motion state s of each participant object at each time step t i,t and the corresponding object type d i A corresponding traffic participant data is formed; and the traffic participant data of all participant objects at each time step t form a corresponding traffic participant data set;

[0098] Step A5, and the object motion state s of the vehicle object at each time step t i,t Form a corresponding vehicle state data;

[0099] Step A6: Identify the number of signal light objects. If the number of signal light objects is zero, set a blank signal light data as the corresponding signal light data for time step t. If the number of signal light objects is one, the coordinates and signal light state of the unique signal light object at each time step t form a corresponding signal light data. If the number of signal light objects is greater than one, then at each time step t, the signal light object located in front of the travel path of the current vehicle object coordinates and closest to the current vehicle object coordinates is used as the corresponding current single-step signal light, and the coordinates and signal light state of the current single-step signal light at the current time step form a signal light data corresponding to the current time step.

[0100] In step A7, multiple types of map elements on the local map of the vehicle are extracted to form a corresponding map element set; the map element set is used as the map perception data corresponding to each time step t; the multiple types of map elements include at least lane lines, lane center lines, traffic signs / markings / markings, and road edge lines;

[0101] In step A8, the traffic participant data set and traffic light data corresponding to each time step t form a corresponding environmental perception data; the environmental perception data, vehicle status data and map perception data corresponding to each time step t form a corresponding next-step perception data; the obtained n next-step perception data are sorted in chronological order to form a corresponding next-step perception data sequence; and the vehicle target position and the next-step perception data sequence form a corresponding preheating data packet and are sent back to the gRPC client 11.

[0102] In another specific implementation of the embodiment of the present invention, the scene unit 21 is specifically configured to refresh the currently loaded simulation world data volume according to the data volume copy sent by the simulation unit 23 and reset the copy of the simulation unit 23 based on the refresh result:

[0103] Step B1, the simulation object set of the data body copy {O i} as the corresponding current object set; and set the current object set to the simulation object O whose index i matches the vehicle identifier i As the corresponding current vehicle object; and the current object is concentrated on each object type d i The simulation object O of the car i As the corresponding other vehicle objects; and the M object motion states s that are updated by the current vehicle object i,t Composed of the corresponding vehicle history trajectory; and the object motion trajectory of each other vehicle object {s i,t}, extract the sub-motion trajectories synchronized with the historical trajectory of the ego vehicle as the corresponding historical trajectories of other vehicles; record the historical trajectories of other vehicles on the same road section / traffic intersection as the historical trajectory of the ego vehicle as the first historical trajectory; and record the historical trajectories of other vehicles on different road sections / traffic intersections as the second historical trajectories;

[0104] Step B2, the built-in vehicle dynamics model predicts the object motion state s of the vehicle in the next M time steps based on the vehicle's historical trajectory. i,t Make predictions and use the predicted M object motion states s i,t Update the object motion trajectory of the current vehicle object i,t};

[0105] Step B3, the built-in vehicle dynamics model predicts the object motion state s of the vehicle in the next M time steps according to each second historical trajectory. i,t Make predictions and use the predicted M object motion states s i,t Update the corresponding object motion trajectory of other vehicle objects {s i,t};

[0106] Step B4, the built-in traffic participant behavior model analyzes the driving behavior type of other vehicle objects corresponding to each first historical trajectory in the next M time steps based on the historical trajectory of the vehicle and all the first historical trajectories to obtain the corresponding other vehicle behavior type; and the built-in vehicle dynamics model analyzes the object motion state s of the vehicle in the next M time steps based on each other vehicle behavior type and its corresponding first historical trajectory. i,t Make predictions and use the predicted M object motion states s i,t Update the corresponding object motion trajectory of other vehicle objects {s i,t};

[0107] Here, other vehicle behavior types include slowing down and going straight, accelerating and going straight, going straight at a constant speed, changing lanes to the left, changing lanes to the right, overtaking, turning around, avoiding to the left, avoiding to the right, sudden braking, and stopping;

[0108] Step B5 , the updated data body copy is loaded to become the latest simulation world data body; and the updated data body copy is sent back to the simulation unit 23 as the latest data body copy on the simulation unit 23 .

[0109] In another specific implementation of the embodiment of the present invention, the scene unit 21 is specifically configured to call the latest data volume copy from the simulation unit 23 to refresh the currently loaded simulation world data volume and generate corresponding simulation data based on the refresh result and store it in the simulation database 4:

[0110] Step C1, call the latest data body copy from the simulation unit 23 as the corresponding current copy; and set the simulation object set {O i} as the corresponding current object set; and set the current object set to the simulation object O whose index i matches the vehicle identifier i As the corresponding current vehicle object; and the current object is concentrated on each object type d i The simulation object O of the car i As the corresponding other vehicle objects; and the M object motion states s that are updated by the current vehicle object i,t Composition of the corresponding historical trajectory of the vehicle;

[0111] Step C2, and record the last time step t of the vehicle's historical trajectory as the corresponding current step t* ; and for the current single step t * Is it equal to the total number of steps N? T If yes, the current copy is used as the corresponding first copy; if not, the current object is set to each simulation object O i The object motion trajectory {s i,t}'s sub-motion trajectory Delete and use the deleted current copy as the corresponding first copy;

[0112] Step C3, and loading the first copy into the latest simulation world data body;

[0113] Step C4, and use the first copy as the corresponding post-test data body; and use the test scene identifier as the corresponding simulation scene identifier, and the vehicle identifier as the corresponding test vehicle identifier; and use the current time as the corresponding simulation timestamp; and the obtained simulation timestamp, simulation scene identifier, test vehicle identifier and post-test data body to form a corresponding simulation data and store it in the simulation database 4.

[0114] In another specific implementation of the embodiment of the present invention, the simulation unit 23 is specifically configured to: when performing a time-step vehicle motion state simulation based on the data volume copy and each received step instruction according to the Runge-Kutta method and refreshing the data volume copy based on the simulation result: i,t＝C As the corresponding starting motion state s C ; and extract the corresponding steering wheel angle, throttle opening and closing, brake opening and closing from the current step instruction as the corresponding starting steering wheel angle, starting throttle opening and closing, starting brake opening and closing; and according to the Runge-Kutta method according to the starting motion state s C , the initial steering wheel angle, the initial throttle opening and closing degree, and the initial brake opening and closing degree for the motion state s at the next time step t = C + 1 C+1 Make predictions; and based on the predicted motion state s C+1 The object motion state s of the simulated car object in the data volume copy i,t＝C+1 Reset; and add 1 to the single-step counter C.

[0115] In another specific implementation of the embodiment of the present invention, the simulation unit 23 is specifically configured to generate environmental perception data and vehicle status data based on the latest copy and send them to the gRPC server 22 and send the latest vehicle label route to the map unit 24:

[0116] Step D1, the simulation object set {O i} as the corresponding current object set; and set the current object set to the simulation object O whose index i matches the simulation vehicle identifieri As the corresponding current simulation vehicle object; and the current object is concentrated into object type d i A simulation object O that is not a signal light and is not the current simulation vehicle object i As the corresponding participant object; and set the current object to object type d i The simulation object O of the traffic light i As the corresponding signal light object; and based on the single-step counter C, the object motion state s of the current simulation vehicle object i,t＝C As a corresponding vehicle state data;

[0117] Step D2, and mark the coordinates of the vehicle state data as the current vehicle coordinates; and record the signal light object whose coordinates are located in front of the driving road of the current vehicle coordinates, closest to the current vehicle coordinates, and whose distance from the current vehicle coordinates is less than a preset first distance threshold as the current single-step signal light; and confirm whether the current single-step signal light does not exist, if so, set an empty signal light data as the corresponding signal light data, otherwise, set the signal light motion state s of the current single-step signal light i,t＝C As the corresponding signal light data;

[0118] Here, the first distance threshold is a preset threshold parameter;

[0119] Step D3, and the object motion state s of each participant object i,t＝C The coordinates of the participants are taken as the corresponding coordinates of the participants; and the coordinates of the participants whose distance from the current vehicle coordinates is less than the preset second distance threshold are marked as first coordinates; and the object type d corresponding to each first coordinate is i and the object motion state s i,t＝C Form a corresponding traffic participant data; and form a corresponding traffic participant data set from all the traffic participant data obtained;

[0120] Here, the second distance threshold is a preset threshold parameter;

[0121] Step D4: The traffic participant data set and the traffic light data obtained this time are combined into a corresponding environmental perception data; and the environmental perception data and the vehicle status data obtained this time are sent to the gRPC server 22;

[0122] Step D5, and the object motion trajectory of the current simulated vehicle object {s i,C-1≤t The coordinates of the vehicle are extracted and sorted in order to form a corresponding vehicle label route and sent to the map unit 24.

[0123] In another specific implementation of an embodiment of the present invention, the map unit 24 is specifically used to generate corresponding map perception data based on the global high-precision map and the self-vehicle label route sent by the simulation unit 23 and send it to the gRPC server 22: use the local high-precision map through which the self-vehicle label route passes in the global high-precision map as the corresponding current local map; and extract multiple types of map elements on the current local map to form a corresponding current element set; and send the current element set as the corresponding map perception data to the gRPC server 22; the multiple types of map elements include at least lane lines, lane center lines, traffic signs / signs / markings, and road edge lines.

[0124] (3) Scene database 3:

[0125] The scenario database 3 of the embodiment of the present invention is used to store multiple test scenarios.

[0126] Each test scenario includes a unique scenario identifier and a simulation world data volume of a traffic scenario. Here, the simulation world data volume of the embodiment of the present invention includes a high-precision map of the simulation environment and a simulation object set {O i}, 1≤index i≤N O , N O is the total number of objects; where:

[0127] 1) High-precision maps of the simulation environment include high-precision maps of all roads in the simulation environment;

[0128] 2) Simulation object set {O i}Includes multiple simulation objects O i ;Simulation object O i Include object type d i 、Object motion trajectory {s i,t}, 1≤ time step t≤N T , N T is the total number of steps, and the time interval between each two adjacent time steps t is fixed to interval L; object type d i At least include cars, motorcycles, bicycles, pedestrians, obstacles, and traffic lights; object type d i For cars, motorcycles, bicycles, pedestrians, obstacles, s i,t The object motion state of the current simulation object at time t, which includes at least coordinates, orientation angle, velocity and acceleration; object type d i When it is a signal light, s i,t is the traffic light motion state of the current simulation object at time t, the traffic light motion state at least includes coordinates and traffic light state, and the traffic light state at least includes red light, green light, and yellow light.

[0129] (IV) Simulation database 4:

[0130] The simulation database 4 of the embodiment of the present invention is used to store a plurality of simulation data; each simulation data is composed of a simulation timestamp, a simulation scene identifier, a test vehicle identifier and a post-test data body.

[0131] Here, each simulation data and post-test data body records the motion status of all participants and the ego vehicle during the entire simulation process; if the ego vehicle's trajectory in the test scene is used as the label trajectory and the ego vehicle's trajectory in the post-test data body is used as the test trajectory, the function and performance of the algorithm can be verified by comparing the ego vehicle's test-label trajectory; if the trajectories of other participants in the test scene are used as reference label trajectories, the safety of the algorithm can be verified by comparing the ego vehicle's test trajectory with the reference label trajectory; if the traffic lights in the test scene and the speed limit elements in the map are combined to check the ego vehicle's test trajectory, the compliance of the algorithm can be verified; by analyzing the speed, acceleration and other information at each point on the ego vehicle's test trajectory, the comfort of the algorithm can also be evaluated.

[0132] An embodiment of the present invention provides a traffic scenario simulation system. As can be seen from the above, the system includes: a vehicle manufacturer algorithm service, a scenario simulation service, a scenario database, and a simulation database. The algorithm and simulation sides communicate data via the gRPC protocol. The algorithm and simulation sides are implemented using Docker technology and deployed in separate pods. The publish-subscribe topic mechanism of the ROS architecture provides a data exchange channel between the internal components of the algorithm and simulation sides. After the algorithm sends a start command to the simulation side, it performs a preheating operation based on a preheating data packet sent by the simulation side. After the preheating is complete, it sends a step command to the simulation side and, based on the next step of the feedback, performs the prediction-planning-decision-control prediction process for the next time step. This cycle repeats until the algorithm sends a stop command to the simulation side, completing the entire testing process. The simulation side's scenario unit processes the start and end commands: 1) Upon receiving the start command, the scenario unit selects scenario data from the scenario database for loading, extracts the initial scenario segment, generates a preheating data packet, and sends it to the algorithm side. 2) Upon receiving the stop command, the scenario unit generates test scenario data based on the latest simulation scenario and stores it in the simulation database. On the simulation side, the simulation unit processes step commands. Each time a step command is received, the simulation unit simulates the vehicle's motion state for the next time step based on the control parameters in the command (such as steering wheel angle, throttle position, brake position), the real-time environmental state of the simulation scene (the motion state of other traffic participants, signal light status), and the real-time state of the ego vehicle. During the simulation, the Runge-Kutta method is used to process the prediction process to improve data accuracy. Based on the simulation results, the ego vehicle's state in the simulation scene is refreshed. The next step of perception data, composed of information such as the environmental state, the latest ego vehicle state, and high-precision maps, is fed back to the algorithm. Furthermore, the simulation unit updates the motion state of other traffic participants in the simulation scene after completing a specified number (M) of ego vehicle state refreshes to ensure the overall consistency of the simulated world. This embodiment of the present invention achieves the separation of algorithm and simulation by introducing Docker technology. Step commands ensure a tight coupling between the algorithm and the simulation environment. The linkage between the scene unit and the simulation unit improves the scene's realism, flexibility, and overall consistency. The present invention not only ensures the test efficiency, but also reduces the difficulty of development and deployment, and improves the flexibility, versatility and maintainability of the system.

[0133] Professionals should also be further aware that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0134] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0135] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A traffic scene simulation system, characterized in that: The system includes: a vehicle factory algorithm service, a scenario simulation service, a scenario database, and a simulation database; The depot algorithm service is connected to the scenario simulation service based on the gRPC protocol; the depot algorithm service and the scenario simulation service are implemented based on container technology, and the depot algorithm service and the scenario simulation service are respectively deployed in different container groups; The depot algorithm service is used to generate a corresponding start instruction according to the test task information input by the user and send it to the gRPC server of the scenario simulation service; and preheat the algorithm service based on the preheating data packet sent by the gRPC server and send the first step step instruction to the gRPC server at the end of the preheating; The vehicle manufacturer algorithm service is further configured to complete a single-step vehicle control prediction based on the next-step perception data sent by the gRPC server after each step instruction is sent, and obtain the next step instruction or end instruction and send it to the gRPC server; The scene simulation service includes a scene unit, the gRPC server, a simulation unit, a map unit, and a simulation ROS service; the scene unit is connected to the scene and the simulation database respectively; the simulation ROS service is used to provide a data interaction channel for the scene unit, the gRPC server, the simulation unit, and the map unit through the topic publish-subscribe mechanism of the ROS architecture; The gRPC server is used to forward the start instruction to the scene unit, and send the warm-up data packet sent back by it to the vehicle factory algorithm service; it is also used to forward the step instruction to the simulation unit, and send the next step perception data corresponding to the environmental perception data and vehicle status data sent back by the simulation unit and the map perception data sent by the map unit to the vehicle factory algorithm service; it is also used to forward the end instruction to the scene unit; The scene unit is used to load the simulation world data body and prepare preheating data according to the startup instruction and the scene database to obtain the preheating data packet and send it back to the gRPC client; and initialize the data body copy of the simulation unit and the global high-precision map of the map unit; The scene unit is further configured to refresh the currently loaded simulation world data volume according to the data volume copy sent by the simulation unit and reset the simulation unit copy based on the refresh result; The scene unit is further configured to, upon receiving the end instruction, call the latest copy of the data volume from the simulation unit to refresh the currently loaded simulation world data volume and generate corresponding simulation data based on the refresh result and store it in the simulation database; The simulation unit is configured to set a continuous counter initialized to 0 when the step instruction is received for the first time; perform a single-step time-step ego-vehicle motion state simulation based on the data body copy and the step instruction received each time according to the Runge-Kutta method, and refresh the data body copy based on the simulation result; add 1 to the continuous counter each time the copy refresh is completed; and identify whether the continuous counter after adding 1 is a preset number M; if not, use the current data body copy as the latest copy; if so, clear the continuous counter, send the current data body copy to the scene unit, and then use the simulated world data body sent back by the scene unit this time as the latest copy and save it; generate the environmental perception data and the ego-vehicle state data based on the latest copy, send them to the gRPC server, and send the latest ego-vehicle label route to the map unit; the preset number M is a positive integer; The map unit is used to generate the corresponding map perception data based on the global high-precision map and the vehicle label route sent by the simulation unit and send it to the gRPC server.

2. The traffic scene simulation system according to claim 1, characterized in that: The depot algorithm service includes a gRPC client, a prediction unit, a planning unit, a decision unit, a control unit, and an algorithm ROS service; the algorithm ROS service is used to provide a data interaction channel for the gRPC client, the prediction unit, the planning unit, the decision unit, and the control unit through the message topic publish-subscribe mechanism of the ROS architecture; The gRPC client is connected to the gRPC server of the scenario simulation service based on the gRPC protocol; The scenario database is used to store multiple test scenarios; Each of the test scenarios includes a unique scenario identifier and a simulation world data volume of a traffic scenario; the simulation world data volume includes a high-precision map of the simulation environment and a simulation object set {O i }, 1≤index i≤N O , N O The simulation environment high-precision map includes a high-precision map of all roads in the simulation environment; the simulation object set {O i }Includes multiple simulation objects O i ; The simulation object O i Include object type d i 、Object motion trajectory {s i,t }, 1≤ time step t≤N T , N T is the total number of steps, and the time interval between each two adjacent time steps t is fixed to interval L; the object type d i At least including cars, motorcycles, bicycles, pedestrians, obstacles, and traffic lights; the object type d i For cars, motorcycles, bicycles, pedestrians, obstacles, s i,t The object motion state of the current simulation object at time t, the object motion state at least includes coordinates, orientation angle, velocity and acceleration; the object type d i When it is a signal light, s i,t The motion state of the traffic light of the current simulation object at time t, wherein the motion state of the traffic light at least includes coordinates and the state of the traffic light, and the state of the traffic light at least includes red light, green light, and yellow light; The preheating data packet consists of the target position of the vehicle and the next step perception data sequence; the next step perception data sequence is composed of n next step perception data sorted in chronological order, n is a preset positive integer, n<N T The next-step perception data includes the environmental perception data, the vehicle status data and the map perception data; the environmental perception data includes a traffic participant data set and signal light data; the traffic participant data set is composed of a plurality of traffic participant data, and the traffic participant data includes object type, coordinates, heading angle, speed and acceleration; the signal light data includes coordinates and the signal light status; the vehicle status data includes coordinates, heading angle, speed and acceleration; the map perception data includes at least lane lines, lane center lines, traffic signs / markings / markings, and road edge lines in the four directions of the front, back, left and right of the road where the vehicle is traveling; The simulation database is used to store a plurality of simulation data; each simulation data consists of a simulation timestamp, a simulation scene identifier, a test vehicle identifier and a post-test data body.

3. The traffic scene simulation system according to claim 2, characterized in that: The vehicle factory algorithm service is specifically used to generate a corresponding start instruction based on the test task information input by the user and send it to the gRPC server of the scenario simulation service: The gRPC client receives the test task information input by the user, extracts the corresponding test scenario identifier and vehicle identifier from the test task information, and sends the test start instruction carrying the test scenario identifier and vehicle identifier to the gRPC server.

4. The traffic scene simulation system according to claim 2, characterized in that: The depot algorithm service is specifically used to preheat the algorithm service based on the preheating data packet sent by the gRPC server and send the first step instruction to the gRPC server at the end of the preheating: The gRPC client receives the preheating data packet sent by the gRPC server, and extracts the corresponding vehicle target position and the next-step perception data sequence from the preheating data packet; and sending the target position of the vehicle to the planning unit; and using each of the next-step perception data in the next-step perception data sequence as the corresponding current perception data; and extracting the corresponding traffic participant data set, the traffic light data, the vehicle status data, and the map perception data from the current perception data; And send the current traffic participant data set, the traffic light data and the map perception data to the prediction unit; and send the current vehicle status data and the map perception data to the planning unit; and wait for receiving the vehicle control instruction sent by the control unit; and after receiving the vehicle control instruction, continue to process the next next-step perception data as the new current perception data until the last next-step perception data is processed, and send the vehicle control instruction sent by the control unit for the last time as the step instruction of the first step to the gRPC server.

5. The traffic scene simulation system according to claim 2, characterized in that: The vehicle manufacturer algorithm service is specifically used to complete a single-step vehicle control prediction based on the next step perception data sent by the gRPC server after each step instruction is sent, and obtain the next step instruction or end instruction to send to the gRPC server: After sending each step instruction, the gRPC client waits for the next step perception data sent by the RPC server based on a preset first waiting time; and confirms whether the next step perception data can be received within the first waiting time; if it is confirmed that the next step perception data is not received within the first waiting time, the end instruction is sent to the gRPC server; if it is confirmed that the next step perception data is received within the first waiting time, the corresponding traffic participant data set, the signal light data, the vehicle status data and the map perception data are extracted from the current next step perception data, and the current traffic participant data set, the signal light data and the corresponding The map perception data is sent to the prediction unit, and the current vehicle state data and the map perception data are sent to the planning unit, and the vehicle control instruction sent by the control unit is waited for and received, and whether the vehicle control instruction can be received within the preset second waiting time is confirmed. If it is confirmed that the vehicle control instruction is not received within the second waiting time, the end instruction is sent to the gRPC server. If it is confirmed that the vehicle control instruction is received within the second waiting time, the current vehicle control instruction is sent to the gRPC server as the next step instruction; the instruction parameters of the vehicle control instruction include at least the steering wheel angle, the throttle opening and closing, and the brake opening and closing.

6. The traffic scene simulation system according to any one of claims 4 and 5, characterized in that: The prediction unit is configured to, upon receiving a set of the traffic participant data set, the traffic light data, and the map perception data, predict the movement trajectory of each traffic participant within a specified future time period based on the traffic participant data set, the traffic light data, and the map perception data received this time, to obtain a corresponding participant trajectory; and forming a corresponding participant trajectory set from all the obtained participant trajectories and sending it to the planning unit; the time length of the future specified period is greater than the interval L; The planning unit is configured to refresh the stored ego vehicle target position based on the current ego vehicle target position each time the ego vehicle target position is received from the gRPC client; The planning unit is further configured to refresh the stored vehicle state data and map perception data based on the current vehicle state data and map perception data each time a set of the vehicle state data and the map perception data sent by the gRPC client is received; The planning unit is further configured to, upon receiving each participant trajectory set sent by the prediction unit, plan a movement trajectory of the ego vehicle within the future specified time period based on the current participant trajectory set and the most recently stored ego vehicle target position, the ego vehicle state data, and the map perception data, obtain a corresponding ego vehicle trajectory, and send the plan to the decision unit; The decision unit is configured to determine the driving behavior type of the vehicle based on the current trajectory of the vehicle each time the vehicle trajectory is received from the planning unit to obtain the corresponding vehicle behavior type; and sending the vehicle behavior type and the vehicle trajectory to the control unit; the vehicle behavior type at least includes decelerating and going straight, accelerating and going straight, going straight at a constant speed, changing lanes to the left, changing lanes to the right, overtaking, turning around, avoiding to the left, avoiding to the right, sudden braking, and stopping; The control unit is configured to, upon receiving a set of the vehicle behavior type and the vehicle trajectory sent by the decision unit, predict the lateral and longitudinal driving control parameters of the vehicle at the next time step based on the current vehicle behavior type and the vehicle trajectory, obtain the corresponding vehicle control instruction, and send it to the gRPC client.

7. The traffic scene simulation system according to claim 2, characterized in that: The scene unit is specifically configured to, when the simulation world data body is loaded and preheating data is prepared according to the startup instruction and the scene database to obtain the preheating data packet and send it back to the gRPC client: Extracting and saving the corresponding test scene identifier and vehicle identifier from the startup instruction; and using the simulation environment high-precision map and the simulation world data volume of the test scene in the scene database that match the scene identifier with the test scene identifier as the corresponding current high-precision map and current data volume; and loading the current data volume; And the simulation object set {O i } as the corresponding current object set; and in the current object set, the simulation object O whose index i matches the self-vehicle identifier i Extract it as the corresponding vehicle object; and collect the object types d in the current object collection i The simulation object O that is not a traffic light and whose index i does not match the vehicle identifier i Extract it as the corresponding participant object; and collect the object types d in the current object i The simulation object O of the signal light i Extract it as the corresponding traffic light object; and extract the object motion trajectory of the vehicle object in the current high-precision map The local high-precision map passed by is used as the corresponding local map of the vehicle; and the object motion trajectory is The last motion state of the object The coordinates of as the corresponding target position of the vehicle; The object motion trajectories of the vehicle object, each of the participant objects, and each of the signal light objects are {s i,t }, the object or signal light motion state s where the time step t is greater than n i,t Delete; and the object motion trajectory {s i,1≤t≤n } Deleting the participant and traffic light objects that do not intersect with the map area of ​​the local map of the vehicle; The object motion state s of each participant object at each time step t is i,t and the corresponding object type d i forming a corresponding traffic participant data; The traffic participant data of all the participant objects at each time step t form a corresponding traffic participant data set; The object motion state s of the vehicle object at each time step t is i,t Forming a corresponding vehicle state data; and identifying the number of the signal light objects; if the number of the signal light objects is zero, setting an empty signal light data as the corresponding signal light data for the time step t; If the number of signal light objects is 1, the coordinates of the unique signal light object at each time step t and the signal light state constitute one corresponding signal light data; if the number of signal light objects is greater than 1, then at each time step t, the signal light object located in front of the travel path of the current vehicle object coordinates and closest to the current vehicle object coordinates is used as the corresponding current single-step signal light, and the coordinates of the current single-step signal light at the current time step and the signal light state constitute one corresponding signal light data for the current time step; Extracting multiple types of map elements from the local map of the vehicle to form a corresponding map element set; and using the map element set as the map perception data corresponding to each time step t; the multiple types of map elements at least include lane lines, lane center lines, traffic signs / markings / markings, and road edge lines; The traffic participant data set and the traffic light data corresponding to each time step t form a corresponding piece of environmental perception data; The environmental perception data, the vehicle status data, and the map perception data corresponding to each time step t form a corresponding next-step perception data; the obtained n next-step perception data are sorted in chronological order to form a corresponding next-step perception data sequence; and the vehicle target position and the next-step perception data sequence form a corresponding preheating data packet and send it back to the gRPC client.

8. The traffic scene simulation system according to claim 7, characterized in that: The scene unit is specifically used to send the vehicle identification and the currently loaded simulation world data body to the simulation unit when initializing the data body copy of the simulation unit and the global high-precision map of the map unit; and send the vehicle local map to the map unit.

9. The traffic scene simulation system according to claim 8, characterized in that: The simulation unit is further configured to save the received simulation world data volume as the latest local copy of the data volume upon receiving the vehicle identifier and the simulation world data volume sent by the scene unit; The received self-vehicle identification is saved as the latest local simulation vehicle identification; and the simulation object set based on the data body copy {O i The simulation object O whose index i matches the simulation vehicle identifier i As the corresponding simulation vehicle object; Initialize the single-step counter C to n; and use the motion trajectory of the sub-object of the simulation vehicle object {s i,1≤t≤n } is an observation quantity, which initializes the model parameters of the built-in vehicle dynamics model; and after the model parameter initialization is completed, waits for the reception of the first step instruction.

10. The traffic scene simulation system according to claim 8, characterized in that: The map unit is further configured to save the local map of the vehicle received this time as the latest local global high-precision map when receiving the local map of the vehicle sent by the scene unit.

11. The traffic scene simulation system according to claim 9, characterized in that: The simulation unit is specifically configured to perform a single-step time-step ego-vehicle motion state simulation according to the data volume copy and the stepping instruction received each time according to the Runge-Kutta method and refresh the data volume copy based on the simulation result: The object motion state s of the simulated vehicle object in the data volume copy is converted based on the single-step counter C. i,t＝C As the corresponding starting motion state s C ; The corresponding steering wheel angle, throttle opening and closing, and brake opening and closing are extracted from the current stepping instruction as the corresponding starting steering wheel angle, starting throttle opening and closing, and starting brake opening and closing; and the Runge-Kutta method is used according to the starting motion state s C , the initial steering wheel angle, the initial throttle opening and closing degree, and the initial brake opening and closing degree for the motion state s at the next time step t = C + 1 C+1 Make a prediction; and based on the predicted motion state s C+1 The object motion state s of the simulated vehicle object in the data volume copy i,t＝C+1 Reset; and add 1 to the single-step counter C.

12. The traffic scene simulation system according to claim 9, characterized in that: The simulation unit is specifically configured to, when generating the environmental perception data and the vehicle state data based on the latest copy and sending them to the gRPC server and sending the latest vehicle label route to the map unit: The simulation object set of the latest copy {O i } as the corresponding current object set; and the simulation object O whose index i matches the simulation vehicle identifier in the current object set i As the corresponding current simulation vehicle object; and the current object is concentrated in the object type d i The simulation object O that is not a signal light and is not the current simulation vehicle object i As the corresponding participant object; and the object type d of the current object i The simulation object O of the signal light i As the corresponding signal light object; and based on the single-step counter C, the object motion state s of the current simulation vehicle object i,t＝C As a corresponding vehicle state data; The coordinates of the vehicle state data are marked as the current vehicle coordinates; the signal light object whose coordinates are located in front of the driving road of the current vehicle coordinates, closest to the current vehicle coordinates, and whose distance from the current vehicle coordinates is less than a preset first distance threshold is recorded as the current single-step signal light; and whether the current single-step signal light does not exist is confirmed, if so, an empty signal light data is set as the corresponding signal light data, otherwise the signal light motion state s of the current single-step signal light is recorded. i,t＝C As the corresponding signal light data; and the object motion state s of each participant object i,t＝C The coordinates of the participants are taken as the corresponding coordinates of the participants; and the coordinates of the participants whose distance from the current vehicle coordinates is less than the preset second distance threshold are marked as first coordinates; and the object type d corresponding to each first coordinate is i and the object motion state s i,t＝C forming a corresponding traffic participant data; and forming the corresponding traffic participant data set from all the traffic participant data obtained; The traffic participant data set and the traffic light data obtained this time are combined into a corresponding piece of environmental perception data; And send the environmental perception data and the vehicle status data obtained this time to the gRPC server; And the object motion trajectory of the current simulated vehicle object {s i,C-1≤t The coordinates of the vehicle are extracted and sorted in order to form the corresponding vehicle label route and sent to the map unit.

13. The traffic scene simulation system according to claim 12, characterized in that: The map unit is specifically used to, when generating the corresponding map perception data based on the global high-precision map and the self-vehicle labeled route sent by the simulation unit and sending it to the gRPC server: use the local high-precision map through which the self-vehicle labeled route passes in the global high-precision map as the corresponding current local map; and extract multiple types of map elements on the current local map to form a corresponding current element set; and send the current element set as the corresponding map perception data to the gRPC server; the multiple types of map elements include at least lane lines, lane center lines, traffic signs / markings / markings, and road edge lines.

14. The traffic scene simulation system according to claim 2, characterized in that: The scene unit is specifically configured to, when refreshing the currently loaded simulation world data volume according to the data volume copy sent by the simulation unit and resetting the simulation unit copy based on the refresh result: The simulation object set of the data body copy {O i } as the corresponding current object set; and the simulation object O whose index i matches the vehicle identifier in the current object set i As the corresponding current vehicle object; and the current object is concentrated in each of the object types d i The simulation object O of the car i As the corresponding other vehicle objects; and the M object motion states s that are most recently updated by the current self-vehicle object i,t Composed of the corresponding vehicle history trajectory; and each of the object motion trajectories of the other vehicle objects {s i,t } extract the sub-motion trajectory synchronized with the historical trajectory of the own vehicle as the corresponding historical trajectory of other vehicles; and record the historical trajectory of other vehicles that are on the same road section / traffic intersection as the historical trajectory of the own vehicle as the first historical trajectory; and record the historical trajectory of other vehicles that are not on the same road section / traffic intersection as the historical trajectory of the own vehicle as the second historical trajectory; The built-in vehicle dynamics model predicts the object motion state s of the vehicle in the next M time steps based on the vehicle's historical trajectory. i,t Predict and use the predicted M motion states s of the object i,t Update the object motion trajectory of the current vehicle object i,t }; The built-in vehicle dynamics model predicts the object motion state s of the vehicle in the next M time steps according to each of the second historical trajectories. i,t Predict and use the predicted M motion states s of the object i,t Update the object motion trajectory of the corresponding other vehicle object {s i,t }; The built-in traffic participant behavior model analyzes the driving behavior type of the other vehicle objects corresponding to each of the first historical trajectories in the future M time steps based on the historical trajectory of the vehicle and all the first historical trajectories to obtain the corresponding other vehicle behavior type; and the built-in vehicle dynamics model analyzes the object motion state s of the vehicle in the future M time steps based on each of the other vehicle behavior types and their corresponding first historical trajectories. i,t Predict and use the predicted M motion states s of the object i,t Update the object motion trajectory of the corresponding other vehicle object {s i,t The other vehicle behavior types include slowing down and going straight, accelerating and going straight, going straight at a constant speed, changing lanes to the left, changing lanes to the right, overtaking, turning around, avoiding to the left, avoiding to the right, sudden braking, and stopping; The updated data body copy is loaded to become the latest simulation world data body; and the updated data body copy is sent back to the simulation unit as the latest data body copy on the simulation unit.

15. The traffic scene simulation system according to claim 2, characterized in that: The scene unit is specifically configured to, when the latest copy of the data volume is called from the simulation unit to refresh the currently loaded simulation world data volume and generate corresponding simulation data based on the refresh result and store it in the simulation database: Call the latest copy of the data body from the simulation unit as the corresponding current copy; and set the simulation object set of the current copy {O i } as the corresponding current object set; and the simulation object O whose index i matches the vehicle identifier in the current object set i As the corresponding current vehicle object; and the current object is concentrated in each of the object types d i The simulation object O of the car i As the corresponding other vehicle objects; and the M object motion states s that are most recently updated by the current self-vehicle object i,t Composition of the corresponding historical trajectory of the vehicle; The last time step t of the vehicle's historical trajectory is recorded as the corresponding current step t * ; and for the current single step t * Is it equal to the total number of steps N? T If yes, the current copy is used as the corresponding first copy; if not, the current object is set to each of the simulation objects O i The object motion trajectory {s i,t }'s sub-motion trajectory Deleting the current copy after deletion and using it as the corresponding first copy; and loading the first copy into the latest simulation world data body; and using the first copy as the corresponding post-test data body; and using the test scene identifier as the corresponding simulation scene identifier and the vehicle identifier as the corresponding test vehicle identifier; and taking the current time as the corresponding simulation timestamp; The obtained simulation timestamp, simulation scenario identifier, test vehicle identifier and post-test data body form a corresponding simulation data which is stored in the simulation database.

Images