Unmanned driving function test modeling method and terminal based on multi-dimensional decision-making

Abstract

The present invention discloses a method and terminal for modeling unmanned driving function test based on multidimensional decision making, and the steps are as follows: collecting road condition pictures of different types of intersections, determining unmanned driving functions to be tested, and inputting the functions to be tested into a table of functions to be tested; completing static context tree and dynamic context tree modeling respectively according to static scene elements and dynamic scene elements in the scene; generating a function output generation tree; generating a multidimensional decision table model for unmanned driving function test under a specific scene; combining the function output view in the multidimensional decision table model with the time axis to generate a corresponding three-dimensional spatiotemporal data graph under a specific scene. The present invention expands the decision table into four dimensions, takes into account the interaction between the unmanned vehicle and the surrounding environment, and realizes the time series test and action sequence test of the unmanned vehicle.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned driving test modeling, and specifically relates to an unmanned driving function test modeling method and terminal based on multi-dimensional decision-making. Background Art

[0002] With the booming development of information technology today, driverless cars have become a hot trend in the automotive industry. However, driverless cars are not a single technology. Their algorithms include three major components: sensing, perception, and decision-making. Driverless car decisions are based on the understanding of the vehicle's surroundings, and behavior plans can be generated in real time during the decision-making phase. The success of a strategy is the superposition of a series of actions over a long period of time. In recent years, the testing field has been to solve complex problems in a continuous environment, among which vehicle decision reliability testing is an essential part of the current industry implementation. However, existing methods cannot consider time series decisions when performing optimization, which makes such methods ineffective when dealing with long-term decisions.

[0003] Moreover, ordinary decision tables have difficulty in utilizing information such as the interaction between individuals and the environment and time, and lack the ability to adapt to the complex and continuous dynamic driving environment of unmanned vehicles, making the results of unmanned driving black box tests inaccurate. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide an unmanned driving function test modeling method and terminal based on multi-dimensional decision-making, so as to solve the problem that the unmanned driving black box testing method in the prior art does not take into account the multi-dimensional space-time continuous environment, resulting in inaccurate test results.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] The present invention provides a method for modeling an unmanned driving function test based on multi-dimensional decision making, and the steps are as follows:

[0007] 1) Collect road condition pictures of different types of intersections, determine the unmanned driving functions to be tested, and enter the functions to be tested into the function table to be tested;

[0008] 2) extracting main information from the traffic condition pictures of the different types of intersections, i.e., extracting static scene elements and dynamic scene elements, and completing static context tree and dynamic context tree modeling respectively according to the static scene elements and dynamic scene elements in the scene;

[0009] 3) Mark the test points in the test scene of the unmanned driving function to be tested, record the changes of dynamic scene elements around the test points over time and the output of the unmanned driving function, and generate a function output spanning tree;

[0010] 4) The static context view, dynamic context view, and function output view are spliced ​​in pairs to generate a multidimensional decision table model for unmanned driving function testing in a specific scenario; the function output view in the multidimensional decision table model is combined with the time axis to generate a corresponding three-dimensional spatiotemporal data graph under a specific scenario, recording the function realization of the unmanned vehicle on the spatiotemporal trajectory.

[0011] Furthermore, the step 1) specifically includes:

[0012] 11) Collect real traffic pictures of different types of roads and intersections on the road, and divide the time into three periods: morning peak, evening peak, and normal time; among them, the road types are divided into expressways, elevated roads, urban roads, and rural roads; the intersection types are divided into crossroads and T-shaped intersections;

[0013] 12) Locate the possible position of the vehicle. The unmanned driving functions to be tested are divided into direction change, speed change, and light control. The functions to be tested are entered into the function table to be tested.

[0014] Furthermore, the step 2) specifically includes:

[0015] 21) The main information extracted is static scene elements and dynamic scene elements, i.e., context; the static scene elements appearing in the real road condition picture are recorded, i.e., static context; the static scene elements are weather, number of lanes, lane conditions, and traffic signs; the static scene elements in the picture are recorded and stored in the static context table;

[0016] Process the pictures collected in step 1): use the target recognition method of the opencv library in python to extract the dynamic scene elements in the picture, that is, the dynamic context; the dynamic scene elements are NPC vehicles, pedestrians, traffic lights, non-motor vehicles, and roadblocks around the unmanned vehicle. Record the distance and action of the dynamic scene elements and the unmanned vehicle to be tested, and use the workbook.write() method in xlwt to store the recorded results in the dynamic context table;

[0017] 22) Static context and dynamic context are modeled using context tree. The context tree G CT =(N CT , E CT , R CT ), where N CT is a finite set of non-empty nodes; E CT is a set of unordered vertex pairs, called edges; R CT It is a set of relationships. The plt library of networkx is used to draw the tree. Add_nodes_from() is used to fill in static and dynamic context elements, and add_edges_from() is used to set the links between elements.

[0018] All static contexts are classified by using the xlwt library to read information from the static context table, and are divided into weather, number of lanes, and lane conditions. This classification method makes the static context an intermediate node, that is, a static context classification tree is formed; all dynamic contexts are classified by using the xlwt library to read information from the dynamic context table, and the contexts are divided into NPC vehicles, pedestrians, traffic lights, non-motor vehicles, and roadblocks. This classification method makes the dynamic context an intermediate node, that is, a dynamic context classification tree is formed;

[0019] For different functions to be tested in step 1), a combination of different static context classification tree leaf nodes is selected as a basic scenario, and different dynamic contexts are added to the basic scenario to form multiple complete unmanned driving test scenarios;

[0020] The spanning tree is a graph in which there is only one path between nodes. The DFS method is used to select the only leaf node in the last layer of the classification tree to generate a context spanning tree. By adding dynamic context to the basic scene with fixed static context, multiple static context and dynamic context spanning tree pairs can be combined, and the scene configuration is generated using the MFC method in the xlwt library and stored in the specific scene table. Among them, one static context spanning tree can correspond to multiple dynamic context spanning trees.

[0021] Furthermore, the step 3) specifically includes:

[0022] 31) Mark the test point of the unmanned driving function test scenario, that is, the expected output position of the function; define the test range, that is, a circular area with a radius of 20m centered on the expected output position of the function;

[0023] 32) Within the test range, record the changes of dynamic scene elements around the unmanned vehicle over time and update the dynamic context classification tree in real time;

[0024] 33) Recording the actual function output of the unmanned vehicle to be tested at the test point and storing it in the actual output column of the to-be-tested function table;

[0025] 34) The context tree modeling method in step 22) is used to generate a function output classification tree. All functions to be tested are classified, and the xlwt library is used to read information from the function table to be tested. The functions to be tested are divided into direction change, speed change, and light control, so that the functions to be tested become intermediate nodes, that is, a function output classification tree is formed; the DFS method is used to select the only leaf node in the last layer of the classification tree to generate a function output generation tree; the actual function output is compared with the function to be tested to obtain the unmanned driving function test results.

[0026] Furthermore, the step 4) specifically includes:

[0027] 41) Fill the corresponding nodes of the static context generation tree into the static context table, use excel2img() in the python script to take a screenshot of the table content, and generate a static context classification view;

[0028] 42) Combine the dynamic context generation tree with its corresponding time and store it in the dynamic context table corresponding to the fixed static context scene, use excel2img() in the python script to take a screenshot of the table content and generate a dynamic context classification view;

[0029] 43) The output spanning tree and its corresponding time combination are stored in the unmanned driving test function table corresponding to the fixed static context scene, and the table content is screenshotted using excel2img() in the python script to generate an output classification view;

[0030] 44) The obtained static context classification view, dynamic context classification view, output classification view and timestamp are combined, a div box is created using html, and the length, width and height are set before splicing the three table images to form a multidimensional decision table model;

[0031] 45) The functions, function output positions and timestamps in the classification view of the multidimensional decision table model output are plotted into a three-dimensional graph using the bar3d() method in the matplotlib library in Python; wherein the x and y axes represent the function output positions, and the z axis represents the function output time. Each cube formed represents the function output of the unmanned driving at that position, generating a three-dimensional spatiotemporal data graph of the unmanned vehicle in a specific scenario, recording the function realization of the unmanned vehicle in the spatiotemporal trajectory.

[0032] The present invention also provides a modeling terminal, comprising:

[0033] one or more processors;

[0034] A memory for storing one or more programs;

[0035] When the one or more programs are executed by the one or more processors, the one or more processors implement the unmanned driving function test modeling method based on multi-dimensional decision-making.

[0036] Beneficial effects of the present invention:

[0037] The present invention expands the traditional two-dimensional decision table into four dimensions, takes into account the interaction between the unmanned vehicle and the surrounding environment, realizes the time series test and action sequence test of the unmanned vehicle, and has the following advantages:

[0038] 1. Decision tables can be used to design complete test cases. The proposed multi-dimensional decision table model expands the traditional two-dimensional decision table into four dimensions. It not only refines the context of the environment around the autonomous driving vehicle, but also adds time to the decision table model to solve the problem that the existing search algorithm cannot use the details of the interaction between the individual and its operating environment at every moment, and realize sequential testing. Design scenarios for each function of the vehicle to achieve real-time testing to determine the functional correctness of the autonomous driving vehicle in different environments.

[0039] 2. The present invention uses HTML to display a three-dimensional model of a multidimensional decision table. The time axis runs through the diagonal of the original three-dimensional cube model. Any item in the first three dimensions can be mapped to the time axis to achieve overall visualization of the four-dimensional decision table model. After the decision table model is added with the time item, any dimension combined with the time dimension can form a planar space-time data graph. Visualizing the natural, social environment, and various parts of vehicle decision-making is more conducive to testers to understand the impact of different levels of autonomous driving space-time on the test from different angles. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the method of the present invention;

[0041] Figure 2 It is a static and dynamic context classification tree structure diagram in the present invention;

[0042] Figure 3 A tree example diagram for generating a static context when testing the cruise function in the present invention;

[0043] Figure 4 It is the static context view of the multidimensional decision table model in the present invention and the corresponding dynamic context and function output view;

[0044] Figure 5 It is the overall diagram of the multidimensional decision table model and the three-dimensional spatiotemporal data diagram in the present invention. DETAILED DESCRIPTION

[0045] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0046] Reference Figure 1 As shown, a method for modeling an unmanned driving function test based on multi-dimensional decision making of the present invention comprises the following steps:

[0047] 1) Collect road condition pictures of different types of intersections, determine the unmanned driving functions to be tested, and enter the functions to be tested into the function table to be tested; specifically including:

[0048] 11) Collect real traffic pictures of different types of roads and intersections on the road, and divide the time into three periods: morning peak, evening peak, and normal time; among them, the road types are divided into expressways, elevated roads, urban roads, and rural roads; the intersection types are divided into crossroads and T-shaped intersections;

[0049] 12) Locate the possible position of the vehicle. The unmanned driving functions to be tested are divided into direction change, speed change, and light control. The functions to be tested are entered into the function table to be tested.

[0050] 2) extracting main information from the road condition pictures of the different types of intersections, i.e., extracting static scene elements and dynamic scene elements, and completing static context tree and dynamic context tree modeling respectively according to the static scene elements and dynamic scene elements in the scene; the step 2) specifically includes:

[0051] 21) Static scene elements include weather, number of lanes, lane conditions, traffic signs, etc. The static environment context is represented by a static context classification tree, which represents the fixed environment elements of the unmanned driving scene. The classification tree structure refers to Figure 2 As shown. There are two types of relationships between classification tree nodes: SELECT-1 and SELECT-M. If the relationship between NP and NC in the classification tree is SELECT-1, NP must contain only one of its child nodes and the corresponding link in the spanning tree; if the relationship between NP and NC in the classification tree is SELECT-M, NP must contain its M child nodes and the corresponding links in the spanning tree.

[0052] Static context can be broken down into natural environment parameters Environment info, geographic information parameters RS info, lane information SV info, etc.; identify and classify different contexts and their parameters, and use static environment context classification trees to display classifications and static environment spatiotemporal data graphs to visualize static environments;

[0053] S=SELECT-M(Static Context,<Environment info,RS info,SV info,…> )

[0054] Instance 1-M =SELECT-1(Environment info,<(rainy,1),(cloudy,3),……>)

[0055] Sign 1-M =SELECT-1(Signs,<turn left,turn right,……> )

[0056] Instance a-z=SELECT-M(SV info,<(type 1 ,1),(type 2 ,2),……>)

[0057] G SCST1 =SELECT-1(info,<(Instance 1-M ,(x1,y1),Instance a-z ),(Instance 1-M ,(x2,y2),

[0058] Instance a-z ),……>)

[0059] Among them, Instance 1-M is an example of a natural environment parameter, represented by a tuple with weather and visibility (1-5); 1-M is a traffic sign instance; a-z is a lane situation instance, represented by the lane type and the lane number (1-5). The static context classification tree is composed of the above three instances. SELECT-1 is used to obtain the static context spanning tree, corresponding to Figure 4 The static context view shown.

[0060] 22) The dynamic scene elements are NPC vehicles, pedestrians, traffic lights (traffic lights, pedestrian lights, variable lane lights, etc.), non-motor vehicles, roadblocks, etc. around the unmanned vehicle. The location of these elements from the ego vehicle, the distance d, and the action action are recorded as dynamic context. Specifically, it is represented by a dynamic context classification tree, and the object is represented as More info; identify and classify different dynamic contexts and their parameters, and use the dynamic environment context classification tree to display the classification results.

[0061] D=SELECT-M(More info,<Traffic lights,pedestrians,NPCs,……> )

[0062] TL 1-M =SELECT-1(Trafic lights,<(t1,(x1,y1),d1,red),(t2,(x2,y2),d2,yellow),…>)

[0063] P 1-M =SELECT-M(Pedestrians,<(t1,(x1,y1),d1,action1),(t2,(x2,y2),d2,action2),…>)

[0064] NPC 1-M =SELECT-M(NPCs,<(t1,(x1,y1),d1,action1),(t2,(x2,y2),d2,action2),…>)

[0065] G DCST1 =SELECT-1(info,<(TL 1-M ,P 1-M ,NPC 1-M ),(TL 1-M ,P 1-M ,NPC 1-M ),……>)

[0066] Where t represents time, TL 1-M Represents a traffic light instance; P 1-M For pedestrian instances, NPC 1-M is an NPC vehicle instance. Both are dynamically mutable objects, represented by a four-tuple with action. The dynamic context classification tree is constructed by the above instances. SELECT-1 is used to obtain the dynamic context spanning tree. Figure 3 Shown is a static context spanning tree when the function under test is cruising.

[0067] 23) For the function to be tested, a combination of fixed static objects is set as a basic scene, and different dynamic scene elements are added to this basic scene to form multiple complete unmanned driving test scenes.

[0068] 24) Generate multiple static context and dynamic context spanning tree pairs corresponding to the test scenario at the same time. One static context can correspond to multiple dynamic context spanning trees, that is, Figure 4 One static context view corresponds to multiple dynamic contexts.

[0069] 3) Mark the test points in the test scene of the unmanned driving function to be tested, record the changes of dynamic scene elements around the test points over time, the output of the unmanned driving function, and generate a function output spanning tree; specifically including:

[0070] 31) Manually mark the test points of the unmanned driving function test scenario, that is, the expected output position of the function; define the test range, that is, a circular area with a radius of 20m centered on the expected output position of the function;

[0071] 32) Within the test range, record the changes of dynamic scene elements around the unmanned vehicle over time and update the dynamic context classification tree in real time;

[0072] 33) Record the output of the unmanned driving function at the test point and draw the output as an output spanning tree. The unmanned driving function output represents the function of the unmanned driving in each test scenario, including direction change Direction, speed change Speed, lighting control and other decisions;

[0073] Tasks = SELECT - M (Tasks,<D,S,W,LC> )

[0074] D 1-M =SELECT-1(Direction,<(t1,turn left),(t2,turn right),(t3,straight)……>)

[0075] S 1-M =SELECT-1(Speed,<(t1,+20km / s),(t2,-2km / s),(t3,0km / s)……>)

[0076] W 1-M =SELECT-1(Wiper,<(t1,<on,fast> ),(t2,<on,medium> )……>)

[0077] LC 1-M =SELECT-1(Lights Control,<(t1,fog light),(t2,outline markerlamp),(t3,dippedheadlight)……>)

[0078] G OCST1 =SELECT-M(Output,<(t1,D 1-M ,S 1-M ,W 1-M ,LC 1-M ),(t2,D 1-M ,S 1-M ,W 1-M ,LC 1-M ),……>)

[0079] Where t represents time, D 1-M Indicates the vehicle's driving direction; S 1-M is the current speed of the vehicle, which is the change of the vehicle speed compared with 2s ago; W 1-M is an example of the vehicle wiper operation, where fast, medium, and slow represent the wiper speeds; LC 1-Mis a vehicle light control instance, indicating the type of vehicle light on, which is off by default. The output classification tree is represented by a five-tuple with the above instance. SELECT-1 is used to obtain the dynamic context spanning tree, corresponding to Figure 4 Function output context view shown;

[0080] 34) The context tree modeling method in step 22) is used to generate a function output classification tree. All functions to be tested are classified, and the xlwt library is used to read information from the function table to be tested. The functions to be tested are divided into direction change, speed change, and light control, so that the functions to be tested become intermediate nodes, that is, a function output classification tree is formed; the DFS method is used to select the only leaf node in the last layer of the classification tree to generate a function output generation tree; the actual function output is compared with the function to be tested to obtain the unmanned driving function test results.

[0081] 4) Combine the static context view, dynamic context view, and function output view in pairs to generate a multidimensional decision table model for autonomous driving function testing in a specific scenario; combine the function output view with the time axis in the multidimensional decision table model to generate a corresponding three-dimensional spatiotemporal data graph in a specific scenario, and record the function realization on the spatiotemporal trajectory of the autonomous vehicle; specifically include:

[0082] 41) Fill the corresponding nodes of the static context generation tree into the static context table, use excel2img() in the python script to take a screenshot of the table content, and generate a static context classification view;

[0083] 42) Combine the dynamic context generation tree with its corresponding time and store it in the dynamic context table corresponding to the fixed static context scene, use excel2img() in the python script to take a screenshot of the table content and generate a dynamic context classification view;

[0084] 43) The output spanning tree and its corresponding time combination are stored in the unmanned driving test function table corresponding to the fixed static context scene, and the table content is screenshotted using excel2img() in the python script to generate an output classification view;

[0085] 44) The obtained static context classification view, dynamic context classification view, output classification view and timestamp are combined, and a div box is created using html. After setting the length, width and height, the three table images are spliced ​​to form a multidimensional decision table model, such as Figure 5 ;

[0086] 45) The functions, function output positions and timestamps in the output classification view of the multidimensional decision table model are plotted into a three-dimensional graph using the bar3d() method in the matplotlib library in Python; where the x and y axes represent the function output positions, and the z axis represents the function output time. Each cube formed represents the function output of the unmanned vehicle at that position, generating a three-dimensional spatiotemporal data graph of the unmanned vehicle in a specific scenario. Figure 5 , recording the space-time trajectory of unmanned vehicles.

[0087] The present invention has many specific application paths. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principle of the present invention. These improvements should also be regarded as the protection scope of the present invention.

Claims

1. A method for testing unmanned driving functions based on multi-dimensional decision making, characterized in that: Here are the steps: 1) Collect road condition pictures of different types of intersections, determine the unmanned driving functions to be tested, and enter the functions to be tested into the function table to be tested; 2) extracting main information from the traffic condition pictures of the different types of intersections, i.e., extracting static scene elements and dynamic scene elements, and completing static context tree and dynamic context tree modeling respectively according to the static scene elements and dynamic scene elements in the scene; 3) Mark the test points in the test scene of the unmanned driving function to be tested, record the changes of dynamic scene elements around the test points over time and the output of the unmanned driving function, and generate a function output spanning tree; 4) The static context classification view, the dynamic context classification view, and the output classification view are spliced ​​in pairs to generate a multidimensional decision table model for unmanned driving function testing in a specific scenario; Combine the function output view and time axis in the multidimensional decision table model to generate a corresponding three-dimensional spatiotemporal data graph under a specific scenario, and record the function realization of the unmanned vehicle on the spatiotemporal trajectory; The step 2) specifically includes: 21) The main information extracted is static scene elements and dynamic scene elements, i.e., context; the static scene elements appearing in the real road condition picture are recorded, i.e., static context; the static scene elements are weather, number of lanes, lane conditions, and traffic signs; the static scene elements in the picture are recorded and stored in the static context table; Process the pictures collected in step 1): use the target recognition method of the opencv library in python to extract the dynamic scene elements in the picture, that is, the dynamic context; the dynamic scene elements are NPC vehicles, pedestrians, traffic lights, non-motor vehicles, and roadblocks around the unmanned vehicle. Record the distance and action of the dynamic scene elements and the unmanned vehicle to be tested, and use the workbook.write() method in xlwt to store the recorded results in the dynamic context table; 22) Static context and dynamic context are modeled using context tree. The context tree G CT =(N CT , E CT , R CT ), where N CT is a finite set of non-empty nodes; E CT is an unordered set of vertex pairs, called edges; R CT It is a set of relationships. The plt library of networkx is used to draw the tree. Add_nodes_from() is used to fill in static and dynamic context elements, and add_edges_from() is used to set the links between elements. All static contexts are classified by using the xlwt library to read information from the static context table, and are divided into weather, number of lanes, and lane conditions. This classification method makes the static context an intermediate node, that is, a static context classification tree is formed; all dynamic contexts are classified by using the xlwt library to read information from the dynamic context table, and the contexts are divided into NPC vehicles, pedestrians, traffic lights, non-motor vehicles, and roadblocks. This classification method makes the dynamic context an intermediate node, that is, a dynamic context classification tree is formed; For different functions to be tested in step 1), a combination of different static context classification tree leaf nodes is selected as a basic scenario, and different dynamic contexts are added to the basic scenario to form multiple complete unmanned driving test scenarios; The spanning tree is a graph in which there is only one path between nodes. The DFS method is used to select the only leaf node in the last layer of the classification tree to generate a context spanning tree. By adding dynamic context to the basic scene with fixed static context, multiple static context and dynamic context spanning tree pairs can be combined, and the scene configuration is generated by the MFC method in the xlwt library and stored in the specific scene table. Among them, one static context spanning tree can correspond to multiple dynamic context spanning trees. The step 3) specifically includes: 31) Mark the test point of the unmanned driving function test scenario, that is, the expected output position of the function; define the test range, that is, a circular area with a radius of 20m centered on the expected output position of the function; 32) Within the test range, record the changes of dynamic scene elements around the unmanned vehicle over time and update the dynamic context classification tree in real time; 33) Recording the actual function output of the unmanned vehicle to be tested at the test point and storing it in the actual output column of the to-be-tested function table; 34) Using the context tree modeling method in step 22), a function output classification tree is generated. All functions to be tested are classified, and the xlwt library is used to read information from the function table to be tested. The functions to be tested are divided into direction change, speed change, and light control, so that the functions to be tested become intermediate nodes, that is, a function output classification tree is formed; the DFS method is used to select the only leaf node in the last layer of the classification tree and then generate a function output spanning tree; the actual function output is compared with the function to be tested to obtain the unmanned driving function test result; The step 4) specifically includes: 41) Fill the corresponding nodes of the static context generation tree into the static context table, use excel2img() in the python script to take a screenshot of the table content, and generate a static context classification view; 42) Combine the dynamic context generation tree with its corresponding time and store it in the dynamic context table corresponding to the fixed static context scene, use excel2img() in the python script to take a screenshot of the table content and generate a dynamic context classification view; 43) The output spanning tree and its corresponding time combination are stored in the unmanned driving test function table corresponding to the fixed static context scene, and the table content is screenshotted using excel2img() in the python script to generate an output classification view; 44) Combine the obtained static context classification view, dynamic context classification view, output classification view and timestamp, use HTML to create a div box, set the length, width and height, and then splice the three table images to form a multidimensional decision table model; 45) The functions, function output positions and timestamps in the classification view of the multidimensional decision table model output are plotted into a three-dimensional graph using the bar3d() method in the matplotlib library in Python; wherein the x and y axes represent the function output positions, and the z axis represents the function output time. Each cube formed represents the function output of the unmanned driving at that position, generating a three-dimensional spatiotemporal data graph of the unmanned vehicle in a specific scenario, recording the function realization of the unmanned vehicle in the spatiotemporal trajectory.

2. The unmanned driving function test modeling method based on multi-dimensional decision-making according to claim 1 is characterized in that: The step 1) specifically includes: 11) Collect real traffic pictures of different types of roads and intersections on the road, and divide the time into three periods: morning peak, evening peak, and normal time; among them, the road types are divided into expressways, elevated roads, urban roads, and rural roads; the intersection types are divided into crossroads and T-shaped intersections; 12) Locate the possible position of the vehicle. The unmanned driving functions to be tested are divided into direction change, speed change, and light control. The functions to be tested are entered into the function table to be tested.

3. A modeling terminal, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 2.

Images