Intelligent driving decision-making method, vehicle driving control method and device, and vehicle

Abstract

This application relates to an intelligent driving technology, and specifically provides an intelligent driving judgment method, including: first obtaining a game object of a host vehicle; then performing multiple releases of the multiple strategy spaces from multiple strategy spaces of the host vehicle and the game object; determining a strategy feasible region of the host vehicle and the game object based on each released strategy space; and determining a driving judgment result of the host vehicle based on the strategy feasible region. The judgment result is a feasible behavior action of the host vehicle. As described above, by releasing the strategy space multiple times, the judgment result may be obtained when as few strategy spaces as possible are released while ensuring the judgment accuracy. This reduces the amount of computing and reduces the requirements for the computing power of the hardware.

Description

[Technical Field]

[0001] This application relates to intelligent driving technology, and in particular to an intelligent driving decision method, a vehicle driving control method and device, and a vehicle. [Background technology]

[0002] With the development of artificial intelligence technology, autonomous driving technology is gradually becoming more widespread, thereby reducing the burden on drivers. For example, the Society of Automotive Engineers International (SAE International) provides five levels of autonomous driving, from L1 to L5. L1 indicates driver assistance, assisting the driver in completing several driving tasks and completing a single driving operation. L2 indicates partial automation, which automatically performs acceleration, deceleration, and steering simultaneously. L3 indicates conditional automation. In certain environments, a vehicle may automatically accelerate, decelerate, and steer without the driver's input. L4 indicates high automation. A vehicle may be driven without a driver, but there are limitations. For example, the vehicle speed cannot exceed a certain value, and the driving area is relatively fixed. L5 indicates fully automated and fully adaptive driving, which is applicable to all driving scenarios. The higher the level, the more advanced the autonomous driving function.

[0003] Currently, automated driving technologies that exceed the L2 level and require proper driving by a human driver are generally considered intelligent driving. When a vehicle performs intelligent driving, it must be able to timely and accurately recognize surrounding obstacles, such as oncoming vehicles, crossing vehicles, stopped vehicles, and pedestrians, determine the vehicle's driving behavior and driving trajectory, and perform actions such as accelerating, steering, and lane changes. Summary of the Invention

[0004] This application provides an intelligent driving judgment method, a vehicle driving control method and device, a vehicle, etc., which can minimize the computing power consumption required to implement vehicle driving judgment while ensuring the accuracy of the judgment.

[0005] A first aspect of this application provides an intelligent driving judgment method, the method including: obtaining a game object of a host vehicle; performing a plurality of releases of a plurality of strategic spaces from a plurality of strategic spaces of the host vehicle and the game object; after performing one of the plurality of releases, determining a strategy feasible region of the host vehicle and the game object based on each released strategic space; and determining a driving judgment result of the host vehicle based on the strategy feasible region.

[0006] The strategy feasible region of the ego-vehicle and non-game objects includes executable behavioral actions of the ego-vehicle with respect to the non-game objects. As described above, by releasing multiple strategy spaces multiple times, the strategy feasible region is obtained when releasing as few strategy spaces as possible while ensuring judgment accuracy (judgment accuracy may be, for example, the probability of execution of the judgment result). In this way, behavior-action pairs are selected from the strategy feasible region as the judgment result, thereby minimizing the number of operations and times to release the strategy space and reducing the computing power requirements of the hardware.

[0007] In a possible implementation of the first aspect, the dimensions of the plurality of strategy spaces include at least one of a vertical sampling dimension, a horizontal sampling dimension, and a temporal sampling dimension.

[0008] As described above, multiple strategy spaces are expanded based on the vertical sampling dimension, horizontal sampling dimension, or temporal sampling dimension. The multiple strategy spaces include a vertical sampling strategy space expanded in the vertical sampling dimension of the ego-vehicle and / or game object, a horizontal sampling strategy space expanded in the horizontal sampling dimension of the ego-vehicle and / or game object, a temporal dimension strategy space expanded in the temporal sampling dimension of the ego-vehicle and / or game object, or a strategy space formed by any combination of two or three of the vertical sampling dimension, horizontal sampling dimension, or temporal sampling dimension. The temporal dimension strategy space corresponds to the strategy space expanded in each of the multiple single-frame derivations included in one step of judgment. In each single-frame derivation, this expanded strategy space may include a vertical sampling strategy space and / or a horizontal sampling strategy space.

[0009] As mentioned above, the corresponding strategy space may be expanded in at least one sampling dimension based on the traffic scenario, and the strategy space may be freed up.

[0010] In a possible implementation of the first aspect, performing multiple releases of multiple strategy spaces includes performing the releases in the order of vertical sampling dimension, horizontal sampling dimension, and temporal sampling dimension.

[0011] As described above, the strategy space that is released multiple times in the order of releasing the longitudinal sampling dimension, releasing the horizontal sampling dimension, and releasing the temporal sampling dimension is the following strategy space: a longitudinal sampling strategy space that is spread over a group of values ​​of the longitudinal sampling dimension of the ego-vehicle; a longitudinal sampling strategy space that is spread over another group of values ​​of the longitudinal sampling dimension of the ego-vehicle; a longitudinal sampling strategy space that is spread over both a group of values ​​of the longitudinal sampling dimension of the ego-vehicle and a group of values ​​of the longitudinal sampling dimension of the game object ... horizontal sampling strategy space that is spread over a group of values ​​of the horizontal sampling dimension of the ego-vehicle and a vertical sampling strategy space that is spread over the longitudinal sampling dimension of the ego-vehicle and / or the vertical sampling dimension of the game object. a strategy space that spans both the horizontal sampling strategy space that spans another group of values ​​of the horizontal sampling dimension of the ego vehicle and the vertical sampling strategy space that spans the vertical sampling dimension of the ego vehicle and / or the vertical sampling dimension of the game object; a strategy space that spans both the horizontal sampling strategy space that spans the group of values ​​of the horizontal sampling dimension of the ego vehicle and the group of values ​​of the horizontal sampling dimension of the game object and the vertical sampling strategy space that spans the vertical sampling dimension of the ego vehicle and / or the vertical sampling strategy space of the game object; a strategy space that spans both the horizontal sampling strategy space that spans both the group of values ​​of the horizontal sampling dimension of the ego vehicle and the group of values ​​of the horizontal sampling dimension of the game object and the vertical sampling strategy space that spans the vertical sampling dimension of the ego vehicle and / or the vertical sampling dimension of the game object; and a horizontal sampling strategy space that spans both the group of values ​​of the horizontal sampling dimension of the ego vehicle and the group of values ​​of the horizontal sampling dimension of the game object and the vertical sampling strategy space that spans the vertical sampling dimension of the ego vehicle and / or the vertical sampling dimension of the game object;The strategy space includes a strategy space that spans both the longitudinal sampling dimension of the ego-vehicle and / or the longitudinal sampling dimension of the game object. Additionally, after the strategy feasible regions of the ego-vehicle and the game object are determined based on each of the released strategy spaces and the driving decision result of the ego-vehicle is determined based on the strategy feasible regions, the released temporal dimension strategy space includes strategy spaces that span each of multiple single frame derivations included in one step decision. In each single frame derivation, the strategy space that spans may include a vertical sampling strategy space, a horizontal sampling strategy space, and a strategy space that spans both the vertical sampling strategy space and the horizontal sampling strategy space.

[0012] In this way, multiple strategy spaces are sequentially released, i.e., firstly changing the vehicle acceleration in the longitudinal direction, and then adjusting the vehicle offset in the lateral direction, thereby better matching the vehicle's driving habits and better satisfying the driving safety requirements. Finally, a decision result with better temporal consistency may be further determined from the multiple feasible regions based on the derivation of multiple frames released in the temporal dimension.

[0013] In a possible implementation of the first aspect, when the strategy feasible area of ​​the ego-vehicle and game object is determined, a total cost value of the behavior action pair in the strategy feasible area is determined based on one or more of the ego-vehicle or game object's safety cost value, right of way cost value, lateral offset cost value, passability cost value, comfort cost value, inter-frame association cost value, and risk area cost value.

[0014] As described above, one or more cost values ​​may be selected as needed to calculate a total cost value, which is used to determine the feasible region.

[0015] In a possible implementation, when the total cost value of a behavior-action pair is determined based on two or more cost values, each of the cost values ​​has a different weight.

[0016] As described above, different weight values ​​may focus on driving safety, right-of-way, accessibility, comfort, risk, etc. Each weight value may be flexibly set to enhance the flexibility of intelligent driving decisions. In some possible implementations, the weight values ​​may be assigned in the following order: safety weight > right-of-way weight > lateral offset weight > accessibility weight > comfort weight > risk zone weight > inter-frame association weight.

[0017] In a possible implementation of the first aspect, when there are two or more game objects, the driving decision result of the host vehicle is determined based on the strategy feasibility areas of the host vehicle and each game object.

[0018] As mentioned above, when there are multiple game objects, the strategy feasibility region for each game object is obtained separately, and the final strategy feasibility region is determined based on the intersection of the strategy feasibility regions. In this specification, the intersection refers to behaviors that include the same actions of the host vehicle.

[0019] In a possible implementation of the first aspect, the method further includes obtaining non-game objects of the host vehicle; determining a strategy feasible area of ​​the host vehicle and the non-game objects, wherein the strategy feasible area of ​​the host vehicle and the non-game objects includes executable behavioral actions of the host vehicle relative to the non-game objects; and determining a driving decision result of the host vehicle based at least on the strategy feasible area of ​​the host vehicle and the non-game objects.

[0020] As mentioned above, when non-game objects are present, the final judgment result is related to the non-game objects.

[0021] In a possible implementation of the first aspect, the strategy feasible area of ​​the driving decision result of the host vehicle is determined based on the intersection of the strategy feasible areas of the host vehicle and each game object, or the strategy feasible area of ​​the driving decision result of the host vehicle is determined based on the intersection of the strategy feasible areas of the host vehicle and each game object and the strategy feasible areas of the host vehicle and each non-game object.

[0022] As described above, when multiple game objects and the host vehicle exist, the driving decision result of the host vehicle and the final strategy feasible area may be obtained based on the intersection of the strategy feasible areas of the host vehicle and the multiple game objects.When multiple game objects, multiple non-game objects, and the host vehicle exist, the driving decision result of the host vehicle and the final strategy feasible area may be obtained based on the intersection of the strategy feasible areas of the host vehicle, the multiple game objects, and the multiple non-game objects.

[0023] In a possible implementation of the first aspect, the method further includes acquiring a non-game object of the ego-vehicle, and constraining a longitudinal sampling strategy space corresponding to the ego-vehicle or constraining a lateral sampling strategy space corresponding to the ego-vehicle based on the motion status of the non-game object.

[0024] As described above, the longitudinal sampling strategy space corresponding to the ego vehicle is constrained, i.e., the range of values ​​of the longitudinal sampling dimension of the ego vehicle that are used when the longitudinal sampling strategy space is expanded is constrained. The longitudinal sampling strategy space corresponding to the ego vehicle is constrained, i.e., the range of values ​​of the lateral sampling dimension of the ego vehicle that are used when the lateral sampling strategy space is expanded is constrained.

[0025] As described above, the range of values ​​for the longitudinal acceleration or lateral offset of the vehicle in the expanded strategy space may be restricted based on the motion status of non-game objects such as position, velocity, etc. This reduces the number of behavioral actions in the strategy space and further reduces the amount of movement.

[0026] In a possible implementation of the first aspect, the method further includes obtaining a non-game object of the game object of the ego-vehicle, and constraining a longitudinal sampling strategy space corresponding to the game object of the ego-vehicle or constraining a lateral sampling strategy space corresponding to the game object of the ego-vehicle based on the motion status of the non-game object.

[0027] As mentioned above, the vertical sampling strategy space corresponding to the ego vehicle's game object is constrained, i.e., the range of values ​​of the vertical sampling dimension corresponding to the ego vehicle's game object that are used when the vertical sampling strategy space is expanded is constrained. The horizontal sampling strategy space corresponding to the ego vehicle's game object is constrained, i.e., the range of values ​​of the horizontal sampling dimension of the ego vehicle's game object that are used when the horizontal sampling strategy space is expanded is constrained.

[0028] As described above, the range of values ​​for the longitudinal acceleration or lateral offset of the game object of the vehicle in the expanded strategy space may be restricted based on the motion status of non-game objects such as position, velocity, etc. This reduces the number of behavioral actions in the strategy space and further reduces the amount of movement.

[0029] In a possible implementation of the first aspect, when the intersection is an empty set, a conservative driving decision of the host vehicle is made, the conservative decision including an action to safely stop the host vehicle or an action to safely slow down the host vehicle for driving.

[0030] As described above, if the strategy feasible region of the host vehicle is null, the host vehicle may travel safely.

[0031] In a possible implementation of the first aspect, the game object or non-game object is determined by attention.

[0032] As mentioned above, game and non-game objects may be determined based on the attention allocated to the ego-vehicle by each obstacle, which may be implemented algorithmically or via neural network inference.

[0033] In a possible implementation of the first aspect, the method further includes displaying, via a human-computer dialogue interface, at least one of the driving decision result of the vehicle, a strategy feasible area of ​​the decision result, a driving trajectory of the vehicle corresponding to the driving decision result of the vehicle, or a driving trajectory of a game object corresponding to the driving decision result of the vehicle.

[0034] In this way, the driving judgment results corresponding to the vehicle or the game may be displayed with rich content on the human-computer dialogue interface, making dialogue with the user more friendly.

[0035] A second aspect of the present application provides an intelligent driving decision-making device, including: an acquisition module configured to acquire a game object of a host vehicle; and a processing module configured to perform a plurality of releases of a plurality of strategic spaces from a plurality of strategic spaces of the host vehicle and the game object, and after performing one of the plurality of releases, determine a strategy feasible region of the host vehicle and the game object based on each released strategic space, and determine a driving decision result of the host vehicle based on the strategy feasible region.

[0036] In a possible implementation of the second aspect, the dimensions of the plurality of strategy spaces include at least one of a vertical sampling dimension, a horizontal sampling dimension, and a temporal sampling dimension.

[0037] In a possible implementation of the second aspect, performing multiple releases of multiple strategy spaces includes performing the releases in the order of vertical sampling dimension, horizontal sampling dimension, and temporal sampling dimension.

[0038] In a possible implementation of the second aspect, when the strategy feasible area of ​​the ego-vehicle and game object is determined, a total cost value of the behavior-action pair in the strategy feasible area is determined based on one or more of the ego-vehicle or game object's safety cost value, right-of-way cost value, lateral offset cost value, passability cost value, comfort cost value, inter-frame association cost value, and risk area cost value.

[0039] In a possible implementation, when the total cost value of a behavior-action pair is determined based on two or more cost values, each of the cost values ​​has a different weight.

[0040] In a possible implementation of the second aspect, when there are two or more game objects, the driving decision result of the host vehicle is determined based on the strategy feasibility areas of the host vehicle and each game object.

[0041] In a possible implementation of the second aspect, the acquisition module is further configured to acquire non-game objects of the ego-vehicle. The processing module is further configured to determine a strategy feasible area of ​​the ego-vehicle and the non-game objects, the strategy feasible area of ​​the ego-vehicle and the non-game objects including executable behavior actions of the ego-vehicle with respect to the non-game objects, and determine a driving decision result of the ego-vehicle based at least on the strategy feasible area of ​​the ego-vehicle and the non-game objects.

[0042] In a possible implementation of the second aspect, the processing module is further configured to determine a strategy feasible area of ​​the driving decision result of the host vehicle based on the intersection of the strategy feasible areas of the host vehicle and each game object, or to determine a strategy feasible area of ​​the driving decision result of the host vehicle based on the intersection of the strategy feasible area of ​​the host vehicle and each game object and the strategy feasible area of ​​the host vehicle and each non-game object.

[0043] In a possible implementation of the second aspect, the acquisition module is further configured to acquire non-game objects of the ego-vehicle, and the processing module is further configured to constrain a longitudinal sampling strategy space corresponding to the vehicle or constrain a lateral sampling strategy space corresponding to the ego-vehicle based on a motion situation of the non-game objects.

[0044] In a possible implementation of the second aspect, the acquisition module is further configured to acquire non-game objects of the ego-vehicle game object, and the processing module is further configured to constrain a longitudinal sampling strategy space corresponding to the ego-vehicle game object or constrain a lateral sampling strategy space corresponding to the ego-vehicle game object based on the motion status of the non-game objects.

[0045] In a possible implementation of the second aspect, when the intersection is an empty set, a conservative driving decision of the host vehicle is made, the conservative decision including an action to safely stop the host vehicle or an action to safely slow down the host vehicle for driving.

[0046] In a possible implementation of the second aspect, the game object or non-game object is determined by attention.

[0047] In a possible implementation of the second aspect, the processing module is further configured to display, via the human-computer dialogue interface, at least one of the driving decision result of the vehicle, a strategy feasible area of ​​the decision result, a driving trajectory of the vehicle corresponding to the driving decision result of the vehicle, or a driving trajectory of a game object corresponding to the driving decision result of the vehicle.

[0048] A third aspect of this application provides a vehicle driving control method including acquiring an obstacle outside the vehicle, determining a driving judgment result of the vehicle in relation to the obstacle according to any of the methods in the first aspect, and controlling the driving of the vehicle based on the judgment result.

[0049] A fourth aspect of the present application provides a vehicle driving control device including an acquisition module configured to acquire an obstacle outside the vehicle, and a processing module configured to determine a driving judgment result of the vehicle in relation to the obstacle according to any of the methods in the first aspect, the processing module being further configured to control the driving of the vehicle based on the judgment result.

[0050] A fifth aspect of the present application provides a vehicle including the vehicle driving control device according to the fourth aspect and a driving system. The vehicle driving control device controls the driving system.

[0051] A sixth aspect of the present application provides a computing device including a processor and a memory, wherein the memory stores program instructions that, when executed by the processor, enable the processor to implement any of the intelligent driving decision-making methods of the first aspect, or the vehicle cruise control method of the third aspect.

[0052] A seventh aspect of the present application provides a computer-readable storage medium. The computer-readable storage medium stores program instructions that, when executed by a processor, enable the processor to implement any of the intelligent driving decision-making methods of the first aspect, or the vehicle cruise control method of the third aspect.

[0053] These and other aspects of the present application will become clearer and easier to understand in the description of the embodiment(s) that follows. [Brief explanation of the drawings]

[0054] The features and relationships between features of this application will be further described below with reference to the accompanying drawings. The accompanying drawings are all examples, and some features are not drawn to scale. Additionally, in some of the accompanying drawings, common features that are not essential to the field of this application may be omitted. Alternatively, additional functions that are not essential to the application are shown. The combination of features shown in the accompanying drawings is not intended to limit this application. Additionally, in this specification, the same reference numerals represent the same content. Specifically, the accompanying drawings are described below.

[0055] [Figure 1] 1 is a schematic diagram of a traffic scenario in which a vehicle is traveling on a road surface according to an embodiment of the present application;

[0056] [Figure 2] 1 is a schematic diagram of a vehicle to which an embodiment of the present application is applied.

[0057] [Figure 3A] 1A-1C are schematic diagrams of game objects and non-game objects in different traffic scenarios according to an embodiment of the present application. [Figure 3B] 1A-1C are schematic diagrams of game objects and non-game objects in different traffic scenarios according to an embodiment of the present application. [Figure 3C] 1A-1C are schematic diagrams of game objects and non-game objects in different traffic scenarios according to an embodiment of the present application. [Figure 3D] 1A-1C are schematic diagrams of game objects and non-game objects in different traffic scenarios according to an embodiment of the present application. [Figure 3E] 1A-1C are schematic diagrams of game objects and non-game objects in different traffic scenarios according to an embodiment of the present application.

[0058] [Figure 4] 2 is a flowchart of an intelligent driving decision method according to an embodiment of the present application.

[0059] [Figure 5] Figure 4 is a flowchart for obtaining game objects.

[0060] [Figure 6] 5 is a flowchart for obtaining the determination result of FIG. 4.

[0061] [Figure 7] FIG. 1 is a schematic diagram of multi-frame derivation according to an embodiment of the present application.

[0062] [Figure 8A] FIG. 2 is a schematic diagram of a cost function according to an embodiment of the present application; [Figure 8B] FIG. 2 is a schematic diagram of a cost function according to an embodiment of the present application; [Figure 8C] FIG. 2 is a schematic diagram of a cost function according to an embodiment of the present application; [Figure 8D] FIG. 2 is a schematic diagram of a cost function according to an embodiment of the present application; [Figure 8E] FIG. 2 is a schematic diagram of a cost function according to an embodiment of the present application; [Figure 8F] FIG. 2 is a schematic diagram of a cost function according to an embodiment of the present application;

[0063] [Figure 9] 4 is a flowchart of cruise control according to another embodiment of the present application.

[0064] [Figure 10] 1 is a schematic diagram of a projection scenario according to one implementation of the present application.

[0065] [Figure 11] 1 is a flowchart of cruise control according to a particular implementation of the present application.

[0066] [Figure 12] 1 is a schematic diagram of an intelligent driving decision device according to an embodiment of the present application;

[0067] [Figure 13] 1 is a flowchart of a vehicle driving control method according to an embodiment of the present application.

[0068] [Figure 14] 1 is a schematic diagram of a vehicle cruise control device according to one embodiment of the present application.

[0069] [Figure 15] 1 is a schematic diagram of a vehicle according to one embodiment of the present application.

[0070] [Figure 16] 1 is a schematic diagram of a computing device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE INVENTION

[0071] The technical solutions provided in this application are described below with reference to the accompanying drawings and embodiments. It should be understood that the system architectures and service scenarios in the embodiments of this application are mainly intended to describe possible implementations of the technical solutions of this application and should not be construed as unique limitations on the technical solutions of this application. Those skilled in the art may know that the technical solutions provided in the embodiments of this application can also be applied to similar technical challenges as system structures evolve and new service scenarios emerge.

[0072] It should be understood that the intelligent driving decision-making solutions provided in the embodiments of this application include intelligent driving decision-making methods and devices, vehicle driving control methods and devices, vehicles, electronic devices, computing devices, computer-readable storage media, and computer program products. Because the problem-solving principles of the technical solutions are the same or similar, in the following description of specific embodiments, some repeated parts may not be described herein, but it should be considered that the specific embodiments can be mutually referenced and combined with each other.

[0073] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by those skilled in the art to which this application pertains. In the event of any discrepancy, the meanings set forth in this specification or acquired by the recorded content of this application shall prevail. Additionally, the terms used in this specification are intended to describe the purpose of the embodiments of this application, but are not intended to limit this application.

[0074] FIG. 1 illustrates a traffic scenario in which vehicles travel on a road surface. As illustrated in FIG. 1, in the traffic scenario, a north-south road and an east-west road form intersection A. A first vehicle 901 is located on the south side of intersection A and travels from south to north. A second vehicle 902 is located on the north side of intersection A and travels from north to south. A third vehicle 903 is located on the east side of intersection A and travels from east to south. That is, the third vehicle 903 turns left at intersection A and enters the north-south road. A fourth vehicle 904 is located behind the first vehicle 901 and is also traveling from south to north. A fifth vehicle 905 is parked on the north-south road side near the southeast corner of intersection A. That is, the fifth vehicle 905 is located on the roadside ahead of the first vehicle 901. It is assumed that the intelligent driving function is enabled for the first vehicle 901. In this case, the first vehicle 901 may detect the current traffic scenario, determine a driving strategy for the current traffic scenario, and control the vehicle's driving based on the determination result. For example, the vehicle may be controlled to accelerate, decelerate, or change lanes based on the determined cut-in, yield, and avoidance policies. An intelligent driving decision solution involves making decisions regarding driving strategies based on a game system. For example, the first vehicle 901 determines its own driving strategy in a traffic scenario in which an oncoming second vehicle 902 is driving using a game system. In complex traffic scenarios, it is difficult to determine a driving strategy using a game system. For example, in the traffic scenario shown in FIG. 1, the first vehicle 901 faces the oncoming second vehicle 902, a third vehicle 903 crossing intersection A on one side, and a fifth vehicle 905 parked on the side of the road ahead of the first vehicle 901. When the driving strategy is determined using a game system, the game objects of the first vehicle 901 are both the second vehicle 902 and the third vehicle 903. Therefore, game decision requires the use of a multidimensional game space, for example, a multidimensional game space spanning the lateral and longitudinal driving dimensions of each of the first vehicle 901, the second vehicle 902, and the third vehicle 903. The use of a multidimensional game space leads to an explosive increase in the number of solutions to game decision, resulting in a geometric increase in computing load and posing significant challenges to existing hardware computing capabilities.Therefore, at present, due to the limited computing power of the hardware, it is difficult to implement the commercialization of intelligent driving scenarios using multi-dimensional game spaces for game judgment.

[0075] One embodiment of this application provides an improved intelligent driving decision-making solution. When this solution is applied to intelligent vehicle driving, the basic principle of the solution includes identifying obstacles in a current traffic scenario relative to the ego-vehicle. The obstacles may include game objects of the ego-vehicle and non-game objects of the ego-vehicle. For a single game object of the ego-vehicle, a strategy space spanning a single sampling dimension or multiple sampling dimensions is released multiple times from the multidimensional game space of the ego-vehicle and the single game object. Additionally, each time the strategy space is released, a solution for the ego-vehicle and the single game object of the ego-vehicle in the strategy space is searched for. When a solution exists, i.e., when there is a game result, i.e., when the ego-vehicle and the single game object of the ego-vehicle have a strategy feasible region in the strategy space, a driving decision result for the ego-vehicle may be determined based on the game result, and the driving of the ego-vehicle may be controlled based on the decision result. In this case, the unreleased strategy space may no longer be released from the multidimensional game space. As described above, for multiple game objects of the ego-vehicle, a strategy feasible region for the ego-vehicle for each game object may be determined separately. Additionally, the vehicle's behavior is used as an indicator to obtain the vehicle's driving strategy from the intersection between the vehicle's strategy feasibility area and each game object (the intersection means that the same behavior of the vehicle is included). This method can obtain the optimal driving decision with the minimum number of searches in the multidimensional game space while ensuring decision accuracy (decision accuracy may be, for example, the execution probability of the decision result), and can minimize the use of strategy space as much as possible. This reduces the demand for hardware computing power, making it easier to commercialize on vehicles.

[0076] The subject for implementing the intelligent driving decision-making solution in this embodiment of the present application may be a powered, autonomously mobile intelligent agent. The intelligent agent may make game decisions with other objects in a traffic scenario based on the intelligent driving decision-making solution provided in this embodiment of the present invention, generate semantic-level decision labels and a predicted driving path for the intelligent agent, and then perform suitable lateral and longitudinal motion planning. For example, the intelligent device may be a vehicle with autonomous driving capabilities, an autonomously mobile robot, etc. In this specification, the term "vehicle" includes general motor vehicles, such as land transportation devices including sedans, sport utility vehicles (SUVs), sport utility vehicles (MPVs), automated guided vehicles (AGVs), buses, trucks, and other cargo or passenger vehicles, water transportation devices including various ships and boats, and aircraft. Motor vehicles also include hybrid vehicles, electric vehicles, fuel-powered vehicles, plug-in hybrid electric vehicles, fuel cell vehicles, and other alternative-fuel vehicles. A hybrid vehicle is a vehicle with two or more power sources, and electric vehicles include battery electric vehicles, long-range electric vehicles, etc. An autonomously mobile robot may also be one of the vehicles.

[0077] An example in which the intelligent driving decision-making solution provided in this embodiment of the present application is provided to a vehicle will be described below. As shown in Fig. 2, when the intelligent driving decision-making solution is applied to a vehicle, a vehicle 10 may include an environmental information acquisition device 11, a control device 12, and a driving system 13, and may further include a communication device 14, a navigation device 15, or a display device 16 in some embodiments.

[0078] In this embodiment, the environmental information acquisition device 11 may be configured to acquire external environmental information of the vehicle. The environmental information acquisition device 11 may include one or more of a camera, a laser radar, a millimeter-wave radar, an ultrasonic radar, a Global Navigation Satellite System (GNSS), and the like. The camera may include a conventional RGB (Red, Green, Blue) camera sensor, an infrared camera sensor, and the like. The acquired external environmental information of the vehicle includes road surface information and objects on the road surface. The objects on the road surface include surrounding vehicles, pedestrians, and the like, and may specifically include vehicle motion status information. The motion status information may include vehicle speed, acceleration, steering angle information, trajectory information, and the like. In some embodiments, the motion status information of surrounding vehicles may be acquired via the communication device 14 of the vehicle 10. The external environmental information of the vehicle acquired by the environmental information acquisition device 11 may be used to form a world model including roads (corresponding to road surface information), obstacles (corresponding to objects on the road surface), and the like.

[0079] In some embodiments, the environmental information acquisition device 11 may be an electronic device, such as a data transmission chip, that receives external environmental information of the vehicle transmitted from a camera sensor, an infrared night vision camera sensor, a laser radar, a millimeter-wave radar, an ultrasonic radar, etc. The data transmission chip may be, for example, a bus data transceiver chip, a network interface chip, etc. Alternatively, the data transmission chip may be, for example, a wireless transmission chip, such as a Bluetooth chip or a Wi-Fi chip. In some other embodiments, the environmental information acquisition device 11 may alternatively be integrated into the control device 12 and function as an interface circuit, a data transmission module, etc. integrated into a processor.

[0080] In this embodiment, the control device 12 may be configured to make a decision regarding an intelligent driving strategy based on the acquired external environment information of the vehicle (including the constructed world model) and generate a decision result. For example, the decision result may include acceleration, braking, steering (including lane changes or steering), or a predicted driving trajectory of the vehicle in a short period of time (e.g., within a few seconds). In some embodiments, the control device 12 may generate a corresponding command based on the decision result to control the driving system 13, execute driving control of the vehicle via the driving system 13, and control the vehicle to drive along the predicted driving trajectory based on the decision result. In the embodiments of this application, the control device 12 may be an electronic device, such as a processor of an on-board processing device such as a head unit, a domain controller, a mobile data center (MDC), or an on-board computer, or may be a conventional chip such as a central processing unit (CPU) or a microprocessor (MCU).

[0081] In this embodiment, the propulsion system 13 may include a power system 131, a steering system 132, and a braking system 133, which are described below.

[0082] The power system 131 may include a drive electrical control unit (ECU) and a drive source. The drive ECU controls the drive source to control the drive force (e.g., torque) of the vehicle 10. For example, the drive source may be an engine, a drive motor, or the like. The drive ECU may control the drive source based on the operation of an accelerator pedal performed by the driver, or may control the drive source in accordance with a command sent from the control device 12 to control the drive force. The drive force of this drive source is transmitted to wheels via a transmission or the like to drive the vehicle 10.

[0083] The steering system 132 may include a steering electric control unit (ECU) and an electric power steering (EPS). The steering ECU may control the EPS motor based on the steering wheel operation performed by the driver or may control the EPS motor according to a command sent from the control device 12 to control the direction of the wheels (specifically, the steering wheels). Additionally, the steering ECU may change the torque distribution and braking force distribution between the left and right wheels to perform the steering operation.

[0084] The power system 133 may include a brake electronic control unit (ECU) and a braking mechanism. The braking mechanism operates the braking unit via a brake motor, a hydraulic mechanism, or the like. The driving ECU may control the driving source based on the operation of the brake pedal performed by the driver, or may control the driving source according to a command sent from the control device 12 to control the driving force. When the vehicle 10 is an electric vehicle or a hybrid vehicle, the braking system 133 may further include an energy recovery braking mechanism.

[0085] In this embodiment, a communication device 14 may also be included. The communication device 14 may exchange data with external objects via wireless communication to obtain data necessary for the vehicle 10 to make intelligent driving decisions. In some embodiments, the communicable external objects may include a cloud server, a mobile terminal (e.g., a mobile phone, a portable computer, a tablet), a roadside device, surrounding vehicles, etc. In some embodiments, the data necessary for the decision may include user profiles of vehicles around the vehicle 10 (i.e., other vehicles). The user profiles reflect the driving habits of the drivers of the other vehicles and may further include the location of the other vehicles, movement status information of the other vehicles, etc.

[0086] In this embodiment, a communication device 15 may also be included. The navigation device 15 may include a Global Navigation Satellite System (GNSS) receiver and a map database. The navigation device 15 may determine the location of the vehicle 10 using satellite signals received by the GNSS receiver, generate a route to the destination based on map information in the map database, and provide information about the route (including the location of the vehicle 10) to the control device 12. Furthermore, the navigation device 15 may also include an Inertial Measurement Unit (IMU) and perform more accurate positioning of the vehicle 10 based on both information from the GNSS receiver and information from the IMU.

[0087] In this embodiment, a communication device 16 may also be included. For example, the display device 16 may be a display screen installed at a central control position in the vehicle cockpit, or may be a head-up display (HUD) device. In some embodiments, the control device 12 may display the determination result on the display device 16 in the vehicle cockpit in a manner that is understandable to a user, for example, in the form of a predicted driving trajectory, arrows, text, etc. In some embodiments, when the predicted driving trajectory is displayed, the display device in the vehicle cockpit may further display the predicted driving trajectory in the form of a partial enlarged view, with reference to the vehicle's current traffic scenario (e.g., a graphical traffic scenario). The control device 12 may also display information about the route to the destination provided by the navigation device 15.

[0088] In some embodiments, an audio playback system may also be included, where the user is prompted with the decisions made in the current traffic scenario.

[0089] The intelligent driving decision-making method provided in one embodiment of this application is described below. For convenience of description, in this embodiment of this application, the intelligent driving vehicle in the traffic scenario and implementing the intelligent driving decision-making method provided in this embodiment of this application is referred to as the ego-vehicle. From the ego-vehicle's perspective, other objects in the traffic scenario that affect or may affect the driving of the ego-vehicle are referred to as the ego-vehicle's obstacles.

[0090] In this embodiment of the present application, the host vehicle has a specific behavior determination capability and may generate a driving strategy to change the motion situation of the host vehicle. The driving strategy includes acceleration, braking, and steering (including lane change or steering). The host vehicle further has a driving behavior execution capability, including executing the driving strategy and driving based on the determined predicted driving trajectory.

[0091] In some embodiments, an obstacle to the host vehicle may also have a behavior determination capability and change the obstacle movement situation. For example, the obstacle may be an autonomously moving vehicle, a pedestrian, etc. Alternatively, the obstacle to the host vehicle may not have a behavior determination capability or change the obstacle movement situation. For example, the obstacle may be a vehicle parked on the side of the road (a vehicle that has not started moving), a pier with a limited road width, etc. In other words, obstacles to the host vehicle include pedestrians, bicycles, automobiles (motorcycles, passenger cars, cargo trucks, large trucks, buses, etc.), etc. The automobile may include an intelligent driving vehicle capable of executing an intelligent driving decision-making method.

[0092] Based on whether a game interaction relationship with the ego-vehicle is established, obstacles for the ego-vehicle may be further classified as game objects for the ego-vehicle, non-game objects for the ego-vehicle, or irrelevant obstacles for the ego-vehicle. Specifically, the strength of interaction between the ego-vehicle and game objects, non-game objects, and irrelevant obstacles gradually weakens from strong interaction to no interaction. It should be understood that game objects, non-game objects, and irrelevant obstacles may change from one another in multiple traffic scenarios corresponding to different decision moments. Depending on the position or movement status of the irrelevant obstacle for the ego-vehicle, the irrelevant obstacle becomes completely irrelevant to the future behavior of the ego-vehicle. There will be no future trajectory or intention conflict between the ego-vehicle and the irrelevant obstacle. Therefore, unless otherwise specified, obstacles in the embodiments of this application are game objects for the ego-vehicle and non-game objects for the ego-vehicle.

[0093] The non-game objects of the host vehicle will have future trajectory or intention conflicts with the host vehicle. Therefore, while the non-game objects of the host vehicle impose constraints on the future behavior of the host vehicle, the non-game objects of the host vehicle do not respond to future trajectory or intention conflicts between the non-game objects of the host vehicle and the host vehicle. Instead, the host vehicle must unilaterally adjust its own motion status to resolve future trajectory or intention conflicts between the non-game objects and the host vehicle. In other words, no game interaction relationship is established between the non-game objects of the host vehicle and the host vehicle. In other words, the non-game objects of the host vehicle are not affected by the driving behavior of the host vehicle and remain in their predetermined motion status. The non-game objects do not adjust their motion status to resolve future trajectory or intention conflicts between the non-game objects and the host vehicle.

[0094] A game interaction relationship is established between the ego-vehicle and the ego-vehicle's game object. The ego-vehicle's game object responds to potential future trajectory or intention conflicts between the ego-vehicle's game object and the ego-vehicle. At the moment when game judgment is initiated, a trajectory or intention conflict exists between the ego-vehicle's game object and the ego-vehicle. In the game process, the ego-vehicle and the ego-vehicle's game object each adjust their motion situations to gradually resolve potential trajectory or intention conflicts between them while ensuring safety. When the vehicle used as the ego-vehicle's game object adjusts its motion situation, automatic adjustments may be performed using the vehicle's intelligent driving function, or manual driving adjustments may be performed by the vehicle's driver.

[0095] To further understand game objects and non-game objects, the game objects and non-game objects of the ego-vehicle will now be described by way of example with reference to the schematic diagrams of several traffic scenarios in Figures 3A to 3E.

[0096] As shown in Figure 3A, the host vehicle 101 goes straight through an unprotected intersection. The oncoming vehicle 102 (to the left of the host vehicle 101) turns left and crosses the unprotected intersection. In this case, there is a trajectory conflict or intention conflict between the oncoming vehicle 102 and the host vehicle 101, and the oncoming vehicle 102 is a game object of the host vehicle 101.

[0097] As shown in Figure 3B, the host vehicle 101 travels straight ahead. The approaching vehicle 102 on the left crosses the lane of the host vehicle 101 and passes through. In this case, there is a trajectory conflict or an intention conflict between the approaching vehicle 102 on the left and the host vehicle 101, and the approaching vehicle 102 on the left is a game object of the host vehicle 101.

[0098] As shown in Figure 3C, the host vehicle 101 travels straight ahead. A vehicle 102 approaching in the same direction (to the right of the host vehicle 101) enters the host vehicle's lane or an adjacent lane. In this case, a trajectory conflict or an intention conflict exists between the host vehicle 101 and the vehicle 102 approaching in the same direction, a game dialogue relationship is established between the vehicle 102 and the host vehicle 101, and the vehicle 102 approaching in the same direction is a game object of the host vehicle 101.

[0099] As shown in FIG. 3D , the host vehicle 101 is traveling straight, an oncoming vehicle 103 is traveling straight in the adjacent lane to the left of the host vehicle 101, and a stationary vehicle 102 (to the right of the host vehicle 101) is present in the adjacent lane to the right of the host vehicle 101. In this case, a trajectory conflict or intention conflict exists between the oncoming vehicle 103 and the host vehicle 101, a game dialogue relationship is established between the oncoming vehicle 103 and the host vehicle 101, and the oncoming vehicle 103 is a game object of the host vehicle 101. The position of the stationary vehicle 102 will conflict with the trajectory of the host vehicle 101 in the future. However, based on the acquired external environment information, the interactive game judgment process may determine that the stationary vehicle 102 will not transition to a moving state, or even transition to a moving state, but has a higher right-of-way and will not establish a game dialogue relationship with the host vehicle 101. Therefore, the stationary vehicle 102 becomes a non-game object of the host vehicle. To resolve this trajectory conflict between the two, the running behavior and movement conditions of the host vehicle 101 are adjusted separately.

[0100] As shown in FIG. 3E, the host vehicle 101 changes lanes from the current lane to the right and enters the adjacent lane on the right. The adjacent lane on the right contains a first straight-moving vehicle 103 (to the right front of the host vehicle 101) and a second straight-moving vehicle 102 (to the right rear of the host vehicle 101). The first straight-moving vehicle 103 has a higher right of way than the host vehicle 101, does not establish a game interaction relationship with the host vehicle 101, and is a non-game object of the host vehicle 101. The second straight-moving vehicle 102 (to the right rear of the host vehicle 101) will have a trajectory conflict with the host vehicle 101 in the future, and establishes a game interaction relationship with the host vehicle 101. The second straight-moving vehicle 102 is a game object of the host vehicle 101.

[0101] Please refer to Figure 1 and Figure 2, and refer to the flowchart shown in Figure 4. Hereinafter, an intelligent driving decision-making method provided in one embodiment of this application will be described. The method includes the following steps:

[0102] S10: The host vehicle acquires the game object of the host vehicle.

[0103] In some embodiments, in the flowchart shown in FIG. 5, this step may include the following substeps:

[0104] S11: The host vehicle acquires external environment information of the vehicle, and the acquired external environment information includes the movement status and relative position information between the host vehicle and obstacles in the road scenario.

[0105] In this embodiment, the vehicle may acquire external environmental information about the vehicle through an environmental information acquisition device of the vehicle, such as a camera sensor, an infrared night vision camera sensor, a laser radar, a millimeter-wave radar, an ultrasonic radar, or a GNSS. In some embodiments, the vehicle may acquire external environmental information about the vehicle by communicating with a roadside device or a cloud server via a communication device of the vehicle. The roadside device may have a camera or a communication device and may acquire information about vehicles around the roadside device. The cloud server may receive and store information reported from each roadside device. In some embodiments, external environmental information about the vehicle may be acquired using a combination of the two methods described above.

[0106] S12: The host vehicle identifies the game object of the host vehicle from the obstacle based on the acquired motion status of the obstacle, or the motion status of the obstacle over a certain period of time, or the formed travel trajectory of the obstacle, and the relative position information of the obstacle.

[0107] In this embodiment, in step S12, non-game objects of the host vehicle may be identified from the obstacles, or non-game objects of the host vehicle's game objects may be identified from the obstacles.

[0108] In some embodiments, game objects or non-game objects may be identified according to preset decision rules. According to these decision rules, for example, if an obstacle's driving trajectory or driving intention conflicts with the driving trajectory or driving intention of the host vehicle, and the obstacle has behavior-determining capabilities and can change its motion state, the obstacle is a game object of the host vehicle. If an obstacle's driving trajectory or driving intention conflicts with the driving trajectory or driving intention of the host vehicle, and the obstacle's motion state does not actively change to actively avoid the conflict, the obstacle is a non-game object of the host vehicle. In some embodiments, the obstacle's driving trajectory or driving intention may be determined based on the lane in which the obstacle is traveling (straight lane or turning lane), whether the turn signal is on, the vehicle's head direction, etc.

[0109] For example, in Figures 3A-3C, both the crossing obstacle and the lane-entering obstacle share a large intersection angle with the trajectory of the host vehicle, and therefore have a driving trajectory conflict and are further classified as game vehicles. The oncoming vehicle 103 passing through the narrow lane in Figure 3D and the rear vehicle 104 in the adjacent lane into which the host vehicle is entering in Figure 3E have an intention conflict and are further classified as game vehicles. The stationary vehicle 102 on the right front passing through the narrow lane in Figure 3D and the front vehicle 103 in the adjacent lane into which the host vehicle is entering in Figure 3E have a trajectory conflict or intention conflict. However, because the host vehicle has a lower right of way than the other vehicle, the other vehicle does not take any action to resolve the conflict. Additionally, the behavior of the host vehicle cannot change the behavior of the other vehicle. In this case, the other vehicle is a non-game vehicle.

[0110] In some embodiments, the ego-vehicle obtains obstacle information from the vehicle's external environment information perceived or acquired by the ego-vehicle according to some known algorithm, and identifies from the obstacles game objects of the ego-vehicle, non-game objects, or non-game objects of game objects.

[0111] In some embodiments, the algorithm may be, for example, a classification neural network based on deep learning. Because identifying the type of obstacle is equivalent to classification, the judgment result may be determined after inference is performed using the classification neural network. The classification neural network may use a convolutional neural network (CNN), a recurrent neural network (RNN), a bidirectional encoder representations from transformers (BERT), or the like. When the classification neural network is trained, sample data may be used to train the neural network. The sample data may be images or video clips of vehicle driving scenarios labeled with classification labels. The classification labels may include game objects, non-game objects, and non-game objects of game objects.

[0112] In some embodiments, the aforementioned algorithms may also use attention-related algorithms, such as a modeled attention model. The attention model is used to output an attention value assigned by each obstacle to the ego-vehicle. The attention value relates to the degree of intent or trajectory conflict between the obstacle and the ego-vehicle. For example, obstacles that have intent or trajectory conflict with the ego-vehicle may allocate more attention to the ego-vehicle. Obstacles that do not have intent or trajectory conflict with the ego-vehicle may allocate less attention, or no attention, to the ego-vehicle. Obstacles that have a higher right of way than the ego-vehicle may also allocate less attention, or no attention, to the ego-vehicle. If an obstacle allocates sufficient attention (e.g., higher than a threshold) to the ego-vehicle, the obstacle may be identified as a game object of the ego-vehicle. If an obstacle allocates much less attention (e.g., lower than a threshold) to the ego-vehicle, the obstacle may be identified as a non-game object of the ego-vehicle.

[0113] In some embodiments, the attention model may be constructed using a mathematical model such as y = softmax(a1x1+a2x2+a3x3,...), where softmax represents normalization, a1, a2, a3,... are weighting coefficients, and x1, x2, x3,... are relationship parameters between the obstacle and the ego-vehicle, such as longitudinal distance, lateral distance, vehicle speed difference, vehicle acceleration difference, and vehicle position relationship (forward, behind, left, right, etc.). Additionally, x1, x2, x3,... may also be normalized values, i.e., values ​​between 0 and 1. In some embodiments, the attention model may be implemented using a neural network. In this case, the output of the neural network is the attention value assigned to the ego-vehicle by the corresponding identified obstacle.

[0114] S20: For an interactive game task between the host vehicle and a game object, perform a plurality of strategic space releases from a plurality of strategic spaces of the host vehicle and the game object, and after performing one of the plurality of releases, determine a strategic feasibility region of the host vehicle and the game object based on each of the released strategic spaces, and determine a driving decision result of the host vehicle based on the strategic feasibility region.

[0115] In some embodiments, the determination result refers to a feasible behavior-action pair between the ego-vehicle and a game object in the strategy feasible region. In this case, in step S20, a determination process for a single vehicular interactive game between the ego-vehicle and any game object is completed, and a strategy feasible region for the ego-vehicle and the game object is determined. In some embodiments, performing multiple releases of the multiple strategy spaces includes performing sequential releases of each of the strategy spaces. That is, only one strategy space is released each time the release is performed, and multiple strategy spaces are cumulatively released after the sequential releases are performed. In this case, each strategy space spans at least one sampling dimension.

[0116] In some embodiments, based on general vehicle control safety or driving habits, the manner in which the ego-vehicle first accelerates and decelerates in the current lane is usually prioritized over the manner of lane change. Therefore, when multiple strategy spaces are released, the process of releasing multiple strategy spaces may optionally perform the release sequentially in the order of the vertical sampling dimension, the horizontal sampling dimension, and the temporal sampling dimension. Different dimensions may expand different strategy spaces. For example, when the vertical sampling dimension is released, the vertical sampling strategy space expands; when the horizontal sampling dimension is released, the horizontal sampling strategy space expands; the strategy space expands both in the horizontal sampling strategy space and in the vertical sampling strategy space expanded by the vertical sampling dimension; or when the temporal sampling dimension is released, multiple strategy spaces formed by multi-frame derivation expand.

[0117] In some embodiments, combinations of spaces in portions of different strategy spaces may also be sequentially released. For example, during a first release, a local space in the vertical sampling strategy space and a local space in the horizontal sampling strategy space are first sequentially released. During a second release, the remaining space in the vertical sampling strategy space and the remaining local space in the horizontal sampling strategy space are sequentially released.

[0118] In some embodiments, the multiple strategy spaces that are cumulatively unlocked may include a vertical sampling strategy space, a horizontal sampling strategy space, a strategy space that expands by combining the vertical sampling strategy space and the horizontal sampling strategy space, a strategy space that expands by combining the vertical sampling strategy space and the horizontal sampling strategy space each with a temporal sampling dimension, and a strategy space that expands by combining the vertical sampling dimension, the horizontal sampling dimension, and the temporal sampling dimension.

[0119] In some embodiments, as described above, the dimensions forming the strategy space may include a longitudinal sampling dimension, a lateral sampling dimension, or a temporal sampling dimension. With reference to a vehicle driving scenario, the dimensions are the longitudinal acceleration dimension, the lateral offset dimension, and the derivation depth corresponding to the multiple single-frame derivations included in one step of decision making. Correspondingly, the longitudinal sampling dimension used when the longitudinal sampling strategy space expands includes at least one of the longitudinal acceleration of the ego-vehicle and the longitudinal acceleration of the game object. The lateral sampling dimension used when the lateral sampling strategy space expands includes at least one of the lateral offset of the ego-vehicle and the lateral offset of the game object. The temporal sampling dimension includes multiple strategy spaces formed by successive multiple-frame derivations at corresponding successive points in time (i.e., sequentially increasing the derivation depth). A combination of the three dimensions may form multiple strategy spaces.

[0120] In this case, the values ​​of the strategy space expanded in each horizontal or vertical sampling dimension correspond to the sampling behavior of the ego-vehicle or game object, i.e., the behavior action Corresponds to.

[0121] In some embodiments, in the flowchart shown in FIG. 6, step S20 may include the following sub-steps S21 to S26.

[0122] S21: The first strategic space is released, and the strategic space of the player's vehicle and game objects is released. release Based on the strategy space, behavior-action pairs formed by a plurality of values ​​of at least one sampling dimension of the ego-vehicle and a plurality of values ​​of at least one sampling dimension of the game object are obtained one by one.

[0123] In some embodiments, when the strategy space is first released, a longitudinal sampling dimension is released that includes the longitudinal acceleration of the ego vehicle and the longitudinal acceleration of the game object, and the released strategy space that spans the longitudinal sampling dimension is the longitudinal sampling strategy space. For convenience of explanation, the strategy space is referred to as releaseThe strategy space may be referred to as a longitudinal sampling strategy space. In this case, the strategy space is a behavior action pair formed by the longitudinal acceleration of the ego vehicle to be evaluated and the longitudinal acceleration of the game object of the ego vehicle (i.e., another vehicle). In this case, multiple sampling values ​​may be set for each sampling dimension. Among these sampling values, multiple consecutive sampling values ​​with uniform sampling intervals may form a sampling range. Among these sampling values, multiple sampling values ​​scattered across the sampling dimension are discrete sampling points. For example, in the longitudinal acceleration dimension of the ego vehicle, uniform sampling is performed at a predetermined sampling interval to obtain multiple sampling values ​​in the longitudinal acceleration dimension of the ego vehicle. The number of multiple sampling values ​​is denoted as M1. That is, there are M1 longitudinal acceleration sampling actions for the ego vehicle. In the longitudinal acceleration dimension of the game object, uniform sampling is performed at a predetermined sampling interval to obtain multiple sampling values ​​in the longitudinal acceleration dimension of the game object. The number of multiple sampling values ​​is denoted as N1. That is, there are N1 longitudinal acceleration sampling actions for the game object. Therefore, the longitudinal sampling strategy space released the first time contains M1*N1 behavior-action pairs of the ego vehicle and the game object, obtained by combining the longitudinal acceleration sampling actions of the ego vehicle and the game object. For specific examples of the strategy space, see Table 1 or Table 2 below. The first row and first column of Table 1 or Table 2 are the longitudinal acceleration sampling values ​​of the ego vehicle and the game object (i.e., another vehicle O in the table), respectively. In Table 1, the game object is the crossing game vehicle. In Table 2, the game object is the oncoming game vehicle.

[0124] S22: Derive each behavior-action pair in the released strategy space into the traffic sub-scenario currently constructed by the ego-vehicle and game objects, and determine the cost value corresponding to each behavior-action pair.

[0125] In this case, the ego-vehicle and each game object further construct a separate traffic sub-scenario, each of which is a subset of the road scenario in which the ego-vehicle is located.

[0126] In some embodiments, each behavior in the strategy space action The cost value corresponding to the pair is the behavior performed by the ego-vehicle and the game object. action The cost value is determined based on at least one of a safety cost value, a comfort cost value, a lateral offset cost value, a passability cost value, a right-of-way cost value, a risk area cost value, and an inter-frame association cost value corresponding to the pair.

[0127] In some embodiments, a weighted sum of the aforementioned cost values ​​may be used. In this case, to distinguish between each cost value, the calculated weighted sum may be referred to as a total cost value. The smaller the total cost value, the greater the decision gain corresponding to the behavior-action pair performed by the ego-vehicle and the game object, indicating a higher probability that the behavior-action pair will be used as a decision result. The aforementioned cost values ​​are further described below.

[0128] S23: Add behavior action pairs whose cost values ​​are less than or equal to the cost threshold to the strategy feasible area of ​​the ego-vehicle and the game object, and the strategy feasible area is the game result of the ego-vehicle and the game object when the first strategy space is released.

[0129] A strategy feasible region is a set of executable behavior-action pairs. For example, in Table 1 below, table entries whose table content is Cy or Cg form a strategy feasible region.

[0130] S24: When the strategy feasible region (i.e., the game outcome) in the current strategy space is not null, use at least one feasible behavioral action pair of the host vehicle and the game object in the strategy feasible region as the decision result of the host vehicle and the game object, and end the release of the current strategy space.

[0131] When the strategy feasibility region is null, it indicates that there is no solution in the current strategy space. In this case, the strategy space is released a second time. That is, the next strategy space among the multiple strategy spaces is released. In this embodiment, the horizontal sampling dimension is released a second time, the horizontal sampling strategy space is expanded with the horizontal offset, and both the newly released horizontal sampling strategy space and the first released vertical sampling strategy space are used as the current strategy space. In this case, the strategy space currently used for the interactive game between the ego-vehicle and the game object is the strategy space expanded by combining the vertical sampling strategy space and the horizontal sampling strategy space.

[0132] The lateral sampling strategy space spans the ego-vehicle's lateral offset dimension and the game object's lateral offset dimension. For example, in the ego-vehicle's lateral acceleration dimension, uniform sampling is performed at a predetermined sampling interval to obtain multiple sampling values ​​in the ego-vehicle's lateral acceleration dimension. The number of multiple sampling values ​​is denoted as Q, i.e., there are Q lateral acceleration sampling behaviors for the ego-vehicle. In the game object's lateral offset dimension, uniform sampling is performed at a predetermined sampling interval to obtain multiple sampling values ​​in the game object's lateral offset dimension. The number of multiple sampling values ​​is denoted as R, i.e., there are R lateral offset sampling behaviors for the game object.

[0133] In this case, in the strategy space currently used for interactive games between the ego-vehicle and a game object, each behavioral action pair between the ego-vehicle and the game object is formed by the lateral offset sampling action of the ego-vehicle, the lateral offset sampling action of the game object, the longitudinal acceleration sampling action of the ego-vehicle, and the longitudinal acceleration sampling action of the game object.

[0134] Assume that the current strategy space is formed by Q values ​​of the ego vehicle's lateral offset, R values ​​of the game object's lateral offset, M2 values ​​of the ego vehicle's longitudinal acceleration, and N2 values ​​of the game object's longitudinal acceleration. The second released strategy space of the ego vehicle and the game object contains M2*N2*Q*R behavior-action pairs. See Table 3 below for a specific example. Each table entry in the lateral sampling strategy space at the top of Table 3 is associated with a table entry in the longitudinal sampling strategy space at the bottom of Table 3. In the table for the lateral sampling strategy space at the top of Table 3, the game object (i.e., another vehicle O in the table) is an oncoming vehicle.

[0135] S25: After the second strategy space release, each released behavior-action pair is derived into the traffic sub-scenario constructed by the ego-vehicle and game objects based on the current strategy space, the cost value corresponding to each behavior-action pair is determined, the strategy feasibility area is determined, and the game result is determined. For this step, see steps S22 and S23.

[0136] S26: When the strategy feasible region (i.e., the game outcome) in step S25 is not null, select a behavior-action pair from the strategy feasible region as the judgment result, and end the release of the current strategy space.

[0137] When the strategy feasibility region is null, it indicates that there is no solution in the current strategy space. In this case, the strategy space is opened for the third time. That is, the next strategy space among the multiple strategy spaces is opened. In this way, according to the above-mentioned method, other strategy spaces may continue to be opened sequentially, and the game outcome and decision results can continue to be determined.

[0138] In some embodiments, for a strategy space that is released multiple times, a strategy space that spans the i-th group of values ​​in the longitudinal acceleration dimension of the ego-vehicle and / or game object and the i-th group of values ​​in the lateral offset dimension of the ego-vehicle and / or game object may be released first. Additionally, when no strategy feasible region exists in the strategy space, a strategy space that spans the i+1-th group of values ​​in the longitudinal acceleration dimension of the ego-vehicle and / or game object and the i+1-th group of values ​​in the lateral offset dimension of the ego-vehicle and / or game object is subsequently released. That is, the strategy space that is released multiple times changes the local positions of the ego-vehicle and / or game object separately in a game space that spans all values ​​in the longitudinal acceleration dimension of the ego-vehicle and / or game object and all values ​​in the lateral offset dimension of the ego-vehicle and / or game object (i is a positive integer). The above uses the i-th group of values ​​as an example to show that several values ​​in each sampling dimension are sequentially released to sequentially search for strategy feasible regions in different local strategy spaces in the game space and determine a decision result. In this way, the strategy space corresponding to different local spaces is sequentially released, and the optimal decision result can be obtained in the multidimensional game space with the minimum number of searches, minimizing the use of the strategy space as much as possible, thereby reducing the requirements for hardware computing power.

[0139] For example, first, the strategy space that is expanded by the lateral offset value of 0 for the oncoming vehicle, the lateral offset value of 1 for the host vehicle, and all longitudinal acceleration values ​​of the host vehicle and the oncoming vehicle in Table 3 is released. Additionally, when there is no strategy feasible region in the strategy space, the strategy space that is expanded by the lateral offset value of 0 for the oncoming game vehicle, the lateral offset value of 2 or 3 for the host vehicle, and all longitudinal acceleration values ​​of the host vehicle and the oncoming game vehicle is released.

[0140] In some embodiments, if the strategy feasibility region for the ego-vehicle and game objects remains null after the strategy space has been released multiple times following the aforementioned steps, indicating that no solution remains. In this case, a conservative driving decision for the ego-vehicle may be implemented. This conservative decision may include a behavioral action to safely brake the ego-vehicle, a behavior to safely slow the ego-vehicle for driving, or a prompt or warning provided to the driver to take over control of the vehicle.

[0141] As described above, after steps S10 to S20 are performed, one single-frame derivation is completed. In some embodiments, if the strategy feasible region is not null after steps S10 to S20 are performed, the method may further include performing multiple rounds of release in the temporal sampling dimension and performing multi-frame derivation based on the evolution of derivation time (i.e., sequentially increasing the derivation depth, which is multiple consecutive instants). In this case, after one round of release is performed at a derivation instant (or time point) in the multiple rounds of release, one single-frame derivation is completed to determine the strategy feasible region of the ego-vehicle and game object. Additionally, when the strategy feasible region of the ego-vehicle and game object determined in the single-frame derivation is not null, the derived release is performed at the next time for the single-frame derivation at the next time until the multiple rounds of release in the temporal sampling dimension are completed or until the consecutive multi-frame derivations are completed.

[0142] In this case, multiple single-frame derivations are implemented in one step of judgment. As shown in Figure 7, T1 indicates the initial motion status of the ego-vehicle and game object. T2 indicates the motion status of the ego-vehicle and game object after the first frame is derived, i.e., the derivation result of the first frame. Tn indicates the motion status of the ego-vehicle and game object after the (n-1)th frame is derived.

[0143] In some embodiments, each time the temporal sampling dimension is released, the derivation time is moved backward by a preset time interval (e.g., 2 seconds or 5 seconds), i.e., moved to the next moment (also called a time point). Correspondingly, the derivation result of the current frame is used as the initial derivation condition for the next frame to derive the movement situation of the ego-vehicle and game objects at the next moment. Thus, according to this method, the temporal sampling dimension may continue to be released at the next moment to continue deriving multiple consecutive frames and continue to determine game results and judgment results.

[0144] As described above, during the release of the temporal sampling dimension, it is necessary to evaluate the behavioral judgment results of the ego-vehicle and game object determined by deriving two adjacent single frames and determine an inter-frame association cost value, as will be described later. The release of the temporal sampling dimension helps improve the consistency of vehicle behavior. For example, when the intention judgments corresponding to the motion situations or judgment results of the ego-vehicle and game object derived in multiple consecutive frames are the same or similar, the driving behavior of an intelligent driving vehicle implementing an intelligent driving judgment method will be more stable in the time domain, with smaller fluctuations in the driving trajectory and more comfortable vehicle driving.

[0145] The released temporal sampling dimension, i.e., the released strategy spaces at consecutive decision instants, are separately used to obtain strategy feasible regions corresponding to the ego-vehicle and the game objects, and to derive motion situations obtained after the ego-vehicle and the game objects have been sequentially executed based on the feasible time series corresponding to these strategy feasible regions. Long-term derivation may be performed on the motion situations and / or predicted driving trajectories of the ego-vehicle and the game objects, thereby ensuring the temporal consistency of the decision results.

[0146] After the multi-frame derivation is completed, if the overall gain of the multi-frame derivation satisfies the judgment requirement, it may be determined that the game results of each frame are gradually converging to a Nash equilibrium state. In this case, the judgment result of the first frame in the multi-frame derivation may be used as the driving judgment result of the host vehicle.

[0147] In some embodiments, if the overall gain of the multi-frame derivation does not satisfy the decision requirements, the decision result of the first frame may be reselected. The decision result of the first frame corresponds to the decision result of the single-frame derivation. Reselecting the decision result of the first frame means selecting another behavior-action pair as the decision result from the strategy feasibility region of the decision result of the first frame. For the reselected decision result, multi-frame derivation may be performed again to determine whether the reselected decision result can be used as the final decision result.

[0148] In some embodiments, when the judgment result of the first frame is selected for the first time and then reselected, or when it is reselected multiple times, the selection may be performed based on the sorting results of the cost values ​​corresponding to each behavior-action pair, and the judgment result corresponding to the behavior-action pair with a smaller total cost value is preferentially selected.

[0149] In some embodiments, different cost values ​​may have different weights, which may be correspondingly referred to as safety weights, comfort weights, lateral offset weights, passability weights, right-of-way weights, risk zone weights, and inter-frame association weights. Additionally, in some embodiments, weight values ​​may be assigned in the following order: safety weight > right-of-way weight > lateral offset weight > passability weight > comfort weight > risk zone weight > inter-frame association weight. In some embodiments, a normalization process may be performed separately on the cost values, and the range of values ​​is [0,1].

[0150] In some embodiments, the cost values ​​may be obtained through calculations based on different cost functions, which may be correspondingly referred to as a safety cost function, a comfort cost function, a passability cost function, a lateral offset cost function, and a right-of-way cost function.

[0151] In some embodiments, the safety cost value is obtained through a calculation based on a safety cost function that uses the relative distance between the ego-vehicle and another vehicle (i.e., game object) when they interact with each other as an independent variable, and the safety cost value is negatively correlated with the relative distance. For example, a larger relative distance between two vehicles indicates a smaller safety cost value. As shown in FIG. 8A, the safety cost function obtained after the normalization process is a piecewise function as follows, where C dist is the safety cost value, and dist is the relative distance between the ego-vehicle and the game object. For example, the minimum distance is defined as the minimum polygon distance between the ego-vehicle and the game object.

number

[0152] threLow is a lower distance threshold, which is 0.2 in Fig. 8A, and threHigh is an upper distance threshold, which is 1.2 in Fig. 8A. Optionally, the lower distance threshold threLow and the upper distance threshold threHigh may be dynamically adjusted based on the situation of the interaction between the host vehicle and another vehicle, for example, based on the relative speed, relative distance, relative angle, etc. between the host vehicle and another vehicle.

[0153] In some embodiments, the safety cost value defined by the safety cost function is positively correlated with the relative speed or relative angle. For example, when two vehicles encounter each other in opposite directions or laterally (laterally being the direction in which another vehicle crosses the ego vehicle), a higher relative speed or relative interaction angle between the two vehicles indicates a correspondingly higher safety cost value.

[0154] In some embodiments, the comfort cost value of a vehicle (e.g., ego-vehicle or game object) may be obtained through a calculation based on a comfort cost function that uses the absolute value of acceleration change (i.e., jerk) as an independent variable. As shown in Figure 8B, the comfort function obtained after the normalization process is a piecewise function as follows, where C comf is the comfort cost value, and jerk is the change in acceleration of the ego vehicle or game object.

number

[0155] threMiddle is the jerk midpoint threshold, which is 2 in the example of FIG. 8B, and threHigh is the upper jerk threshold, which is 4 in FIG. 8B. middle is the cost gradient of "jerk". That is, the larger the vehicle acceleration change, the worse the comfort and the larger the comfort cost value. Additionally, the comfort cost value increases faster after the vehicle acceleration change is greater than the midpoint threshold.

[0156] In some embodiments, the vehicle acceleration change may be a longitudinal acceleration change, a lateral acceleration change, or a weighted sum of both. In some embodiments, the comfort cost value may be a comfort cost value of the ego vehicle, a comfort cost value of the game object, or a weighted sum of both comfort cost values.

[0157] In some embodiments, the travelability cost value may be obtained through a calculation based on a travelability cost function that uses the speed change of the host vehicle or game object as an independent variable. For example, if a vehicle yields with a relatively large deceleration, it will result in a relatively large speed loss (the difference between the current speed and the future speed, i.e., acceleration) or a relatively long waiting time. In this case, the travelability cost value of the vehicle will be large. For example, if a vehicle yields with a relatively large acceleration, it will result in a relatively large speed increase (the difference between the current speed and the future speed, i.e., acceleration) or a relatively short waiting time. In this case, the travelability cost value of the vehicle will be small.

[0158] In some embodiments, the travelability cost value may also be obtained through a calculation based on a travelability cost function that uses the ratio of the relative speed between the host vehicle and the game object as an independent variable. For example, before the behavior pair is executed, the ratio of the absolute value of the host vehicle's speed to the sum of the absolute values ​​of the host vehicle's and the game object's speed is relatively large, and the ratio of the absolute value of the game object's speed to the sum of the absolute values ​​of the host vehicle's and the game object's speed is relatively small. After the behavior action pair is executed, if the host vehicle yields with a relatively large deceleration, the host vehicle's speed loss is large and the speed ratio is small. In this case, the travelability cost value corresponding to the behavior action pair executed by the host vehicle is relatively large. However, if the game object breaks with a relatively large acceleration after the behavior action pair is executed, the game object's speed increases and the speed ratio increases. In this case, the travelability cost value corresponding to the behavior action executed by the game object is relatively small.

[0159] In some embodiments, the travelability cost value is a weighted sum of the travelability cost value of the ego-vehicle corresponding to the behavior action pair performed by the ego-vehicle, or the travelability cost value of the game object corresponding to the behavior action pair performed by the game object, or both.

[0160] In some embodiments, as shown in FIG. 8C, the traffic cost function obtained after the normalization process is a piecewise function as follows: pass is the travelability cost value, and speed is the absolute value of the vehicle speed.

number

[0161] The absolute value of the speed at the midpoint of the vehicle is speed0, the absolute maximum value of the speed of the vehicle is speed1, and C middle is the cost gradient of the speed. That is, the larger the absolute value of the vehicle speed, the better the passability and the smaller the passability cost value. Additionally, after the absolute value of the vehicle speed becomes larger than the midpoint threshold, the passability cost value decreases faster.

[0162] In some embodiments, the progress right information corresponding to the host vehicle may be determined based on the acquired user profile of the host vehicle or game object. For example, if the driving behavior of the game object is aggressive and more likely to make a cut-in decision, the progress right is high. If the driving behavior of the game object is conservative and more likely to adopt a yielding policy, the progress right is low. A high progress right tends to maintain the specified motion situation or specified driving behavior, and a low progress right tends to change the specified motion situation or specified driving behavior.

[0163] In some embodiments, the user profile may be determined based on the user's gender, age, or historical behavior / action completion status. In some embodiments, a cloud server may obtain the data necessary to determine the user profile and determine the user profile. If a behavior / action pair performed by the host vehicle and / or game object allows a vehicle with a high progression right to change its movement situation, the progression right cost value corresponding to that behavior / action pair will be relatively large and the gain will be relatively small.

[0164] In some embodiments, a larger entitlement cost value is determined based on a behavior decision that causes a change in the motion situation of the vehicle with the higher entitlement, thereby increasing the penalty. In other words, due to this feedback mechanism, a behavior-action pair that causes the vehicle with the higher entitlement to maintain its current motion situation has a larger entitlement gain, i.e., a relatively smaller entitlement cost value.

[0165] In some embodiments, as shown in FIG. 8D, the progress right cost function obtained after the normalization process is a piecewise function as follows, where C roadRight is the right of way cost value, and acc is the absolute value of the vehicle's acceleration.

number

[0166] threHigh is the upper acceleration threshold, which is 1 in FIG. 8D. That is, the higher the vehicle acceleration, the higher the right of way cost value.

[0167] In other words, this right of way cost function allows a vehicle with a high right of way to have a behavior action that maintains the current movement situation with a relatively small right of way cost value, and avoids a vehicle with a high right of way from determining a behavior action pair that changes the current movement situation.

[0168] In some embodiments, the vehicle acceleration may be a longitudinal acceleration or a lateral acceleration. That is, in the lateral offset dimension, a large lateral change also results in a large progress entitlement cost value. In some embodiments, the progress entitlement cost value may be a weighted sum of progress entitlement cost values ​​corresponding to behavior-action pairs performed by the ego-vehicle, progress entitlement cost values ​​corresponding to behavior-action pairs performed by the game object, or both.

[0169] In some embodiments, for example, when a vehicle is in a risk area on a road (where the vehicle has a relatively high driving risk and needs to leave the risk area as soon as possible), a relatively large risk area cost value needs to be applied to the vehicle yielding strategy to increase the penalty. Instead of selecting the vehicle yielding behavior, the vehicle cutting in behavior is selected as the decision result, so that the vehicle leaves the risk area as soon as possible. That is, a decision is made that the vehicle leaves the risk area as soon as possible. This ensures that the vehicle leaves the risk area as soon as possible and does not cause serious impact on traffic.

[0170] In other words, due to the feedback mechanism where the larger the risk area cost value, the smaller the strategic gain, the vehicles in the risk area on the road will not take the yielding behavior, i.e., they will abandon the behavior decision (which has a relatively large risk area cost value) that makes the vehicles in the risk area on the road take the yielding behavior, and select the judgment result (which has a relatively small risk area cost value) that allows the vehicles in the risk area on the road to take the cut-in behavior in order to leave the risk area as soon as possible, thereby preventing the vehicles in the risk area on the road from getting stuck and causing serious impacts on traffic.

[0171] In some embodiments, the risk zone cost value may be a weighted sum of risk zone cost values ​​corresponding to behavior-action pairs performed by the ego-vehicle in the risk zone on the road, or risk zone cost values ​​corresponding to behavior-action pairs performed by game objects in the risk zone on the road, or both. In some embodiments, the lateral offset cost value may be obtained through a calculation based on the lateral offset amount of the ego-vehicle or game object. As shown in Figure 8E, the lateral offset cost function in the right half space obtained after the normalization process is a piecewise function as follows, where C offsetis the lateral offset cost value, and offset is the vehicle's lateral offset in meters. In this case, the formula for the left half-space of the coordinate plane may be obtained by negating the formula for the right half-space.

number

[0172] threMiddle is the lateral offset middle value, e.g., road soft boundary, and C middle is the first lateral offset cost gradient, and threHigh is the upper lateral offset threshold, e.g., the road hard boundary. In other words, the greater the vehicle's lateral offset, the smaller the lateral offset gain and the larger the lateral offset cost value. Additionally, after the vehicle's lateral offset amount becomes larger than the lateral offset intermediate value, the lateral offset cost value increases faster to increase the penalty. For example, after the vehicle's lateral offset amount becomes larger than the upper lateral offset threshold, the lateral offset cost value is set to a fixed value of 1.2 and the penalty is increased.

[0173] In some embodiments, the lateral offset cost value may be a weighted sum of lateral offset cost values ​​corresponding to behavior-action pairs performed by the ego-vehicle, or lateral offset cost values ​​corresponding to behavior-action pairs performed by the game object, or both.

[0174] In the multi-frame derivation step described above, the behavior of the ego-vehicle and game objects determined by deriving two adjacent single frames must be evaluated, and an inter-frame association cost value must be determined.

[0175] In some embodiments, as shown in FIG. 8F , the intention determination of the ego-vehicle in the previous frame K is to interrupt the game object. When the intention determination of the ego-vehicle in the current frame K+1 is to interrupt the game object, the corresponding inter-frame association cost value is relatively small, for example, 0.3. However, because the default value is 0.5, this is a reward. However, when the intention determination of the ego-vehicle in the current frame K+1 is to yield the game object, the corresponding inter-frame association cost value is relatively large, for example, 0.8. However, because the default value is 0.5, this is a penalty. In this case, a strategy in which the intention determination of the ego-vehicle in the current frame interrupts the game object is selected as a feasible solution for the current frame. Based on the penalty or reward for the inter-frame association cost value, it can be ensured that the intention determination of the ego-vehicle in the current frame and the intention determination of the previous frame are consistent, so that the movement situation of the ego-vehicle in the current frame and the movement situation in the previous frame are consistent, and the behavior determination of the ego-vehicle is stabilized in the time domain.

[0176] In some embodiments, the inter-frame association cost value may be obtained through a calculation based on an intent determination of the ego vehicle in a previous frame and an intent determination of the ego vehicle in a current frame, may be obtained through a calculation based on an intent determination of a game object in a previous frame and an intent determination of a game object in a current frame, or may be obtained after weighting is performed based on the ego vehicle and the game object.

[0177] In some embodiments, as shown in FIG. 4, after the judgment result is determined in step S20, the following steps S30 and / or S40 may be further included.

[0178] S30: The host vehicle generates longitudinal / lateral control amounts based on the judgment result, and the host vehicle's driving system then executes the longitudinal / lateral control amounts to implement the host vehicle's predicted driving trajectory.

[0179] In some embodiments, the control device of the host vehicle generates longitudinal / lateral control amounts based on the determination result and transmits the longitudinal / lateral control amounts to the driving system 13, so that the driving system 13 performs driving control including power control, steering control, and braking control for the vehicle. In this way, the vehicle implements the predicted driving trajectory of the host vehicle based on the determination result.

[0180] S40: The judgment result is displayed on the display device in a manner that can be understood by the user.

[0181] The driving judgment result of the host vehicle includes the behavior of the host vehicle. Based on the behavior of the host vehicle and the motion status of the host vehicle acquired at the judgment start moment in the current frame derivation, an intention judgment of the host vehicle, for example, whether to cut in, yield, or avoid, may be predicted, and a predicted driving trajectory of the host vehicle may further be predicted. In some embodiments, the judgment result is displayed on a display device in the vehicle cockpit in a manner understandable to a user, for example, in the form of a predicted driving trajectory, an arrow indicating the intention judgment, text indicating the intention judgment, etc. In some embodiments, when the predicted driving trajectory is displayed, the display device in the vehicle cockpit may display the predicted driving trajectory in the form of a partial enlarged view with reference to a current traffic scenario (e.g., a graphical traffic scenario) of the vehicle. In some embodiments, an audio playback system may also be included. The audio playback method may prompt the user with the intention judgment or the determined strategy label.

[0182] In some embodiments, taking into account one-way dialogue judgment between the host vehicle and a non-game object of the host vehicle, or one-way dialogue judgment between a game object of the host vehicle and a non-game object of the game object of the host vehicle, in another embodiment shown in Figure 9, the following steps may be further included between steps S10 to S20.

[0183] S15: Constrain the strategy space of the ego-vehicle or game object based on the motion status of the non-game object.

[0184] In some embodiments, the range of values ​​in each sampling dimension of the ego-vehicle may be constrained based on the motion of non-game objects of the ego-vehicle.

[0185] In some embodiments, the range of values ​​in each sampling dimension for the ego-vehicle game object may be constrained based on the motion of non-game objects relative to the ego-vehicle game object.

[0186] In some embodiments, a range may be one or more sampling ranges in a sampling dimension, or may be multiple discrete sampling points. A constrained range may be a partial range.

[0187] In some embodiments, step S15 includes determining a range of values ​​in each sampling dimension of the ego-vehicle under constraints of the motion status of the non-game object in a process of one-way interaction between the ego-vehicle and a non-game object of the ego-vehicle, or determining a range of values ​​in each sampling dimension of the ego-vehicle's game object under constraints of the motion status of the non-game object of the ego-vehicle's game object in a process of one-way interaction between the ego-vehicle's game object and a non-game object of the ego-vehicle's game object.

[0188] The non-game objects do not participate in the interactive game, and the motion status of the non-game objects remains unchanged. Therefore, the value range in each sampling dimension of the host vehicle is constrained based on the motion status of the game object of the host vehicle, and the value range in each sampling dimension of the game object of the host vehicle is constrained based on the motion status of the non-game objects of the game object of the host vehicle, and then step S20 is executed again. This helps to reduce the game space and strategy space in the judgment process of the single-vehicle interactive game between the host vehicle and the game object of the host vehicle, and reduces the computing power used in the judgment process of the interactive game.

[0189] In some embodiments, based on the constraints on the ego-vehicle, step S15 first receives the motion status of the non-game object of the ego-vehicle and observes the feature of the non-game object, then calculates the conflict zone between the ego-vehicle and the non-game object, and determines the feature of the ego-vehicle, i.e., the critical action. As described above, based on the position, speed, acceleration, and / or driving trajectory of the non-game object, the critical action corresponding to the ego-vehicle's intention determination, such as avoidance, cutting in, or yielding, is calculated, and a feasible interval for the non-game object in each sampling dimension of the ego-vehicle, i.e., a constrained value range for the non-game object in each sampling dimension of the ego-vehicle, is generated.

[0190] In some embodiments, based on constraints on the ego-vehicle game object, the ego-vehicle may be modified to generate a constrained range in each sampling dimension of the ego-vehicle game object relative to the non-game object, compared to the aforementioned constraints on the ego-vehicle.

[0191] In some embodiments, when the host vehicle has a non-game object C, an interactive game judgment between the host vehicle A and the host vehicle's game object B is first performed to determine the corresponding strategy feasible area AB, the non-game feasible area AC of the host vehicle and the non-game object C is introduced, the intersection of the strategy feasible area AB and the non-game feasible area AC is obtained to obtain a final strategy feasible area ABC, and the driving judgment result of the host vehicle is determined based on the final strategy feasible area.

[0192] In some embodiments, when the game object B of the host vehicle has a non-game object D, an interactive game judgment between the host vehicle A and the game object B of the host vehicle is first performed to determine the corresponding strategy feasible area AB, the non-game feasible area BD of the game object B of the host vehicle and the non-game object D of the game object B of the host vehicle are introduced, the intersection of the strategy feasible area AB and the non-game feasible area BD is obtained to obtain a final feasible area ABD, and the driving judgment result of the host vehicle is determined based on the final strategy feasible area.

[0193] In some embodiments, when the host vehicle has a non-game object C and the host vehicle's game object B has a non-game object D, an interactive game judgment between the host vehicle A and the host vehicle's game object B is first performed to determine the corresponding strategy feasible area AB, and the non-game feasible area AC between the host vehicle A and the non-game object C and the non-game feasible area BD between the host vehicle's game object B and the non-game object D of the host vehicle's game object B are introduced, and the intersection of the strategy feasible area AB, the non-game feasible area AC, and the non-game feasible area BD is obtained to obtain a final feasible area ABCD, and the driving judgment result of the host vehicle is determined based on the final strategy feasible area.

[0194] The above specifically describes the step of determining a feasible behavioral action pair between the ego-vehicle and the single game object by sequentially unlocking multiple strategy spaces for the ego-vehicle and the single game object. In some embodiments, when the ego-vehicle has two or more game objects, for example, two game objects including a first game object and a second game object, the intelligent driving decision-making method provided in this embodiment of this application includes the following steps:

[0195] Step 1: From the multiple strategy spaces of the host vehicle and the first game object, sequentially unlock the strategy spaces to determine the strategy feasible region of the host vehicle's travel with respect to the first game object. Determining the strategy feasible region of the host vehicle's travel with respect to the first game object is similar to step S20. The details will not be described again here.

[0196] Step 2: From the multiple strategy spaces of the host vehicle and the second game object, sequentially unlock the strategy spaces to determine the strategy feasible region of the host vehicle's travel with respect to the second game object. Determining the strategy feasible region of the host vehicle's travel with respect to the second game object is similar to step S20. The details will not be described again here.

[0197] Step 3: Determine a driving decision result for the host vehicle based on each strategy feasible region of the host vehicle and each game object. In some embodiments, an intersection is obtained based on each strategy feasible region to obtain a final strategy feasible region, and then a decision result is determined based on the strategy feasible region. In some embodiments, the decision result may be the behavior-action pair with the smallest cost value in the strategy feasible region.

[0198] A specific implementation of the intelligent driving decision-making method provided in an embodiment of this application will be described below. This specific implementation will still be described using a traffic scenario in which a vehicle is traveling on a road as an example. As shown in Figure 10, the scenario of this specific implementation is as follows: an ego-vehicle 101 is traveling on a road, the road is a two-way single lane, and an ego-vehicle 103 is traveling in the opposite direction, that is, it is an oncoming game vehicle. There is a vehicle 102 ahead of the ego-vehicle and is about to cross the road, that is, it is a crossing game vehicle. Please refer to the flowchart shown in Figure 11. A driving control method provided in this specific implementation will be described in detail below. This method includes the following steps:

[0199] S110: The host vehicle acquires external environment information of the host vehicle via the environment information acquisition device 11.

[0200] For this step, see step S11, and the details will not be described again here.

[0201] S120: The ego-vehicle determines game objects and non-game objects.

[0202] For this step, see step S12. The details will not be described again here. In this step, it is determined that the game object of the host vehicle is a crossing game vehicle and that the game object is an oncoming game vehicle.

[0203] S130: Sequentially release the strategic spaces from the multiple strategic spaces of the own vehicle and the crossing game vehicle, and determine the game results of the own vehicle and the crossing game vehicle. Specifically, the following steps S131 and S132 may be included.

[0204] S131: According to the principle of first releasing the longitudinal sampling dimension and then releasing the lateral sampling dimension, the longitudinal acceleration dimension of the ego vehicle and the crossing game vehicle is released.

[0205] The longitudinal acceleration dimension of the ego vehicle and the crossing game vehicle is released from the multidimensional game space of the ego vehicle and the crossing game vehicle, expanding the first longitudinal sampling strategy space of the ego vehicle and the crossing game vehicle. Taking into consideration the longitudinal / lateral dynamics, kinematic constraints, relative positional relationship and relative speed relationship between the ego vehicle and the crossing game vehicle, and considering that the two vehicles have the same mobility capability, the value range of the longitudinal acceleration of both the ego vehicle and the crossing game vehicle is determined to be [-4, 3]. The unit is m / s 2 where m represents meters and s represents seconds. Based on the computing power of the vehicle and the preset judgment accuracy, the sampling intervals of the vehicle and the crossing game vehicle are both set to 1 m / s. 2 It is decided that: [Table 1]

[0206] The strategy space that is expanded when displayed as a two-dimensional table is shown in Table 1. In Table 1, the first row lists all values ​​Ae of the longitudinal acceleration of the ego vehicle, and the first column lists all values ​​Ao1 of the longitudinal acceleration of the crossing game vehicle. In other words, the longitudinal sampling strategy space of the ego vehicle and the crossing game vehicle that is released this time contains 8 x 8, or 64, longitudinal acceleration behavior-action pairs of the ego vehicle and the crossing game vehicle.

[0207] S132: According to a predefined method for determining each cost value such as each cost function, calculate a cost value corresponding to each behavior action pair in the longitudinal sampling strategy space of the ego vehicle and the crossing game vehicle, and determine a strategy feasible region.

[0208] In the 64 released behavioral action pairs in Table 1, after the ego vehicle and the crossing game vehicle performed nine sampling actions, the traffic sub-scenario constructed by the ego vehicle and the crossing game vehicle had very poor traffic conditions (e.g., braking and stopping), and the actions were infeasible solutions. In Table 1, these action pairs are identified using the label "0".

[0209] Of the 64 behavioral pairs released, after the ego vehicle and the crossing game vehicle performed 39 sampling actions, the traffic sub-scenario constructed by the ego vehicle and the crossing game vehicle had very poor safety (e.g., collision), and the actions were infeasible solutions. In Table 1, these behavioral pairs are identified using the label “-1”.

[0210] After the ego vehicle and the crossing game vehicle perform 3 + 13 (i.e., 16) sampled actions among the 64 released behavioral action pairs, the traffic sub-scenario constructed by the ego vehicle and the crossing game vehicle has a weighted sum of the safety cost value, comfort cost value, passability cost value, lateral offset cost value, right-of-way cost value, risk area cost value, and inter-frame association cost value that is greater than the preset cost threshold. This is a feasible solution in the strategy space, forming the strategy feasible region for the ego vehicle and the crossing game vehicle.

[0211] The vertical sampling strategy space is freed and no horizontal offset is involved, therefore the horizontal offset cost value is 0. The decision is made for the current frame and the decision results of previous frames are not involved, therefore the inter-frame offset cost value is 0.

[0212] In this case, the interactive game between the ego vehicle and the cross-sampling game vehicle finds enough feasible solutions in the vertical sampling strategy space, and there is no need to continue finding solutions in the horizontal sampling dimension. In this case, the number of behavior-action pairs explored is 64, and this round of the game consumes less computing power and less computing time.

[0213] Additionally, a judgment label may be further added to each behavior-action pair based on a cost value corresponding to each behavior-action pair within the strategy feasible region.

[0214] After the host vehicle and the crossing game vehicle perform the three sampling actions, the behavior judgment of the host vehicle and the crossing game vehicle is that the host vehicle accelerates to move and the crossing game vehicle decelerates to move. Based on the motion status of the host vehicle and the crossing game vehicle acquired at the judgment start moment in the current frame derivation, it may be derived that the host vehicle passes through the conflict area in front of the crossing game vehicle after the host vehicle and the crossing game vehicle perform any one of the three sampling actions. Therefore, the intention judgment corresponding to these behavior-action pairs is determined to be that the host vehicle will cut in on the crossing game vehicle. Correspondingly, the cut-in judgment label of the host vehicle, i.e., "Cg" in Table 1, is set for the three behavior-action pairs.

[0215] After the host vehicle and the crossing game vehicle perform the 13 sampling actions, the behavior judgment of the host vehicle and the crossing game vehicle is that the crossing game vehicle accelerates to move and the host vehicle decelerates to move. Based on the motion status of the host vehicle and the crossing game vehicle acquired at the judgment start moment in the current frame derivation, it may be derived that after the host vehicle and the crossing game vehicle perform any one of the 13 sampling actions, the crossing game vehicle passes through the conflict area in front of the host vehicle. Therefore, the intention judgment corresponding to these behavior-action pairs is determined to be that the crossing game vehicle cuts in on the host vehicle. Correspondingly, a crossing game vehicle cut-in judgment label, i.e., "Cg" in Table 1, is set for the 13 behavior-action pairs.

[0216] S140: Sequentially release strategic spaces from among the multiple strategic spaces of the own vehicle and the oncoming game vehicle, and determine the game results of the own vehicle and the corresponding game vehicle. Specifically, the following steps S141 to S144 may be included.

[0217] S141: According to the principle of first releasing the vertical sampling dimension and then releasing the horizontal sampling dimension, the vertical sampling dimension of the own vehicle and the oncoming game vehicle is released, and the first vertical sampling strategy space of the own vehicle and the oncoming game vehicle is expanded.

[0218] The longitudinal acceleration dimension of the host vehicle and the oncoming game vehicle is released from the multidimensional game space of the host vehicle and the oncoming game vehicle, expanding the first longitudinal sampling strategy space of the host vehicle and the oncoming game vehicle. Taking into consideration the longitudinal / lateral dynamics of the host vehicle and the oncoming vehicle, kinematic constraints, and the relative positional and relative speed relationships between the host vehicle and the oncoming game vehicle, the value range of the longitudinal acceleration of both the host vehicle and the oncoming vehicle is determined to be [-4, 3]. The unit is m / s 2 The sampling intervals for both the host vehicle and the oncoming game vehicle are 1 m / s. 2 It is decided that:

[0219] The strategy space that expands when displayed as a two-dimensional table is shown in Table 2. In Table 2, the first row lists all values ​​Ae of the longitudinal acceleration of the host vehicle, and the first column lists all values ​​Ao2 of the longitudinal acceleration of the oncoming game vehicle. In other words, the longitudinal sampling strategy space of the host vehicle and the oncoming game vehicle that is released this time includes 8 x 8 pairs of longitudinal acceleration behavior actions of the host vehicle and the oncoming game vehicle, i.e., a total of 64 pairs. [Table 2]

[0220] S142: According to a predefined method for determining each cost value such as each cost function, calculate a cost value corresponding to each behavior action pair in the longitudinal sampling strategy space of the host vehicle and the oncoming game vehicle, and determine a strategy feasible region.

[0221] Among the 64 behavioral action pairs released, after the ego vehicle and the oncoming game vehicle performed nine sampling actions, the traffic sub-scenario constructed by the ego vehicle and the oncoming game vehicle had very poor traffic conditions (e.g., braking and stopping), and the actions were infeasible solutions. In Table 2, these action pairs are identified using the label "0".

[0222] Of the 64 behavioral pairs released, after the ego vehicle and the oncoming game vehicle performed 55 sampling actions, the traffic sub-scenario constructed by the ego vehicle and the oncoming game vehicle had very poor safety (e.g., collision), and the actions were infeasible solutions. In Table 2, these behavioral pairs are identified using the label "-1".

[0223] In other words, after the host vehicle and the oncoming vehicle execute the 64 released behavior-action pairs, the traffic sub-scenario constructed by the host vehicle and the oncoming vehicle will have a safety cost value or a passability cost value greater than a preset cost threshold. There is no feasible solution in the strategy space released the first time, and the strategy feasible region of the host vehicle and the oncoming vehicle is null.

[0224] S143: The lateral offset dimension of the host vehicle is released, and the second strategy space of the host vehicle and the oncoming game vehicle is expanded by the longitudinal acceleration dimension of the host vehicle and the oncoming game vehicle.

[0225] Specifically, some values ​​of the lateral offset dimension of the host vehicle and some values ​​of the longitudinal acceleration dimension of the host vehicle and the oncoming vehicle are released, widening the second strategy space for the host vehicle and the oncoming vehicle.

[0226] First, a maximum lateral sampling strategy space spanning the lateral offset dimension of the ego-vehicle and the oncoming vehicle is determined from the multidimensional game space of the ego-vehicle and the oncoming vehicle. Figure 10 is a schematic diagram of the lateral sampling actions determined for each of the two vehicles for lateral offset sampling. In other words, multiple lateral offset behavioral actions correspond to multiple parallel lateral offset trajectories that can be executed by the vehicle.

[0227] Considering the longitudinal and lateral dynamics of the ego vehicle and the oncoming vehicle, kinematic constraints, and the relative position and velocity relationships between the ego vehicle and the oncoming vehicle, the longitudinal offset value range for both the ego vehicle and the oncoming vehicle is determined to be [-3, 3]. The unit is m, where m represents meters. During sampling, the sampling interval for both the ego vehicle and the oncoming vehicle is determined to be 1 m based on the computing power of the ego vehicle and the preset judgment accuracy. In this case, the freed lateral sampling strategy space for the ego vehicle and the oncoming vehicle is displayed as a two-dimensional table, as shown in the upper subtable of Table 3. In the upper subtable of Table 3, the first row lists all values ​​of the lateral offset of the ego vehicle, Oe, and the first column lists all values ​​of the lateral offset of the oncoming vehicle, Oo2. Therefore, the lateral sampling strategy space spanning the lateral offset dimension of the ego vehicle and the oncoming vehicle contains a maximum of 7 × 7, i.e., 49 lateral offset behavior pairs between the ego vehicle and the oncoming vehicle.

[0228] When a vehicle is traveling, it cannot independently perform lateral offset without taking longitudinal behavioral actions. Therefore, it is necessary to free up multiple behavioral action pairs of the ego-vehicle and the oncoming vehicle in the longitudinal acceleration dimension while freeing up the lateral sampling strategy space of the ego-vehicle and the oncoming vehicle.

[0229] To reduce computing power and resources, only some values ​​of the lateral offset dimension of the ego vehicle are released during this release. Additionally, some values ​​of the lateral offset dimension of the ego vehicle and some values ​​of the longitudinal sampling dimension and longitudinal acceleration dimension of the ego vehicle and the oncoming game vehicle are spread across the second released strategy space. In this case, the lateral offset value of the oncoming vehicle is 0. As shown in the upper subtable of Table 3, from the lateral sampling strategy space, the lateral offset value of the oncoming vehicle is 0, and the lateral offset value of the ego vehicle is -3, -2, -1, 0, 1, 2, or 3, respectively, forming seven lateral offset behavior-action pairs. The seven lateral offset behavior-action pairs are separately combined with the previously released 64 longitudinal acceleration behavior-action pairs of the ego vehicle and the oncoming game vehicle (as shown in Table 2) to obtain 7 x 64, or 448, behavior-action pairs. In this case, in each behavior-action pair, the lateral offset value of the oncoming game vehicle is 0. In the strategy space corresponding to the longitudinal acceleration sampling of the ego vehicle and the oncoming game vehicle, up to 64 behavior-action pairs may be released. In comparison, the number of behavior-action pairs released in this case increases by six times, and is seven times the number released in the first round.

[0230] S144: According to a method for determining each cost value such as each cost function, a cost value corresponding to each behavior action pair in the second released strategy space, which is expanded by several values ​​in the lateral offset dimension of the own vehicle and several values ​​in the longitudinal acceleration dimension of the own vehicle and the oncoming game vehicle, is separately calculated to determine a strategy feasible region.

[0231] As shown in the lower sub-table of Table 3, when the lateral offset value of the host vehicle is 1, in the 64 released longitudinal acceleration behavior action pairs of the host vehicle and the oncoming game vehicle, after the host vehicle and the oncoming game vehicle perform 16 sampling actions, the traffic sub-scenario constructed by the host vehicle and the oncoming game vehicle has too poor trafficability (e.g., braking and stopping), and the actions are infeasible solutions. In Table 3, these actions are identified using the label "0".

[0232] When the lateral offset value of the ego vehicle is 1, among the 64 released longitudinal acceleration behavior action pairs between the ego vehicle and the oncoming game vehicle, after the ego vehicle and the oncoming game vehicle perform 48 sampled actions, the weighted sum of the safety, comfort, passability, lateral offset cost value, right-of-way cost value, risk area cost value, and inter-frame association cost value in the traffic sub-scenario constructed by the ego vehicle and the oncoming game vehicle is greater than the preset cost threshold. This is a feasible solution in the strategy space, forming the strategy feasible region of the ego vehicle and the oncoming game vehicle. In Table 3, the 48 action pairs are identified using the label "1." In this case, the interactive game is in the current frame, and the decision results of the previous frame are not involved. Therefore, the inter-frame association cost value is 0.

[0233] At this time, 48 feasible solutions have been found in the interactive game between the ego vehicle and the oncoming vehicle, and there is no need to continue searching for solutions in the game space between the ego vehicle and the oncoming vehicle. The total number of behavior-action pairs explored is 64, and this round of the game consumes less computing power and less computing time.

[0234] That is, the lateral offset behavior-action pairs in which the lateral offset value of the oncoming vehicle is 0 and the lateral offset value of the host vehicle is 1 are separately combined with the 64 longitudinal acceleration behavior-action pairs of the host vehicle and the oncoming vehicle that were released earlier to obtain 64 behavior-action pairs, among which there are 48 feasible solutions (the feasible solutions are shown in Table 3 with shadows and shading). These feasible solutions are added to the strategy feasible region of the host vehicle and the oncoming game vehicle.

[0235] This is because the ego vehicle and the oncoming vehicle are already staggered laterally after the ego vehicle is laterally offset to the right by 1 m (the ego vehicle is used as the reference, and a lateral offset to the right is positive, and a lateral offset to the left is negative), so in the longitudinal sampling strategy space of the ego vehicle and the oncoming vehicle, the strategy feasibility region covers all cases except when both vehicles are braked to a stop (action pairs are shown with shading in Table 3).

[0236] Also, compared to Table 2, when the longitudinal accelerations of the host vehicle and the oncoming vehicle in the lower sub-table of Table 3 are all -1, the label of the corresponding behavior-action pair is adjusted from "-1" to "0." This is because when the host vehicle is laterally offset 1 m to the right, there is no longer a collision risk between the host vehicle and the oncoming game vehicle. In the traffic scenario constructed by the oncoming game vehicle and the host vehicle to which these behavior-action pairs are mapped, traffic conditions are poor (braking and stopping), and the actions are still infeasible solutions. However, the label is adjusted from "-1" to "0."

[0237] Additionally, in this case, the intention determination may be determined for the host vehicle and the oncoming vehicle, and a strategy label may be set. For details, see step S132. The details will not be described again here. [Table 3]

[0238] In order to free up the strategy space of the ego-vehicle and the oncoming vehicle, multiple sampling values ​​may be selected from the lateral offset dimension of the ego-vehicle, for example, the lateral offset value of the ego-vehicle is 2 or 3, respectively, which opens up more strategy space together with the longitudinal acceleration sampling strategy space of the ego-vehicle and the oncoming vehicle.

[0239] In this embodiment, the lateral offset behavior-action pair in which the lateral offset value of the oncoming game vehicle is 0 and the lateral offset value of the host vehicle is 1, and the 64 longitudinal acceleration behavior-action pairs of the host vehicle and the oncoming game vehicle that were released earlier, are used to expand the strategy space that is released the second time. Additionally, 48 feasible solutions are found in the strategy space. Therefore, there is no need to release other strategy spaces. In this way, the interactive game consumes less computing power and less computing time. [Table 4] TIFF0007707412000010.tif45170

[0240] S150: The common area between the strategy feasible area of ​​the host vehicle and the oncoming game vehicle and the strategy feasible area of ​​the host vehicle and the crossing game vehicle is obtained, and the game result of the host vehicle is determined.

[0241] Based on the determined intersection between the strategy feasible area of ​​the own vehicle and the crossing game vehicle and the strategy feasible area of ​​the own vehicle and the oncoming game vehicle, a common feasible area of ​​both is found, and a feasible solution with the smallest cost value (i.e., the best gain) is found from the common feasible area.

[0242] Table 4 shows the feasible solutions with the smallest cost value (i.e., the best gain) found from the common feasible area between the strategy feasible areas of the own vehicle and the oncoming vehicle in Table 3 and the strategy feasible areas of the own vehicle and the crossing game vehicle in Table 1. The feasible solutions are game decision-action pairs of the own vehicle, the oncoming vehicle, and the crossing game vehicle, and are multidimensional behavior-action pairs combined by the longitudinal acceleration of the own vehicle, the longitudinal acceleration of the oncoming vehicle, the lateral offset of the own vehicle, and the longitudinal acceleration of the crossing game vehicle.

[0243] In other words, the vehicle will move at -2 m / s to give way. 2The vehicle decelerates at a longitudinal acceleration of 1 m / s and offsets laterally to the right by 1 m to avoid the oncoming vehicle. 2 The vehicle accelerates at a longitudinal acceleration of 1 m / s and passes through the competition area. 2 The vehicle accelerates at a longitudinal acceleration of 1 / 2 and passes through the conflict area.

[0244] After the behavioral actions are performed by the own vehicle, the oncoming vehicle, and the crossing game vehicle, the intention determinations are distinct as follows: the crossing game vehicle cuts in on the own vehicle, the oncoming game vehicle cuts in on the own vehicle, the own vehicle drives laterally to the right to avoid the oncoming game vehicle, and the own vehicle yields to the crossing game vehicle.

[0245] S160: Select a judgment result from the game result of the host vehicle, select an action pair corresponding to the minimum cost value, and determine an executable action for the host vehicle based on the action pair, and the executable action for the host vehicle may be used to control the host vehicle to perform the action.

[0246] In some embodiments, pairs of actions may be selected as decisions based on cost values ​​for multiple strategy feasibility regions of game outcomes.

[0247] In some embodiments, for multiple solutions (i.e., behavior-action pairs) in the strategy feasible region of the game outcome, successive multi-frame derivations may be further performed for each solution, i.e., the temporal sampling dimension is released, and a behavior-action pair with relatively good temporal consistency is selected as the driving decision result of the host vehicle. For details, see the description of FIG. 7.

[0248] This application further provides corresponding embodiments of intelligent driving decision-making devices, as shown in Figure 12. For the advantageous effects of the devices or the technical problems to be solved by the devices, please refer to the description in the method corresponding to each device or refer to the summary description. The details will not be described again here.

[0249] In one embodiment of the intelligent driving decision-making apparatus, the intelligent driving decision-making apparatus 100 includes an acquisition module 110 configured to acquire game objects of the host vehicle. Specifically, the acquisition module 110 is configured to perform step S10, or step S110 and step S120, or each optional embodiment corresponding to these steps.

[0250] The intelligent driving decision-making device 100 includes a processing module 120 configured to perform a plurality of strategic space releases from a plurality of strategic spaces for the host vehicle and the game object, determine a strategy feasible region for the host vehicle and the game object based on each released strategic space after performing one of the plurality of releases, and determine a driving decision result for the host vehicle based on the strategy feasible region. Specifically, the processing module 120 is configured to perform steps S20 to S40 or each optional embodiment corresponding to steps S20 to S40.

[0251] In some embodiments, the multiple strategy space dimensions include at least one of a vertical sampling dimension, a horizontal sampling dimension, or a temporal sampling dimension.

[0252] In some embodiments, performing multiple freeings of multiple strategy spaces includes performing the freeing in the order of vertical sampling dimension, horizontal sampling dimension, and temporal sampling dimension.

[0253] In some embodiments, when the strategy feasible region of the ego-vehicle and game object is determined, a total cost value of the behavior-action pair in the strategy feasible region is determined based on one or more of the ego-vehicle or game object's safety cost value, right-of-way cost value, lateral offset cost value, passability cost value, comfort cost value, inter-frame association cost value, and risk area cost value.

[0254] In some embodiments, when the total cost value of a behavior action pair is determined based on two or more cost values, each of the cost values ​​has a different weight.

[0255] In some embodiments, when there are two or more game objects, the driving decision for the host vehicle is based on the respective strategy feasibility regions of the host vehicle and each game object.

[0256] In some embodiments, the acquisition module 110 is further configured to acquire non-game objects for the ego-vehicle. The processing module 120 is further configured to determine a strategy feasible region for the ego-vehicle and the non-game objects, the strategy feasible region for the ego-vehicle and the non-game objects including executable behavioral actions of the ego-vehicle with respect to the non-game objects, and determine a driving decision result for the ego-vehicle based at least on the strategy feasible region for the ego-vehicle and the non-game objects.

[0257] In some embodiments, the processing module 120 is further configured to determine a strategy feasible area of ​​the driving decision result of the host vehicle based on an intersection of the strategy feasible areas of the host vehicle and each of the game objects, or to determine a strategy feasible area of ​​the driving decision result of the host vehicle based on an intersection of the strategy feasible area of ​​the host vehicle and each of the game objects and the strategy feasible area of ​​the host vehicle and each of the non-game objects.

[0258] In some embodiments, the acquisition module 110 is further configured to acquire a non-game object of the ego-vehicle, and the processing module 120 is further configured to constrain a longitudinal sampling strategy space corresponding to the ego-vehicle or constrain a lateral sampling strategy space corresponding to the ego-vehicle based on a motion situation of the non-game object.

[0259] In some embodiments, the acquisition module 110 is further configured to acquire non-game objects of the ego-vehicle game object, and the processing module 120 is further configured to constrain a longitudinal sampling strategy space corresponding to the ego-vehicle game object or constrain a lateral sampling strategy space corresponding to the ego-vehicle game object based on the motion situation of the non-game objects.

[0260] In some embodiments, when the intersection is an empty set, a conservative driving decision is made for the ego vehicle, including taking action to safely stop the ego vehicle or safely slowing the ego vehicle for driving.

[0261] In some embodiments, the game object or non-game object is determined by attention.

[0262] In some embodiments, the processing module 120 is further configured to display, via the human-computer interaction interface, at least one of the driving decision result of the host vehicle, the strategy feasibility area of ​​the decision result, the driving trajectory of the host vehicle corresponding to the driving decision result of the host vehicle, or the driving trajectory of the game object corresponding to the driving decision result of the host vehicle.

[0263] The driving judgment result of the host vehicle may be the judgment result of the current single-frame derivation or may be the judgment result corresponding to each of the executed single-frame derivations. This judgment result may be a behavioral action performed by the host vehicle, a behavioral action performed by a game object, or an intention determination corresponding to the behavioral action performed by the host vehicle, such as Cg or Cy in Table 1, such as cutting in, yielding, or avoiding.

[0264] The strategy feasible region of the determination result may be the strategy feasible region of the current single-frame derivation, or may be the strategy feasible region corresponding to each of the multiple single-frame derivations that have been executed.

[0265] The vehicle's driving trajectory corresponding to the vehicle's driving judgment result may be the vehicle's driving trajectory corresponding to the first single frame derivation in one step of judgment, for example, T1 in Figure 7, or the vehicle's driving trajectory obtained by sequentially connecting multiple single frame derivations performed in one step of judgment, for example, T1, T2 and Tn in Figure 7.

[0266] The driving trajectory of the game object corresponding to the driving judgment result of the vehicle itself may be the driving trajectory of the game object corresponding to the first single frame derivation in one step of judgment, for example, T1 in Figure 7, or it may be the driving trajectory of the game object obtained by sequentially connecting multiple single frame derivations performed in one step of judgment, for example, T1, T2 and Tn in Figure 7.

[0267] As shown in FIG. 13, an embodiment of the present invention further provides a vehicle driving control method, including the following steps:

[0268] S210: Obtain information about obstacles outside the vehicle.

[0269] S220: Based on the obstacle information, determine the vehicle driving judgment result according to any one of the above-mentioned intelligent driving judgment methods.

[0270] S230: Based on the result of the determination, the vehicle's running is controlled.

[0271] As shown in FIG. 14 , an embodiment of the present application further provides a vehicle driving control device 200, including: an acquisition module 210 configured to acquire an obstacle outside the vehicle; and a processing module 220 configured to determine a driving judgment result of the vehicle according to any one of the above-mentioned intelligent driving judgment methods for the obstacle, and the processing module is further configured to control the driving of the vehicle based on the judgment result.

[0272] 15, an embodiment of the present application further provides a vehicle 300 including the above-described vehicle driving control device 200 and a driving system 250. The vehicle driving control device 200 controls the driving system 250. In some embodiments, the driving system 250 may include the driving system 13 of FIG.

[0273] 16 is a schematic diagram of the structure of a computing device 400 according to one embodiment of the present application. The computing device 400 includes a processor 410 and a memory 420, and may further include a communication interface 430.

[0274] It should be understood that the communication interface 430 in the computing device 400 shown in FIG. 16 may be configured to communicate with another device.

[0275] The processor 410 may be connected to a memory 420. The memory 420 may be configured to store program code and data. Thus, the memory 420 may be a storage unit within the processor 410, an external storage unit independent of the processor 410, or a component including a storage unit within the processor 410 and an external storage unit independent of the processor 410.

[0276] Optionally, the computing device 400 may further include a bus. The memory 420 and the communication interface 430 may be connected to the processor 410 via a bus. The bus may be a Peripheral Component Interconnect (PCI) standard bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be classified into an address bus, a data bus, a control bus, etc.

[0277] It should be understood that in this embodiment of the present application, the processor 410 may be a central processing unit (CPU). Alternatively, the processor may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or another programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor, etc. Alternatively, the processor 410 is configured to execute associated programs by using one or more integrated circuits to implement the technical solutions provided in the embodiments of the present application.

[0278] The memory 420 may include read-only memory and random access memory to provide instructions and data to the processor 410. A portion of the processor 410 may further include non-volatile random access memory. For example, the processor 410 may further store device type information.

[0279] When the computing device 400 is running, the processor 410 executes the computer-executable instructions in the memory 420 to perform the operational steps of the methods described above.

[0280] The computing device 400 according to this embodiment of the present application may correspond to an entity that performs the method according to this embodiment, and the above-mentioned and other operations and / or functions of the modules in the computing device 400 are intended to be separately implemented to implement corresponding steps of the method in this embodiment. For simplicity, the details will not be described again here.

[0281] Those skilled in the art will recognize with reference to the embodiments disclosed herein that the units and algorithm steps in the described examples can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether a function is performed by hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to go beyond the scope of this application.

[0282] For convenient and simple description, it will be obviously understood by those skilled in the art that the detailed working processes of the aforementioned systems, devices, and units may refer to the corresponding processes of the aforementioned method embodiments, and the details will not be described again here.

[0283] In some embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the described device embodiments are merely examples. For example, the division into units is merely a logical functional division, and other divisions may be used in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented. Additionally, shown or discussed mutual couplings, direct couplings, or communication connections may be implemented via some interfaces. Indirect couplings or communication connections between devices or units may be implemented in electronic, mechanical, or other forms.

[0284] The units described as separate parts may or may not be physically separate, and the parts shown as units may or may not be physical units, may be located in one location, or may be distributed across multiple network units. Some or all of the units may be selected based on actual requirements to achieve the objectives of the solutions of the embodiments.

[0285] Additionally, the functional units in the embodiments of the present application may be integrated into one processing unit, and each of the units may exist physically alone, or two or more units may be integrated into one unit.

[0286] When a function is implemented in the form of a software functional unit and sold or used as an independent product, the function may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be essentially implemented in the form of a software product, a portion contributing to the prior art may be implemented in the form of a software product, or a portion of the technical solution may be implemented in the form of a software product. A computer software product is stored in a storage medium and includes a plurality of instructions for instructing a computer device (such as a personal computer, a server, or a network device) to perform all or part of the steps of the method described in the embodiments of this application. The storage medium includes any medium capable of storing program code, such as a USB flash drive, a removable hard disk, a read-only memory (ROM), a random-access memory (RAM), a magnetic disk, or an optical disk.

[0287] An embodiment of this application further provides a computer-readable storage medium, which stores a computer program, which, when executed by a processor, is used to perform the aforementioned method, which includes at least one of the solutions described in the aforementioned embodiment.

[0288] A computer storage medium according to this embodiment of this application may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable medium may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exclusive list) of computer-readable storage media include an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this specification, a computer-readable storage medium may be any tangible medium that contains or stores a program for use by or in combination with an instruction execution system, apparatus, or device.

[0289] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, in which computer-readable program code is carried. Such propagated data signals may be in various forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may alternatively be any computer-readable medium other than a computer-readable signal medium. A computer-readable medium may transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0290] The program code contained in the computer readable medium may be transmitted using any suitable medium, including, but not limited to, Wi-Fi, wire, optical cable, RF, etc., or any suitable combination thereof.

[0291] Computer program code for carrying out the operations of this application may be written in one or more programming languages, or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as traditional procedural programming languages ​​such as "C" and similar programming languages. The program code may run entirely on the user computer, some may run on the user computer, some may run as a separate software package, some may run on the user computer and some on a remote computer, or the code may run entirely on a remote computer or server. When remote computers are involved, they may be connected to the user computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet Service Provider).

[0292] In this specification and claims, the terms "first, second, third, etc." or similar terms such as module A, module B, and module C are used merely to distinguish between similar objects and do not represent a particular order of the objects. It will be understood that the specific order or sequence may be interchanged, where possible, such that the embodiments of this application described herein may be implemented in orders other than those illustrated or described herein.

[0293] It should be noted that in the above description, the step numbers such as S116, S124, etc. do not necessarily indicate that the steps are executed in sequence. If possible, the sequence of the steps may be interchanged, or the steps may be executed simultaneously.

[0294] The terms "comprises" and "comprises" used in this specification and claims should not be construed as being limited to the contents listed below and do not exclude other elements or steps. They should be construed as specifying the presence of a referenced feature, whole, step, or part, but not excluding the presence or addition of one or more other features, wholes, steps, or parts, and combinations thereof. Thus, the phrase "a device including apparatus A and apparatus B" should not be limited to a device including only component A and component B.

[0295] The term "one embodiment" or "an embodiment" used in this specification indicates that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of this application. Therefore, the terms "in one embodiment" or "in an embodiment" appearing in this specification do not necessarily refer to the same embodiment, but may refer to the same embodiment. Furthermore, in one or more embodiments, particular features, structures, or characteristics may be combined in any suitable manner, as would be apparent to one skilled in the art from this disclosure. It should be noted that the above are merely exemplary embodiments and technical principles of this application. Those skilled in the art may understand that this application is not limited to the specific embodiments described in this specification, and may make various obvious changes, rearrangements, and substitutions without departing from the scope of protection of this application. Therefore, although this application is described in detail with reference to the above embodiments, this application is not limited to the above embodiments. Many other similar embodiments may be included without departing from the concept of this application, and all fall within the scope of protection of this application.

Claims

1. An intelligent driving decision method executed by a vehicle, comprising: Obtaining a game object of the host vehicle, wherein an obstacle is the game object if its travel trajectory or travel intention conflicts with the travel trajectory or travel intention of the host vehicle, and the obstacle has a behavior determination ability and can change the motion situation of the obstacle; performing a plurality of releases of a plurality of strategy spaces from a plurality of strategy spaces, each of which represents a set of behavior action pairs of the ego vehicle and the game object, wherein each of the releases enables the ego vehicle to additionally obtain sampling values ​​of the ego vehicle and sampling values ​​of the game object in a corresponding one of a longitudinal sampling dimension, a lateral sampling dimension, and a temporal sampling dimension; After performing one of the multiple releases, based on each released strategic space, determine a strategy feasible area of ​​the host vehicle and the game object, which indicates a set of executable behavior action pairs of the host vehicle and the game object, and determine a driving decision result of the host vehicle based on the strategy feasible area; When the strategy feasibility region determined after performing a first of the plurality of releases is null, a second of the plurality of releases is performed.

2. The method of claim 1 , wherein the multiple releases of the multiple strategy spaces are performed in the order of the vertical sampling dimension, the horizontal sampling dimension, and the temporal sampling dimension.

3. When the strategy feasible region of the host vehicle and the game object is determined, the total cost value of each behavior action pair is A method according to any one of claims 1 to 2, wherein the cost value is determined based on one or more of a safety cost value, a right of way cost value, a lateral offset cost value, a passability cost value, a comfort cost value, an inter-frame association cost value, and a risk area cost value of the vehicle or the game object.

4. The method of claim 3 , wherein when the total cost value of the behavior-action pair is determined based on two or more cost values, each of the cost values ​​has a different weight.

5. The method according to claim 1 , wherein when there are two or more game objects, the driving decision result of the host vehicle is determined based on the strategy feasible areas of the host vehicle and each of the game objects.

6. obtaining a non-game object of the host vehicle; determining a strategy feasibility region for the host vehicle and the non-game object; The method of any one of claims 1 to 5, further comprising: determining the driving decision result of the host vehicle based at least on the strategy feasibility area of ​​the host vehicle and the non-game object.

7. The strategy feasible area of ​​the driving determination result of the host vehicle is determined based on an intersection of the strategy feasible areas of the host vehicle and each game object; or The method according to claim 5 or 6, wherein the strategy feasible area of ​​the driving judgment result of the host vehicle is determined based on the intersection of the strategy feasible areas of the host vehicle and each game object and the strategy feasible areas of the host vehicle and each non-game object.

8. obtaining a non-game object of the host vehicle; The method according to any one of claims 2 to 7, further comprising: constraining a range of values ​​of the vertical sampling dimension corresponding to the ego-vehicle or constraining a range of values ​​of the horizontal sampling dimension corresponding to the ego-vehicle based on a motion situation of the non-game object.

9. obtaining a non-game object of the game object of the host vehicle; The method according to any one of claims 1 to 7, further comprising: constraining a range of values ​​of the vertical sampling dimension corresponding to the game object of the ego-vehicle or constraining a range of values ​​of the horizontal sampling dimension corresponding to the game object of the ego-vehicle based on a motion situation of the non-game object.

10. The method of claim 7, wherein when the intersection is an empty set, a conservative driving decision of the host vehicle is performed, and the conservative driving decision includes an action to safely stop the host vehicle or an action to safely slow down the host vehicle for driving.

11. A method as described in claim 1, wherein an attention value of the obstacle is determined to determine whether the obstacle is the game object or a non-game object, and the attention value is related to the degree of intention conflict or trajectory conflict that exists between the obstacle and the vehicle.

12. Through a human-computer dialogue interface, The method of any one of claims 1 to 11, further comprising displaying at least one of the driving judgment result of the host vehicle, the strategy feasible area of ​​the driving judgment result, the driving trajectory of the host vehicle corresponding to the driving judgment result of the host vehicle, or the driving trajectory of the game object corresponding to the driving judgment result of the host vehicle.

13. An intelligent driving decision device, an acquisition module configured to acquire a game object of a host vehicle, the acquisition module being configured to acquire a game object of a host vehicle, the game object being an obstacle if its travel trajectory or travel intention conflicts with the travel trajectory or travel intention of the host vehicle, and the obstacle has a behavior determination ability and can change the motion situation of the obstacle; a processing module configured to perform a plurality of releases of a plurality of strategy spaces from a plurality of strategy spaces, each of the strategy spaces representing a set of behavior action pairs of the ego vehicle and the game object, wherein each of the plurality of releases enables the ego vehicle to obtain additional sampling values ​​of the ego vehicle and sampling values ​​of the game object in a corresponding one of a longitudinal sampling dimension, a lateral sampling dimension, and a temporal sampling dimension; a processing module configured to determine, after performing one of the plurality of releases, a strategy feasible region of the host vehicle and the game object, which indicates a set of executable behavior action pairs of the host vehicle and the game object, based on each released strategy space, and to determine a driving decision result of the host vehicle based on the strategy feasible region; When the strategy feasibility region determined after performing a first of the plurality of releases is null, a second of the plurality of releases is performed.

14. A vehicle driving control method, comprising: acquiring obstacles external to the vehicle; determining a travel decision result of the vehicle for the obstacle using the method according to any one of claims 1 to 12; and controlling driving of the vehicle based on the driving determination result.

Images