Intelligent driving decision method, decision device and vehicle

Abstract

The present application discloses an intelligent driving decision-making method, a decision-making device, and a vehicle, which can be applied in the field of intelligent driving, including: identifying an object with uncertain intention (i.e., a game object) and its n road topologies, and determining the intention probability of the game object traveling along each road topology (a total of n), constructing a sampling game space including m game strategies, calculating the strategy cost of the m game strategies in each road topology (a total of n groups), determining n target strategy costs that meet preset conditions, and determining the decision result of the self-vehicle according to the target game strategy (one of the m). In the case of an object with uncertain intention, the present application uses the observed information to infer the intention probability of the object with unclear intention, and based on this, makes an uncertain interactive game decision to solve the best action of the self-vehicle, so that the self-vehicle can reasonably respond to the jump of the intention of the dynamic object, reduce unreasonable braking and mistaken rushing, and improve the riding comfort, safety, and passability of the self-vehicle.

Description

Technical Field

[0001] The present application relates to the field of intelligent driving, and in particular to an intelligent driving decision-making method, a decision-making device and a vehicle. Background Art

[0002] With the advent of intelligent driving, intelligent driving of vehicles has become a hot research direction. Intelligent driving includes assisted driving and automatic driving, and its key technologies include: mapping and positioning, environmental perception, fusion prediction, decision planning and underlying control.

[0003] At the traditional decision-making level, various manufacturers have also proposed decision-making planning methods for different scenarios, which are mainly divided into high-level semantic decisions (such as lane change decisions / lane keeping decisions, etc.) and obstacle decisions for objects (such as avoiding / following, rushing / giving way, etc.). Among them, the object decision module labels the behavior of obstacles and provides reasonable input for the motion planning module to ensure that the autonomous driving vehicle responds reasonably to the surrounding obstacles. The industry's interactive decision-making is based on the optimal game solution of the more dangerous intentions / behaviors of environmental objects. Although it can strictly guarantee safety, it is because it is an interactive game reasoning for a certain intention that is assumed to be unchanged.

[0004] The above decision-making method will bring two serious problems: First, because the existing decision-making system generally uses interactive game to solve the more dangerous deterministic single intention, only considering the worst intention of the object, the decision results are often conservative and non-human, resulting in a decrease in the vehicle's ability to pass in complex environments; second, because only one intention of the object is considered, when the actual behavior of the object changes (that is, the sudden change of the intention of the other vehicle) and is inconsistent with the set intention, the decision input undergoes a step-by-step discontinuous change, which will lead to the vehicle's decision-making being unrobust and unstable, such as erroneous deceleration or erroneous avoidance, which deteriorates the riding experience. Summary of the invention

[0005] The embodiments of the present application provide an intelligent driving decision-making method, a decision-making device, and a vehicle, which are used to use observed information to infer the intention probability of the dynamic object with unclear intention (i.e., the game object) when the intention of the environmental dynamic object is uncertain / unclear (i.e., the game object), and perform uncertain interactive game decisions based on this intention probability to solve the best action of the vehicle (i.e., determine the decision result of the vehicle based on the target game strategy), so that the autonomous driving vehicle can reasonably respond to the jump in the intention of the dynamic object, thereby reducing unreasonable braking and accidental overtaking, and improving the riding comfort, safety, and passability of the autonomous driving vehicle.

[0006] Based on this, the embodiments of the present application provide the following technical solutions:

[0007] In the first aspect, the embodiment of the present application first provides an intelligent driving decision-making method, which can be used in the field of intelligent driving. The method includes: first, the self-vehicle can determine the game object based on the collected perception data and the topological information obtained from the upstream module, and obtain the real-time state information of the game object (such as one or more of the real-time position information, real-time heading angle, real-time speed, real-time acceleration, etc.), and the game object is an obstacle with n (n≥2) road topologies. After that, the self-vehicle can determine the intention probability of the game object traveling along each road topology according to the real-time state information of the game object, thereby obtaining n intention probabilities. After obtaining the real-time state information of the game object, the self-vehicle can further obtain the real-time state information of the self-vehicle (such as the vehicle information of the self-vehicle collected by the self-vehicle sensor system, etc.), and construct a sampling game space for the game object according to the real-time state information of the self-vehicle and the real-time state information of the game object (and additional road condition information), and the sampling game space includes m (m≥1) game strategies, and the game strategy is an action pair composed of each actionable action of the self-vehicle and the actionable action of the game object. After constructing the sampling game space including m game strategies, for each combination of the lateral and longitudinal actions of the ego vehicle and the lateral and longitudinal actions of the game object, the strategy cost is evaluated under each road topology, thereby obtaining n sets of strategy costs, wherein each set of strategy costs includes m strategy costs corresponding to the m game strategies one by one. Finally, according to the n intention probabilities and n sets of strategy costs, n target strategy costs whose total costs meet the preset conditions are determined, and the decision result of the ego vehicle is determined according to the target game strategy. The n target strategy costs are the strategy costs of the target game strategy under each road topology, and the target game strategy is one of the above m game strategies. It should be noted here that when n=1, the game object described above is an obstacle with clear intention. At this time, the obstacle has only one road topology (i.e., n=1). At this time, the calculated intention probability of the obstacle traveling along this unique road topology is determined, which is 100%. The subsequent game solving and decision results of downstream tasks based on the intention probability are similar and will not be repeated here.

[0008] In the above-mentioned embodiment of the present application, when the intention of the dynamic object in the environment is uncertain / unclear (i.e., the game object), the observed information (i.e., the real-time status information of the game object) can be used to quickly and efficiently identify the probabilities of the various intentions of the game object, and make stable decisions based on this intention probability to solve the best action of the vehicle (i.e., determine the decision result of the vehicle based on the target game strategy), so that the autonomous driving vehicle can reasonably respond to the jump in the intention of the dynamic object, thereby reducing unreasonable braking and accidental overtaking, and ensuring the comfort, safety and passability of the ride.

[0009] In a possible implementation of the first aspect, according to the real-time state information of the game object, one implementation method for determining the probability of the game object's intention to travel along each road topology may be: first, the self-vehicle obtains the characteristic reference information of each road topology owned by the game object, and the characteristic reference information is used to describe the baseline state of the game object's intention to travel along each road topology under the current traffic environment state. Afterwards, the self-vehicle determines the probability of the game object's intention to travel along each road topology based on the real-time state information and the characteristic reference information of the game object.

[0010] In the above-mentioned embodiment of the present application, it is specifically explained that the probability of each intention of the game object can be determined based on the characteristic reference information of the game object in each road topology. The characteristic reference information is easy to obtain, convenient to calculate, and operational.

[0011] In a possible implementation of the first aspect, a method for determining the probability of the game object's intention to travel along each road topology based on the real-time state information and feature reference information of the game object can be: first, a target feature vector is obtained based on the real-time state information of the game object and the feature reference information calculated by the above method. Then, the likelihood probability of the target feature vector is determined based on a probability density distribution data set constructed in advance, wherein the probability density distribution data set includes the distribution of the probability density of the feature vector under different feature vector values. Finally, the probability of the game object's intention to travel along each road topology is determined based on the likelihood probability.

[0012] In the above-mentioned implementation mode of the present application, it is specifically explained that the likelihood probability of the target feature vector can be obtained first, and then the intention probability can be obtained based on the likelihood probability, thereby effectively identifying the intention of the game object, which is simple and effective.

[0013] In a possible implementation of the first aspect, the probability density distribution data set includes: a nonlinear probability density distribution data set, wherein the nonlinear probability density distribution data set includes the distribution of probability density of feature vectors under different speed values ​​and different feature vector values.

[0014] In the above implementation of the present application, it is taken into consideration that the feature vector and the likelihood probability distribution should have different distribution forms at different speeds of the game object, which is more in line with the actual application scenario and improves the accuracy of calculating the intention probability.

[0015] In a possible implementation of the first aspect, an implementation method for determining the intention probability of the game object to travel along each road topology based on the likelihood probability may be: first, obtaining the prior intention probability of the obstacle traveling along each road topology, and then, based on the likelihood probability and the prior intention probability, using a Bayesian inference algorithm, calculating the intention probability of the game object to travel along each road topology.

[0016] In the above-mentioned implementation manner of the present application, the introduction of a priori intention probability can further improve the accuracy of calculating the intention probability.

[0017] In a possible implementation of the first aspect, the feature reference information is obtained based on the target waypoints on each road topology; or, the feature reference information is obtained based on the road congestion level and / or the aggressiveness of the driving style of the game object. It should be noted that in the embodiment of the present application, the feature reference information of the game object on each road topology can be calculated directly based on the road congestion level and / or the aggressiveness of the driving style of the game object; or the initial feature reference information can be first obtained based on the target waypoints on each of the road topologies, and then further optimized and adjusted based on this based on the road congestion level and / or the aggressiveness of the driving style of the game object, so that the optimized feature reference information more reasonably represents the characteristics of each intention under the current traffic environment state, and is more in line with the actual situation.

[0018] In the above-mentioned embodiments of the present application, several implementation methods for obtaining feature reference information are specifically described, which are flexible.

[0019] In a possible implementation of the first aspect, the real-time status information of the game object includes at least any one or more of the following: the real-time position information of the game object, the real-time orientation angle of the game object, the real-time speed of the game object, and the real-time acceleration of the game object; the characteristic reference information includes at least any one or more of the following: the reference position information of the game object, the reference orientation angle of the game object, the reference speed of the game object, and the reference acceleration of the game object.

[0020] In the above-mentioned embodiments of the present application, several possible types of real-time status information and feature reference information of gaming objects are specifically described, which are comprehensive.

[0021] In a possible implementation of the first aspect, when there are no other obstacles around the game object except the vehicle itself, the reference speed of the game object is calculated based on the kinematic constraints of the game object and / or the environmental traffic constraints of the game object.

[0022] In the above implementation of the present application, it is specifically described that when there are no other obstacles, the reference speed of the game object is subject to the following constraints: kinematic constraints and environmental traffic constraints. This is in line with the actual application scenario and is feasible.

[0023] In a possible implementation of the first aspect, the kinematic constraints include at least one or more of the following: lane topology curvature speed limit, start acceleration constraint, and red light deceleration constraint. The environmental traffic constraints include at least one or more of the following: road speed limit of the lane where the game object is located, speed constraint of the preceding vehicle in the lane where the game object is located, and speed constraint of obstacles in lanes other than the lane where the game object is located.

[0024] In the above-mentioned embodiments of the present application, several forms of existence of kinematic constraints and environmental traffic constraints are specifically described, which are extensive.

[0025] In a possible implementation of the first aspect, on the n road topologies owned by the game object, if there is at least one obstacle located in front of the game object (which may be referred to as the first target obstacle, which may be one or more), it is necessary to first determine whether the distance between the first target obstacle and the game object is at a safe distance. If it is at a safe distance, the reference speed of the game object is obtained based on the speed of the one or more first target obstacles.

[0026] In the above implementation of the present application, it is specifically explained that when there are other obstacles, if the obstacles are on the road topology of the game object, the reference speed of the game object is constrained as follows: when the distance to the front vehicle is less than the safe distance, it is constrained by the speed of the front vehicle. This is in line with the actual application scenario and is feasible.

[0027] In a possible implementation of the first aspect, the game object has n road topologies. If there is at least one obstacle (which may be called a second target obstacle, which may be one or more) in the same driving direction as the game object, and the road topology owned by the at least one second target obstacle does not belong to the above n road topologies, the vehicle can first obtain a reference speed of each second target obstacle according to a preset method, and then obtain a reference speed of the game object based on each reference speed.

[0028] In the above implementation of the present application, it is specifically described that when there are other obstacles, if the obstacle and the game object are traveling in the same direction and parallel, the reference speed of the game object is constrained by: the reference speed of the obstacle traveling in parallel. This is in line with the actual application scenario and is feasible.

[0029] In a possible implementation of the first aspect, the game object has n road topologies. If there is at least one obstacle (which may be called a third target obstacle, which may be one or more) in the cross-traveling direction of the game object, the vehicle may first determine whether the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than a preset threshold. If the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than the preset threshold, the reference speed of the game object may be the real-time speed of the game object. It should be noted that the real-time speed of the game object here does not refer to the speed of the game object observed by the vehicle at the beginning, but refers to the actual speed value of the vehicle when the game object is traveling at each current moment.

[0030] In the above implementation of the present application, it is specifically explained that when there are other obstacles, if the obstacle is a crossing vehicle of the game object, the reference speed of the game object is constrained as follows: if the time difference between the two reaching the predicted collision point is greater than the preset threshold, it means that it is relatively safe and the game object can maintain the current speed. This is in line with the actual application scenario and is feasible.

[0031] In a possible implementation of the first aspect, the game object has n road topologies. If there is at least one obstacle (which may be called a third target obstacle, which may be one or more) in the direction of the game object's crossing, the vehicle may first determine whether the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than a preset threshold. If the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is less than or equal to the preset threshold, the reference speed of the game object is obtained based on the real-time speed of the game object, and the reference speed needs to be less than the real-time speed of the game object. Similarly, the real-time speed of the game object here does not refer to the speed of the game object observed by the vehicle at the beginning, but refers to the actual speed value of the vehicle when the game object is traveling at each current moment.

[0032] In the above implementation of the present application, it is specifically explained that when there are other obstacles, if the obstacle is a crossing vehicle of the game object, the reference speed of the game object is constrained as follows: if the time difference between the two reaching the predicted collision point is ≤ a preset threshold, the game object should slow down and give way. This is in line with actual application scenarios and is feasible.

[0033] In a possible implementation of the first aspect, an implementation method for determining n target strategy costs whose total cost meets preset conditions based on n intention probabilities and n groups of strategy costs may be: taking the n intention probabilities as the weights of the n road topologies respectively, calculating the weighted strategy costs corresponding to each game strategy in the n groups of strategy costs, and obtaining m total costs; then, determining the total cost with the smallest value from the m total costs; and finally, further determining n target strategy costs corresponding to the total cost with the smallest value.

[0034] In the above-mentioned implementation mode of the present application, the deterministic interactive game solution is combined with the intention probability to form an uncertain interactive game framework to solve the optimal decision under uncertain intentions, thereby improving the correctness of vehicle decision-making.

[0035] In a possible implementation of the first aspect, an implementation method for calculating the strategy costs of m game strategies under each road topology may be: determining various factors of the strategy costs of the m game strategies under each road topology, such as safety, comfort, passability, road right, lateral deviation, risk area, inter-frame association, etc.; then, calculating the factor cost of each factor in each strategy cost. The method for calculating the factor cost may be to determine the weights of each factor in advance, such as safety weight, comfort weight, passability weight, road right weight, lateral deviation weight, risk area weight, inter-frame association weight, etc. (the values ​​of each weight may be different), and weighting the factor cost of each factor in each strategy cost based on each weight to obtain the strategy costs of the m game strategies under each road topology.

[0036] In the above-mentioned implementation mode of the present application, it is specifically explained that the strategy cost is the weighted sum of the costs of various factors, so that the final strategy cost is more in line with the vehicle driving habits and is conducive to driving safety requirements.

[0037] In a possible implementation of the first aspect, a method for constructing a sampling game space for the game object based on the real-time status information of the vehicle and the real-time status information of the game object may be: first, determining the decision upper limit and decision lower limit of the vehicle and the game object based on the real-time status information of the vehicle and the game object; then, according to preset rules, obtaining the decision strategy of the vehicle and the decision strategy of the game object from the decision upper limit and the decision lower limit; finally, combining the decision strategy of the vehicle with the decision strategy of the game object (also called an action pair or action combination) to obtain m game strategies of the vehicle and the game object, and the m game strategies belong to the sampling game space.

[0038] In the above-mentioned implementation mode of the present application, kinematic deduction is performed on the feasible space of all ego vehicles and game vehicles, and m game strategies are formed according to the future motion interaction situation. The strategic cost of each game strategy can be calculated based on this, so that stable decisions can be made in the change of object intentions, solving the conservative problem of point braking in traditional deterministic decision-making, and ensuring safety, comfort and passability.

[0039] In a possible implementation of the first aspect, an implementation of determining the game object may be: first, the vehicle may first obtain all road topologies within a preset area around the vehicle (such as within a radius of 10m around the vehicle), and then obtain obstacle information of obstacles (such as at least one obstacle) perceived around the vehicle, and calculate the road topology attribution of each obstacle based on the obtained obstacle information and all the road topologies obtained above, and if there is an obstacle with n road topologies, then the obstacle with n road topologies is determined to be the game object. The process of determining the game object may also be referred to as the process of object road topology attribution recognition and intention uncertain object recognition.

[0040] In the above-mentioned implementation mode of the present application, a method for determining the game object is specifically described, which is feasible.

[0041] The second aspect of the embodiment of the present application provides a decision-making device, which has the function of implementing the method of the first aspect or any possible implementation of the first aspect. The function can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions.

[0042] A third aspect of an embodiment of the present application provides a vehicle driving control method, the method comprising: obtaining an obstacle outside the vehicle, and determining a decision result of the vehicle driving according to the method of the first aspect or any possible implementation method of the first aspect for the obstacle, and finally controlling the vehicle driving according to the decision result.

[0043] A fourth aspect of the embodiment of the present application provides a vehicle driving control device, which has the function of implementing the method of the third aspect. The function can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions.

[0044] A fifth aspect of an embodiment of the present application provides a vehicle, comprising the vehicle driving control device and the driving system provided in the fourth aspect, wherein the vehicle driving control device controls the driving system.

[0045] A sixth aspect of an embodiment of the present application provides a computing device, which may include a memory, a processor, and a bus system, wherein the memory is used to store programs, and the processor is used to call the program stored in the memory to execute the method of the first aspect of the embodiment of the present application or any possible implementation of the first aspect, or the processor is used to call the program stored in the memory to execute the method of the third aspect of the embodiment of the present application.

[0046] The seventh aspect of an embodiment of the present application provides an intelligent driving system, which includes a sensing system and a processor, the processor is used to call a program stored in a memory to execute the method of the first aspect of the embodiment of the present application or any possible implementation method of the first aspect, or the processor is used to call the program stored in the memory to execute the method of the third aspect of the embodiment of the present application.

[0047] An eighth aspect of an embodiment of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, the computer can execute the method of the first aspect or any possible implementation of the first aspect, or the computer can execute the method of the third aspect of the embodiment of the present application.

[0048] The ninth aspect of the embodiments of the present application provides a computer program, which, when running on a computer, enables the computer to execute the method of the first aspect or any possible implementation of the first aspect, or enables the computer to execute the method of the third aspect of the embodiments of the present application.

[0049] The tenth aspect of the embodiment of the present application provides a chip, which includes at least one processor and at least one interface circuit, the interface circuit is coupled to the processor, the at least one interface circuit is used to perform a transceiver function, and send instructions to the at least one processor, the at least one processor is used to run a computer program or instruction, which has the function of implementing the method of the first aspect or any possible implementation of the first aspect, or, it has the function of implementing the method of the third aspect, the function can be implemented by hardware, can also be implemented by software, can also be implemented by a combination of hardware and software, the hardware or software includes one or more modules corresponding to the above functions. In addition, the interface circuit is used to communicate with other modules outside the chip, for example, the interface circuit can send the game strategy obtained by the processor on the chip to the downstream control module of the vehicle for motion control (such as avoiding, following, grabbing, giving way, etc.). BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of the system architecture provided for an embodiment of the present application;

[0051] Figure 2 A typical application scenario provided by the embodiment of the present application;

[0052] Figure 3 Another typical application scenario provided by the embodiment of the present application;

[0053] Figure 4 Another typical application scenario provided by the embodiment of the present application;

[0054] Figure 5A schematic diagram of a flow chart of an intelligent driving decision-making method provided in an embodiment of the present application;

[0055] Figure 6 An example diagram of determining a game object provided in an embodiment of the present application;

[0056] Figure 7 Another example diagram of determining a game object provided in an embodiment of the present application;

[0057] Figure 8 A schematic diagram of the relationship between the target feature vector and the likelihood probability distribution in the straight and left turn scenarios provided in the embodiment of the present application;

[0058] Fig. 9 A schematic diagram of the relationship between the target feature vector and the likelihood probability distribution when the speed is different in the straight and left turn scenarios provided in the embodiment of the present application;

[0059] Fig.10 A schematic diagram of the relationship between the speed of a game object and the probability density distribution of the position of the straight-going intention provided in an embodiment of the present application;

[0060] Fig.11 A schematic diagram of the relationship between the speed of a game object and the probability density distribution of the straight-moving intention direction angle provided in an embodiment of the present application;

[0061] Fig.12 A schematic diagram of the probability density distribution relationship between the speed of the game object and the straight-going intention speed provided in the embodiment of the present application;

[0062] Fig.13 A schematic diagram of the lateral feasible space sampling of the vehicle and the game object provided in the embodiment of the present application;

[0063] Fig.14 A schematic diagram of performing lateral position offset sampling and longitudinal acceleration offset sampling on a self-vehicle and a game object provided in an embodiment of the present application;

[0064] Fig.15 A schematic diagram of a security cost function after homogenization provided in an embodiment of the present application;

[0065] Fig.16 A schematic diagram of a comfort cost function after homogenization provided in an embodiment of the present application;

[0066] Fig.17 A schematic diagram of a passability cost function after the homogenization process provided in an embodiment of the present application;

[0067] Fig.18 A schematic diagram of a road right cost function after the homogenization process provided in an embodiment of the present application;

[0068] Fig.19 A schematic diagram of a lateral offset cost function after the normalization process provided in an embodiment of the present application;

[0069] Fig. 20 A schematic diagram of the inter-frame correlation cost function after the normalization process provided in the embodiment of the present application;

[0070] Fig.21 A schematic diagram of a feasible domain of strategies for traversing a game object provided in an embodiment of the present application;

[0071] Fig. 22 Another schematic diagram of performing lateral position offset sampling and longitudinal acceleration offset sampling on the vehicle and the game object provided in an embodiment of the present application;

[0072] Fig.23 A schematic diagram of a core process architecture of an intelligent driving decision-making method provided in an embodiment of the present application;

[0073] Fig.24 A schematic diagram of the structure of a decision-making device provided in an embodiment of the present application;

[0074] Fig.25 A schematic flow chart of a vehicle driving control method provided in an embodiment of the present application;

[0075] Fig.26 A schematic diagram of the structure of a vehicle driving control device provided in an embodiment of the present application;

[0076] Fig. 27 A structural schematic diagram of a vehicle provided in an embodiment of the present application;

[0077] Fig.28 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0078] The embodiments of the present application provide an intelligent driving decision-making method, a decision-making device, and a vehicle, which are used to use observed information to infer the intention probability of the dynamic object with unclear intention (i.e., the game object) when the intention of the environmental dynamic object is uncertain / unclear (i.e., the game object), and perform uncertain interactive game decisions based on this intention probability to solve the best action of the vehicle (i.e., determine the decision result of the vehicle based on the target game strategy), so that the autonomous driving vehicle can reasonably respond to the jump in the intention of the dynamic object, thereby reducing unreasonable braking and accidental overtaking, and improving the riding comfort, safety, and passability of the autonomous driving vehicle.

[0079] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0080] In order to better understand the scheme of the embodiment of the present application, the following first introduces the relevant terms and concepts that may be involved in the embodiment of the present application. It should be understood that the interpretation of the relevant concepts may be limited due to the specific circumstances of the embodiment of the present application, but it does not mean that the present application is limited to the specific circumstances. There may be differences in the specific circumstances of different embodiments, which are not specifically limited here.

[0081] (1) Object Decision

[0082] It means that during the process of autonomous navigation, the autonomous vehicle needs to make decisions about obstacles in the environment and label them with behavioral labels, such as bypassing / following / overtaking an obstacle. This process is called object decision making.

[0083] (2) Game strategy

[0084] The behaviors that the autonomous driving vehicle and the game object may take during the game. In the embodiment of the present application, the game strategy refers to the k1 sampling actions of the game object and the k2 sampling actions of the autonomous driving vehicle, which constitute k1*k2 game strategies. A game strategy can also be called a "game action pair", and the set containing k1*k2 game strategies can be called a sampling game space.

[0085] (3) Reference path

[0086] The reference benchmark used by autonomous vehicles for object decision making, such as the center line of the current road. This path mainly reflects real map information and can guide the driving direction of the autonomous vehicle, but it has no obstacle avoidance function.

[0087] (4) Horizontal

[0088] Parallel to the reference path direction, it corresponds to the obstacle avoidance, detour and other behaviors of the autonomous driving vehicle.

[0089] (5) Vertical

[0090] Along the reference path direction, it corresponds to the acceleration and deceleration behavior of the autonomous driving vehicle.

[0091] (6) Intention

[0092] Environmental traffic participants will have to make higher-level behavioral decisions in the future, such as going straight, turning left, turning right, and U-turns.

[0093] (7) Road topology

[0094] Road geometry information and the associated relationships of each road, such as left-turn lanes, right-turn lanes, etc.

[0095] (8) Self-driving

[0096] An autonomous driving vehicle, that is, a vehicle in which the method of the embodiment of the present application works.

[0097] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0098] First, the system architecture of this application is described. For details, please refer to Figure 1 , the decision-making device provided in the embodiment of the present application can be subordinate to the object decision module in the automatic driving system, or it can be deployed independently, and the present application does not limit this. For the convenience of explanation, the object decision module subordinate to the automatic driving system is used as an example for explanation. Specifically, the object decision module determines whether the self-vehicle will conflict with the obstacle when traveling along the reference path according to the self-vehicle reference path input by the upper module and the obstacle information obtained by the perception / fusion module and the prediction module. In the case where there is a conflict between the self-vehicle and the obstacle, it is divided into game / non-game objects. For the game object, the intelligent driving decision-making method described in the embodiment of the present application is performed; for the non-game object, a rule-based object decision is performed (this part is not repeated in this application); and then the conflict resolution module is used to resolve the scene where there are conflicts in multiple object decision results. The intelligent driving decision-making method provided in the embodiment of the present application will output object decision results, such as behavior labels: rushing / giving way, avoiding obstacles / following, etc. These object decision results are input into the motion planning module, and the motion planning module will perform corresponding motion planning on it according to the behavior label of the obstacle, and finally realize the self-vehicle's rushing / giving way, avoiding obstacles / following actions on the object.

[0099] It should be noted that the intelligent driving decision-making method provided in the embodiment of the present application is an algorithm that can be run on a vehicle-mounted computer system, and the computing power of the computer system can ensure an update rate of at least 30 Hz for the algorithm.

[0100] It should also be noted that Figure 1The system architecture shown is for illustration only. In actual applications, it may include more or fewer subsystems or modules, and each subsystem or module may include multiple components, which are not specifically limited here.

[0101] The application scenario of the embodiment of the present application can be any scenario where the self-vehicle interacts with social vehicles (i.e. other vehicles). In the embodiment of the present application, social vehicles refer to vehicles with at least two road topology attributes, which are also referred to as game objects in the present application. As an example, Figure 2 A typical application scenario of the present application is illustrated, in which the vehicle goes straight and the game object turns left / goes straight; as another example, Figure 3 Another typical application scenario of the present application is illustrated. In this application scenario, the vehicle turns right and the game object turns left / goes straight. As another example, Figure 4 Another typical application scenario of the present application is illustrated, in which the ego vehicle turns left and the game object turns right / goes straight. In these application scenarios, the ego vehicle needs to identify dynamic objects with unclear intentions (dynamic obstacles with unclear intentions for the ego vehicle refer to dynamic objects with two or more possible driving paths) based on inputs such as obstacle information, motion information, road topology information, environmental information, and traffic rules information (such as road rights), and perform probabilistic reasoning on the intentions of such objects, and make uncertain decisions based on the probability of each intention, and decide whether to rush / give way to the obstacle, avoid obstacles / follow the obstacle.

[0102] It should be understood that the above are only a few specific scenarios applied in the embodiments of the present application. The embodiments of the present application are not limited to the above scenarios when applied. They can be applied to any scenario where the vehicle interacts with the game object, and this application will not elaborate on this.

[0103] In combination with the above description, the intelligent driving decision-making method provided by the embodiment of the present application is described below. For details, please refer to Figure 5 , Figure 5 A flowchart of an intelligent driving decision-making method provided in an embodiment of the present application is provided. The method may include the following steps:

[0104] 501. Determine a game object and obtain real-time status information of the game object. The game object is an obstacle identified by the ego-vehicle and belonging to n road topologies, where n≥2.

[0105] First, the ego vehicle determines the game object based on the collected perception data and the topological information obtained from the upstream module, and obtains the real-time status information of the game object. It should be noted that the real-time status information of the game object is based on the game object as the vehicle coordinate system, such as the real-time position information, real-time heading angle, real-time speed, real-time acceleration, etc. of the game object. Among them, the real-time position information of the game object refers to the position of the center of mass of the game object, which can be recorded as (x0, y0), and the real-time heading angle of the game object refers to the heading angle of the front of the game object. In addition, the game object refers to an obstacle with n (n≥2) road topology attributes.

[0106] It should be noted that in some embodiments of the present application, the method of determining the game object includes but is not limited to: first, the vehicle can first obtain all road topologies within a preset area around the vehicle (such as within a radius of 10m around the vehicle), and then obtain obstacle information of obstacles (such as at least one obstacle) perceived around the vehicle, and calculate the road topology attribution of each obstacle based on the obtained obstacle information and all the road topologies obtained above, and if there is an obstacle with n road topologies, then the obstacle with n road topologies is determined to be the game object. The process of determining the game object can also be called the process of object road topology attribution recognition and intention uncertain object recognition.

[0107] To facilitate understanding of the above process, the following is an example to illustrate the process of determining the above game object. Figure 6 , Figure 6 The illustrated scenario is a traffic scenario where the vehicle is moving straight and an obstacle with uncertain intention may turn left or move straight. There are no other dynamic obstacles around the obstacle with uncertain intention.

[0108] like Figure 6 As shown, first, the vehicle can obtain all road topology information within a preset area around the vehicle through the upstream module (such as the navigation module), for example, all road topology information within 20m around the intersection where the vehicle is located, such as the left turn r (1) , go straight in the opposite direction (2) and r (3) , turn right (4)(Indicated by the dotted line with an arrow). By comparing with the reference trajectory information of the own vehicle (indicated by the straight line in front of the own vehicle), the oncoming multi-intention lanes can be screened out. The so-called multi-intention lanes refer to lanes where social vehicles that may have a risk of collision with the own vehicle are located. The present application is to screen out adjacent lanes to the lanes where the own vehicle is located and lanes where the driving trajectory may intersect. Specifically, the road topology that is nearly parallel to the reference trajectory of the own vehicle can be identified first (for some other special intersections, the angle between the road topology and the reference trajectory of the own vehicle can also be considered to be a parallel relationship within a preset range, such as an angle between 150°-180°, which is also considered to be parallel), and r can be obtained. (2) and r (3) , that is, the straight lane topology. At the same time, the left turn lane r can be screened out by the road topology information (1) , through r (1) and r (2) By comparing the waypoints on each topology, we can see that r (1) and r (2) The straight lines before entering the intersection overlap with each other, so it can be considered that r (1) and r (2) It is a multi-purpose lane.

[0109] After that, when the social car is first recognized by the perception module of the ego car, the autonomous driving system of the ego car can obtain the real-time status information such as the position and size of the social car. The position and size of the social car and the waypoint information of the multi-intention lane are input into the collision detection algorithm to obtain the lane affiliation of the social car, that is, in this embodiment, the social car is in the multi-intention lane r (1) and r (2) From this, it can be considered that the social car is a fuzzy and uncertain obstacle, that is, the game object, and its road topology belongs to: multi-intention lane r (1) and r (2) , in the future it may be r (1) Turn left for the reference path, or you may turn left with r (2) Go straight for the reference path.

[0110] It should be noted that Figure 6 The figure shows the process of determining a game object. In the actual application process, each obstacle identified by the vehicle perception module can be judged as a game object based on the above method. If the identified obstacle has n (n≥2) road topology attributes, it is determined to be a game object. Figure 6Only one of them is used as an example for explanation. It should also be noted that when n=1, the game object described above is an obstacle with clear intention. At this time, the obstacle has only one road topology (i.e. n=1). At this time, the calculated probability of the obstacle's intention to travel along this unique road topology is certain, which is 100%. The subsequent game solving and decision-making processes for downstream tasks based on this intention probability are similar and will not be repeated here.

[0111] It should be noted that in the embodiments of the present application, the process of determining the game object is not limited to the above methods, but also includes but is not limited to: 1) a data-driven method, for example, a method of determining the intention of a social car based on an attention model is to input the speed, position, topology and historical information of the social car into the model, output the attention distribution of the social car, and determine its intention based on the attention distribution; 2) an expert experience method, for example, by querying the relationship data between the stored speed, position, topology and other data and the intention. Specifically, this application does not limit the implementation method of the process of determining the game object.

[0112] 502. According to the real-time status information of the game object, determine the intention probability of the game object traveling along each road topology, and obtain n intention probabilities.

[0113] Afterwards, the ego vehicle can determine the intention probability of the game object to travel along each road topology based on the real-time status information of the game object, thereby obtaining n intention probabilities.

[0114] Specifically, first, the vehicle obtains the topology of each road owned by the game object (such as Figure 6 The left turn topology shown (1) and straight topology (2)) feature reference information, which is used to describe the baseline state of the game object's intention to travel along each road topology under the current traffic environment. Afterwards, the vehicle determines the probability of the game object's intention to travel along each road topology based on the game object's real-time state information and feature reference information. Similarly, the feature reference information of the game object on each road topology is also based on the game object as the car coordinate system, such as one or more of the game object's reference position information, reference orientation angle, reference speed, reference acceleration, etc., wherein the game object's reference position information refers to the position of the target waypoint (i.e., the reference point, such as the waypoint closest to the center of mass position of the game object on the left turn topology of the game object) selected by the game object on each road topology, and the reference orientation angle refers to the tangent angle of the reference point on the corresponding road topology (the reference orientation angle is for the game object, with the game object as the coordinate). Therefore, after selecting the target waypoint of the game object on each road topology, the remaining other feature reference information, such as the reference orientation angle, can be obtained based on the target waypoint. It should be noted that in some other embodiments of the present application, the feature reference information of the game object on each road topology can be calculated directly based on the road congestion level and / or the aggressiveness of the game object's driving style; or initial feature reference information can be first obtained based on the target waypoints on each of the road topologies, and then further optimized and adjusted based on the road congestion level and / or the aggressiveness of the game object's driving style, so that the optimized feature reference information more reasonably represents the characteristics of each intention under the current traffic environment and is more in line with the actual situation.

[0115] For ease of understanding, the following is still based on Figure 6 As an example of the application scenario shown in the figure, an implementation method for determining the characteristic reference information of each road topology owned by the game object is described. Figure 6 In the application scenario shown, it can be seen that the road topologies owned by the game objects are r (1) (corresponding to the intention to turn left) and r (2) (corresponding to the intention to go straight), for this application scenario, it is assumed that the feature reference information includes reference position information, reference orientation angle, and reference speed. (1) Feature reference information reference location (exist Figure 6 Abbreviated as P r1 ), reference orientation angle Reference speed Go straight (2) The characteristic reference information is the reference position (exist Figure 6 Abbreviated as P r2 ), reference orientation angle Reference speed The details are shown in Table 1:

[0116] Table 1. Feature reference information for left turn intention and straight going intention

[0117]

[0118] As an example, the calculation process of feature reference information is as follows: Figure 6 As shown, the center of mass of the game object is (x0, y0). (1) All waypoints on the left turn topology r (1) The nearest waypoint can be found on the (not necessarily the nearest waypoint, can be any waypoint, here is just for illustration) As r (1) Reference position on (i.e. P r1 ). Along the r (1) The orientation angle, as r (1) Reference orientation angle on Similarly, we can obtain the straight topology r (2) Reference position on (i.e. P r2 ), reference orientation angle

[0119] As for the reference speed, it can be determined based on whether there are other obstacles around the game object. The following are explained separately:

[0120] 1. There are no other obstacles around the game object except the vehicle itself

[0121] In this case, the reference speed of the game object is calculated based on the kinematic constraints of the game object and / or the environmental traffic constraints of the game object. As an example, the kinematic constraints of the game object include but are not limited to the following factors:

[0122] (1) Lane topology curvature speed limit.

[0123] Lane topology curvature speed limit K is the road topology r (i) The curvature.

[0124] (2) Starting acceleration constraints.

[0125] Starting acceleration restriction a smax It is the maximum starting acceleration of social vehicles. It should be noted that different vehicle types such as trucks, cars, buses, motorcycles, etc. need to be configured with different values.

[0126] (3) Red light speed reduction restrictions.

[0127] Red light speed reduction d is the distance from the current position to the stop line of the intersection, v is the current speed of the social vehicle, and t is the accumulated time from the detection of the red light, which increases frame by frame.

[0128] In addition, in the embodiment of the present application, the environmental traffic constraints of the game object include but are not limited to the following factors:

[0129] (1) The speed limit of the lane where the game object is located.

[0130] The speed limit of the game object's road It can be obtained through the road topology sent down from the upstream.

[0131] (2) The speed constraint of the vehicle ahead in the lane where the game object is located.

[0132] The speed constraint of the vehicle ahead of the game player in the lane exist Figure 6 In the corresponding example, the social car has no preceding car, so

[0133] (3) Speed ​​constraints on obstacles in lanes other than the lane where the game object is located.

[0134] Other lane dynamic objects In this embodiment, there are no other cars around the social car, so

[0135] It should be noted that, in addition to the above methods, the speed constraint of the vehicle in front of the lane where the game object is located is not limited to data-driven / expert experience, and will not be elaborated here.

[0136] Through the above method, the vehicle can calculate each sub-speed constraint and determine a speed value that meets the preset requirements based on each sub-speed constraint as the reference speed. For example, the minimum value of all the above sub-speed constraints can be taken as r (1) and r (2) The respective reference speed and Specifically, as shown in formula (1) and formula (2):

[0137]

[0138]

[0139] 2. The situation where there are other obstacles around the game object besides the vehicle

[0140] In this case, the vehicle needs to further determine the driving status of other obstacles and determine the range of reference speed accordingly.

[0141] For the sake of convenience, the following is a specific example. Figure 7 , Figure 7 The example shows a situation where there are other obstacles Obj1-Obj6 around the game object besides the ego vehicle. The ego vehicle is driving straight at an unprotected straight-ahead intersection, and the opposite game object has both straight-ahead and left-turn lanes on its road topology. There are dynamic obstacles Obj1 and Obj2 on its left-turn topology path, dynamic obstacles Obj3 and Obj4 on its straight-ahead topology path, and straight-ahead dynamic obstacles Obj5 and crossing dynamic obstacles Obj6 on other lanes. Figure 7 ,Since the environment around the game object opposite to the ego vehicle has more other dynamic traffic participants, this will affect the generation of the reference speed of the game object in each ,intention situation.

[0142] Based on this, in the embodiment of the present application, the environmental traffic participants can be divided into two categories according to the relative position relationship between the environmental dynamic traffic participants and the game object: dynamic obstacles on the same road topology path (such as Figure 7 Obj1~Oj4 in ) and dynamic obstacles on different road topology paths (such as Figure 7 Obj5~Oj6 in ). The following Figure 7 Take the following as an example to explain different situations:

[0143] A. There are other obstacles, and the obstacles are located on the road topology of the game object

[0144] On the n road topologies owned by the game object, if there is at least one obstacle located in front of the game object (which can be called the first target obstacle, which can be one or more), it is necessary to first determine whether the distance between the first target obstacle and the game object is at a safe distance. If it is at a safe distance, the reference speed of the game object is obtained based on the speed of the one or more first target obstacles.

[0145] For ease of understanding, the following Figure 7 As an example, the reference speed of the game object in this case is explained: for each dynamic obstacle on the road topology path owned by the game object, the game object should be able to stop at a reasonable deceleration, maintain a safe distance from the vehicle in front, and avoid rear-end collision. Therefore, in the embodiment of the present application, by traversing each front dynamic object on the same road topology path, the reference speed considering the safety of each front dynamic object can be obtained, and then the minimum value of the reference speed of each front dynamic object is taken as the road topology path r (i) The speed constraint of the preceding vehicle As an example, the calculation method of the safety distance can be shown as formula (3):

[0146]

[0147] Among them, a smax For a reasonable deceleration, it can be preset, for example, [-3,-1.5] m / s 2 v is the current real-time speed of the game object, k is a preset value, which can be customized, for example, k = 15. It should be noted that the real-time speed of the game object here does not refer to the speed of the game object observed by the vehicle at the beginning, but refers to the actual speed value of the vehicle when the game object is driving at each current moment.

[0148] In the embodiment of the present application, for the left-turn road topology r (1) , the distance between Obj1 and Obj2 and the game object is less than d risk , so it needs to be considered. The speed constraint of the preceding vehicle can be expressed as formula (4):

[0149]

[0150] Among them, V toObj1 、V toObj2 are the current real-time speeds of the preceding vehicles Obj1 and Obj2, respectively. It should also be noted that the real-time speeds of the preceding vehicles Obj1 and Obj2 here do not refer to the speeds of the preceding vehicles Obj1 and Obj2 observed by the vehicle at the beginning, but refer to the actual speed values ​​of the preceding vehicles Obj1 and Obj2 at each current moment.

[0151] Similarly, for the straight road topology r (2) , the distances between Obj3 and Obj4 and the game object are both less than d risk , so it needs to be considered. The speed constraint of the preceding vehicle can be expressed as formula (5):

[0152]

[0153] Among them, V toObj3 、V toObj4 They are the current real-time speeds of the front vehicles Obj3 and Obj4 respectively. It should also be noted that the real-time speeds of the front vehicles Obj3 and Obj4 here do not refer to the speeds of the front vehicles Obj3 and Obj4 observed by the vehicle at the beginning, but refer to the actual speed values ​​of the front vehicles Obj3 and Obj4 when they are traveling at each current moment.

[0154] It should be noted that in the embodiments of the present application, the calculation of the reference speed of the game object may not be limited to the above-mentioned implementation method based on the safety distance method, but may also be based on the safety time distance method, expert experience system and data-driven method to calculate the reference speed. The present application does not specifically limit this.

[0155] B. There are other obstacles, and the obstacles are not on the road topology of the game object

[0156] For dynamic objects that are not in the road topology of the game object (which can be called dynamic objects in other road topologies), the above safe vehicle distance method cannot be used to calculate the reference speed. Therefore, this application further classifies dynamic objects in other road topologies to explain the calculation method of the reference speed, including but not limited to the following methods:

[0157] a. Other obstacles are moving in the same direction and parallel to the game object

[0158] The game object has n road topologies. If there is at least one obstacle (which can be called a second target obstacle, which can be one or more) in the same driving direction as the game object, and the road topology owned by the at least one second target obstacle does not belong to the above n road topologies, the self-vehicle can first obtain the benchmark speed of each second target obstacle according to a preset method, and then obtain the reference speed of the game object based on each benchmark speed.

[0159] For ease of understanding, the following is still based on Figure 7 Take 2 as an example to explain the reference speed of the game object in this case: Figure 7 It can be seen that Obj5 belongs to this situation, that is, the road topology owned by Obj5 does not belong to the road topology r (1) and r (2) , and Obj5 and the game object are traveling in the same direction and parallel. For Obj5, the present application can first use the method of penetrating virtual walls (the method in another related patent of the present applicant, which will not be repeated here) to calculate the reference speed V of Obj5 toObj5 If there are other obstacles in the safe distance that are parallel to the game object in the same direction besides Obj5, then calculate the reference speed V corresponding to each obstacle that is parallel to the game object in the same direction. toObjx , finally take Obj5 and all V toObjx The minimum value in is taken as the final reference speed of the game object in this case As shown in formula (6):

[0160]

[0161] Among them, x is the serial number of all obstacles running in the same direction and parallel to the game object. Figure 7 In the corresponding example, only Obj5 meets this condition, so the final reference speed is Just for V toObj5 .

[0162] b. Other obstacles are obstacles that cross the game object

[0163] The game object has n road topologies. If there is at least one obstacle (called the third target obstacle, which can be one or more) in the cross-travel direction of the game object, the vehicle can first determine whether the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than a preset threshold, and calculate the reference speed based on the judgment result, which is explained below:

[0164] (1) The time difference between the game object and at least one third target obstacle reaching the predicted collision point is greater than a preset threshold

[0165] Under this judgment result, the reference speed of the game object can be the real-time speed of the game object. It should be noted that the real-time speed of the game object here does not refer to the speed of the game object observed by the vehicle at the beginning, but refers to the actual speed value of the vehicle when the game object is traveling at each current moment.

[0166] Similarly, for ease of understanding, the following is still based on Figure 7 Take 2 as an example to explain the reference speed of the game object in this case: Figure 7 It can be seen that Obj6 belongs to this situation, that is, Obj6 is the crossing vehicle of the game object. The self-vehicle can first calculate the intersection of Obj6's predicted trajectory and the game object's predicted trajectory and the distance s to the intersection. obj6 and s, and calculate the time ttc when Obj6 reaches the intersection based on their current observed speed and acceleration obj6 The time ttc when the game objects reach the intersection is calculated, so as to obtain the time difference tdtc when the two reach the collision point. The calculation of the time difference tdtc is shown in formula (7):

[0167] tdtc=ttc obj6 -ttc (7)

[0168] If tdtc>tdtd thresh (tdtd thresh is the preset threshold), it means that the game object is tdtd ahead of Obj6 thresh This situation is relatively safe, so the reference speed of the game object can remain unchanged, that is:

[0169]

[0170] in, is the reference speed of the game object, v is the current real-time speed of the game object. It should be noted that the real-time speed of the game object here does not refer to the speed of the game object observed by the vehicle at the beginning, but refers to the actual speed value of the vehicle when the game object is traveling at each current moment.

[0171] (2) The time difference between the game object and the at least one third target obstacle reaching the predicted collision point is less than or equal to a preset threshold.

[0172] Under this judgment result, the reference speed of the game object is obtained based on the real-time speed of the game object, and the reference speed needs to be less than the real-time speed of the game object. Similarly, the real-time speed of the game object here does not refer to the speed of the game object observed by the vehicle at the beginning, but refers to the actual speed value of the vehicle when the game object is driving at each current moment.

[0173] Similarly, for ease of understanding, the following is still based on Figure 7 As an example, the reference speed of the game object in this case is explained: According to the above formula (7), the time difference tdtc between Obj6 and the game object reaching the collision point can be obtained. If tdtc <tdtd thresh , it means that the time when the two arrive at the collision point is relatively close. This situation is more dangerous, so the game object should slow down and give way. In this case, the reference speed of the game object It can be calculated by formula (9):

[0174]

[0175] Among them, t is the cumulative time counted from the first time Obj6 enters the safe distance, and the decision module performs cumulative updates every frame; d safe is the safety threshold, i.e., d in front of the collision point safe It should be noted that the calculation method for other crossing vehicles except Obj6 is similar and will not be described here. Finally, the reference speed corresponding to Obj6 and all other crossing vehicles is taken as The minimum value of is used as the final reference speed of the game object in this case As shown in formula (10):

[0176]

[0177] Among them, y is the serial number of all obstacles that cross the game object. Figure 7 In the corresponding example, only Obj6 meets this situation, so the final reference speed is Just for

[0178] Finally, the straight road topology r (2) The speed constraint of the environmental dynamic objects is the minimum value of the speed constraint of the front vehicle and the speed constraint of the dynamic objects in other lanes.

[0179]

[0180] Similarly, for the left-turn road topology r (1) The dynamic object velocity constraints are:

[0181]

[0182] It should be noted that the above-mentioned methods of calculating the reference speed of the game object are only illustrative. The reference speed of the game object is not limited to the above methods, but can also be calculated based on data-driven, expert experience and other methods. This application does not specifically limit this.

[0183] Based on the above, after obtaining the feature reference information of the game object, the intention probability of the game object traveling along each road topology can be determined based on the real-time state information of the game object and the obtained feature reference information. In some embodiments of the present application, one implementation method for determining the intention probability of the game object traveling along each road topology according to the real-time state information and feature reference information of the game object can be: first, according to the real-time state information of the game object and the feature reference information calculated by the above method, a target feature vector is obtained. As an example, assuming that the feature reference information includes reference position information, reference heading angle and reference speed, then the method for calculating the target feature vector can be shown as formula (13):

[0184]

[0185] Among them, f k,pos ,f k,θ ,f k,v is the target feature vector, which can be recorded as (f k,pos ,f k,θ ,f k,v ); They are the game objects in the road topology r (i) The reference position, reference orientation angle and reference speed on the c ,θ c 、V c They are respectively the current observed position of the game object (ie, real-time position information), the observed real-time heading angle, and the observed real-time speed.

[0186] After that, the likelihood probability of the target feature vector is determined based on the previously constructed probability density distribution data set, where the probability density distribution data set includes the distribution of the probability density of the feature vector under different feature vector values. As an example, assuming that the feature reference information includes reference position information, reference orientation angle, and reference speed, then the target feature vector (f k,pos ,f k,θ ,f k,v The likelihood probability of each feature vector in ) can be recorded as P(fk,pos ∣r (i) ), P(f k,θ ∣r (i) ), P(f k,v ∣r (i) ).

[0187] It should be noted that, in the embodiment of the present application, the likelihood probability P(f k,pos ∣r (i) ), which can be qualitatively described as: the current position of the game object (i.e., the center of mass position) is away from the road topology r (i) The farther the reference position is (i.e., f k,pos increases), then the game object will follow the road topology r in the future (i) The lower the probability of driving, that is, P(f k,pos ∣r (i) ) should be followed by f k,pos increases and decreases; similarly, for the likelihood probability P(f k,θ ∣r (i) ), which can be qualitatively described as: the current real-time orientation angle θ of the game object c With road topology (i) The reference orientation angle θ on r(i) Difference (f k,θ ) is larger, the game object will follow the road topology r in the future. (i) The lower the probability of driving, that is, P(f k,θ ∣r (i) )With f k,θ decreases with the increase of; similarly, for the likelihood probability P(f k,v ∣r (i) ), which can be qualitatively described as: the current real-time speed V of the game object c With road topology (i) Reference speed on The larger the gap, the more likely the game object will follow the road topology r in the future. (i) The lower the probability of driving, that is, P(f k,v ∣r (i) )With f k,v decreases with the increase of .

[0188] As an example, the likelihood probability P(f k,pos ∣r (i) ), P(f k,θ ∣r (i) ), P(f k,v ∣r (i) ) and the target feature vector (f k,pos ,f k,θ ,f k,v ) can be expressed as Figure 8 As shown, in Figure 8 In FIG. 1 , a distribution diagram of the characteristic vector-likelihood probability relationship distribution of the game object going straight and turning left is shown when the speed V is Q km / s (such as 8 km / s), where Figure 8 (a) Sub-diagram in the figure, with the horizontal axis being f k,pos , the vertical coordinate is P(f k,pos ∣r (i) ), similarly, for Figure 8 (b) The horizontal axis is f k,θ , the vertical coordinate is P(f k,θ ∣r (i) ),for Figure 8 (c) The horizontal axis is f k,v , the vertical coordinate is P(f k,v ∣r (i) ).

[0189] In the above-mentioned embodiments of the present application, the target feature vector-likelihood probability relationship is designed using a normal distribution (if the environmental factors, aggressiveness, etc. of the game object are also considered, it may be a form of approximate normal distribution). It should be noted that in other embodiments of the present application, the design of this relationship is not limited to expert database generation, based on Datadriven statistical data, etc., and the present application does not limit this.

[0190] It should be noted that from Figure 8 It can be seen that the closer the observed value is to the reference state, the greater the likelihood probability. In addition, considering that the game object is at different speeds, the eigenvector and likelihood probability distribution should have different distribution forms. For this reason, the present application can also design a nonlinear probability density distribution data set based on different speeds. The nonlinear probability density distribution data set includes the distribution of the probability density of the eigenvector under different speed values ​​and different eigenvector values. For details, please refer to Fig. 9 , for different speed values, there are probability density distribution data sets corresponding to the speed. Figure 8 This is a probability density distribution data set when V is Q km / s (such as 8 km / s). Figure 7 Taking the example of as an example (the feature reference information includes reference position information, reference orientation angle and reference speed), the respective situations of the nonlinear probability density distribution data set are explained:

[0191] (1) Probability density distribution of the straight-moving intention position at different speed values

[0192] When the speed V is small (e.g. V = 10 km / s), the position of the game object needs to change significantly to represent the change in its intention (compared to the case of higher speed), so the distribution should be flatter, e.g. Fig.10 On the other hand, when the speed V of the game object is high (such as V = 50 km / s), a slight lateral position shift means an obvious tendency to change the intention, so the probability density distribution should be more concentrated, such as Fig.10 The right sub-figure.

[0193] However, it should be noted that, in the embodiment of the present application, since the observed real-time speed of the game object may be any value, and the pre-constructed nonlinear probability density distribution data set is for a limited number of different speed values, such as V: {5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100} km / s, a total of 11 different V values, the probability density distribution data set, if the current observed real-time speed of the game object is one of the 11 different V values ​​(such as the current observed real-time speed is 50 km / s), then based on V = 50 km / s under the straight intention position probability density distribution data set (that is, Fig.10 As shown in the rightmost sub-figure), take the current target feature vector f k,pos The corresponding P(f k,pos ∣r (2) ) as f k,pos ; if the currently observed real-time speed of the game object is not any of the 11 different V values ​​(for example, the currently observed real-time speed is 45km / s), then based on the two V values ​​adjacent to the currently observed real-time speed (i.e., V = 40km / s and V = 50km / s), the probability density distribution data set of the straight-line intention position is used. As an example, the current target feature vector f under two adjacent V values ​​can be taken. k,pos The corresponding P(f k,pos ∣r (2) ) as the interpolation value of f k,pos Likelihood probability, for example, assuming V = 40km / s, P(f k,pos ∣r (2) )=0.7,V=50km / s,P(f k,pos ∣r (2) )=0.9, then when the speed of the game object is 45km / s, the corresponding P(f k,pos ∣r (2) ) takes the middle value between 0.7 and 0.9, that is, 0.8. As another example, calculate P(f k,pos ∣r (2) ) can also be nonlinear, which will not be described in detail here.

[0194] (2) Probability density distribution of the straight-going intention heading angle at different speed values

[0195] When the speed V is small (such as V = 10 km / s), the orientation angle of the game object needs to change greatly to represent the change of its intention (compared to the higher speed), so the distribution should be flatter, such as Fig.11 On the other hand, when the speed V of the game object is high (such as V = 50 km / s), as long as the steering wheel is slightly turned, there will be an obvious tendency to change the intention, so the probability density distribution at this time should be more concentrated, such as Fig.11 The right sub-figure.

[0196] Similarly, in the embodiment of the present application, since the observed real-time speed of the game object may be any value, and the pre-constructed nonlinear probability density distribution data set is for a limited number of different speed values, such as V: {5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100} km / s, a total of 11 different V values, the probability density distribution data set, if the current observed real-time speed of the game object is one of the 11 different V values ​​(such as the current observed real-time speed is 50 km / s), then based on the straight-ahead intention heading angle probability density distribution data set under V = 50 km / s (i.e. Fig.11 As shown in the rightmost sub-figure), take the current target feature vector f k,θ The corresponding P(f k,θ ∣r (2) ) as f k,θ ; if the currently observed real-time speed of the game object is not any of the 11 different V values ​​(for example, the currently observed real-time speed is 45km / s), then based on the two V values ​​adjacent to the currently observed real-time speed (i.e., V = 40km / s and V = 50km / s), the probability density distribution data set of the straight-line intention heading angle can be used as an example. Under two adjacent V values, the current target feature vector f k,θ The corresponding P(f k,θ ∣r (2) ) as the interpolation value of f k,θ Likelihood probability, for example, assuming V = 40km / s, P(f k,θ ∣r (2) )=0.7,V=50km / s,P(f k,θ ∣r (2) )=0.8, then when the speed of the game object is 45km / s, the corresponding P(f k,θ ∣r (2) ) takes the middle value between 0.7 and 0.8, that is, 0.75.

[0197] (3) Probability density distribution of straight-moving intention speed under different speed values

[0198] When the speed V is small (e.g., V = 10 km / s), a small change in the speed of the game object (compared to a higher speed) can represent a change in its intention, so the distribution should be more concentrated, such as Fig.12 On the other hand, when the speed V of the game object is high (such as V = 50 km / s), a larger speed change is required to represent the change of intention, so the probability density distribution should be flatter, such as Fig.12 The right sub-figure.

[0199] Similarly, in the embodiment of the present application, since the observed real-time speed of the game object may be any value, and the pre-constructed nonlinear probability density distribution data set is for a limited number of different speed values, such as V: {5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100} km / s, a total of 11 different V values, the probability density distribution data set, if the current observed real-time speed of the game object is one of the 11 different V values ​​(such as the current observed real-time speed is 50 km / s), then based on the straight-line intention speed probability density distribution data set under V = 50 km / s (i.e. Fig.12 As shown in the rightmost sub-figure), take the current target feature vector f k,v The corresponding P(f k,v ∣r (2) ) as f k,v ; if the currently observed real-time speed of the game object is not any of the 11 different V values ​​(for example, the currently observed real-time speed is 45km / s), then based on the two V values ​​adjacent to the currently observed real-time speed (i.e., V = 40km / s and V = 50km / s), the probability density distribution data set of the straight-line intention heading angle can be used as an example. Under two adjacent V values, the current target feature vector f k,v The corresponding P(f k,v ∣r (2) ) as the interpolation value of f k,θ Likelihood probability, for example, assuming V = 40km / s, P(f k,v ∣r (2) )=0.8,V=50km / s,P(f k,v ∣r (2) )=0.9, then when the speed of the game object is 45km / s, the corresponding P(f k,v ∣r (2) ) takes the middle value between 0.8 and 0.9, that is, 0.85.

[0200] For the intention to turn left, similarly, it will not be elaborated here. One point that needs to be made clear is that compared with going straight, the left turn topology has a larger curvature, so the reference speed is smaller and the reference heading angle is larger. Therefore, compared with the probability density distribution of going straight, the probability density distribution of the reference speed is more concentrated, and the probability density distribution of the reference heading angle is flatter. This application will not elaborate on this in detail.

[0201] Based on the above, after determining the likelihood probability of the target feature vector according to the previously constructed probability density distribution data set, the intention probability of the game object traveling along each road topology is determined according to the likelihood probability. In some embodiments of the present application, one implementation method for determining the intention probability of the game object traveling along each road topology according to the likelihood probability may be: first, obtain the prior intention probability of the obstacle traveling along each road topology, and then, based on the likelihood probability and the prior intention probability, use the Bayesian reasoning algorithm to calculate the intention probability of the game object traveling along each road topology. The calculation method may be shown as formula (14):

[0202]

[0203] Among them, f k,pos ,f k,θ ,f k,v is the target feature vector, which can be obtained based on the above formula (13); P(f k,pos ∣r (i) ), P(f k,θ ∣r (i) ), P(f k,v ∣r (i) ) are respectively in the road topology r (i) The position probability density, orientation angle probability density, and velocity probability density (such as Figure 8 or Fig. 9 As shown), they represent the situation when the game object has r (i) When the direction of the intention is k,pos ,f k,θ ,f k,v ) likelihood probability; P(f k,pos ∣r (i) ), P(f k,θ ∣r (i) ), P(f k,v ∣r (i) ) is calculated in advance and stored on the vehicle. For each target speed (e.g., V = 10, 30, 50, etc.), a corresponding set of P(f k,pos ∣r (i) ), P(f k,θ ∣r (i) ), P(f k,v ∣r (i)) data distribution set. Whichever V corresponds to the real-time speed of the currently observed game object, the corresponding set of P(f k,pos ∣r (i) ), P(f k,θ ∣r (i) ), P(f k,v ∣r (i) ). If the current real-time speed of the game object is 15km / s, and there is no likelihood probability set at this speed in the database, then the interpolation of the likelihood probabilities under two adjacent V can be taken. For example, when V = 10km / s, P(f k,pos ∣r (i) )=0.10,P(f k,θ ∣r (i) )=0.21,P(f k,v ∣r (i) )=0.30,V=30km / s,P(f k,pos ∣r (i) )=0.19,P(f k,θ ∣r (i) )=0.10,P(f k,v ∣r (i) )=0.39, then the real-time speed of the current game object is 15km / s corresponding to the road topology r (i) The likelihood probabilities are 0.15, 0.16, and 0.35 respectively (see above for details); P r(i) The game object is in the road topology r (i) The prior probability that the road topology r (i) The prior probability of driving comes from the upstream prediction module; P(r k =r (i) ,k=i|f k,pos ,f k,θ ,f k,v ) is the intention r (i) The posterior probability of the game object is that after considering the current observed state and the prior probability of the game object, the game object is calculated according to r (i) Probability of driving intention.

[0204] In summary, the core idea of ​​intention posterior probability reasoning is: through the observed state of the game object (i.e., real-time state information) and the corresponding feature reference state (i.e., feature reference information), the likelihood probability P(f k ∣r (i) ), and then combined with the prior probability of each intention The posterior probability P(r k =r (i) ).

[0205] It should be noted that, in some embodiments of the present application, if the prior probability It is difficult to obtain and the prior probability may not be needed. In this case, as an example, the probability mean can be taken based on the number of road topologies where the target vehicle is located. For example, in the above embodiment of the present application, assuming that the feature reference information includes two road topologies, namely, a straight topology and a left turn topology, to which the game object belongs at the same time, the prior probability can be considered to be 0.5.

[0206] In the above example of the present application, the vehicle can obtain the straight-moving intention probability P(r k =r (2) ) and the left turn intention probability P(r k =r (1) Considering that the game object is in a straight / left turn multi-intention lane, according to traffic rules, the game object cannot turn right, so P(r k =r (1) )+P(r k =r (2) )=1.

[0207] 503. According to the real-time state information of the vehicle and the real-time state information of the game object, a sampling game space is constructed for the game object, where the sampling game space includes m game strategies, where m≥1.

[0208] After obtaining the real-time status information of the game object, the ego vehicle can further obtain the real-time status information of the ego vehicle (such as the vehicle information of the ego vehicle collected by the ego vehicle sensor system, etc.), and construct a sampling game space for the game object based on the real-time status information of the ego vehicle and the real-time status information of the game object (and additional road condition information), and the sampling game space includes m game strategies, m≥1.

[0209] It should be noted that in the embodiment of the present application, one implementation method for constructing a sampling game space for the game object may be: kinematic deduction of the lateral and longitudinal feasible spaces of all vehicles and game objects, wherein the lateral and longitudinal feasible spaces are not limited to continuous distribution, discrete distribution, mixed continuous and discrete distribution, and decision-making considering behavioral uncertainty, etc., and the present application does not limit this. For ease of explanation, the present application adopts a discrete distribution method to construct a sampling game space. Specifically, the sampling game space may include a longitudinal feasible space and a lateral feasible space, which are explained below respectively:

[0210] (1) Vertical feasible space

[0211] The longitudinal game strategy of the ego vehicle and the game object can be characterized by the size of acceleration / deceleration (rushing / yielding), that is, by longitudinal acceleration offset sampling. First, the decision upper limit and decision lower limit of the game strategy (acceleration) are generated, which can be obtained based on the longitudinal dynamics of the vehicle, kinematic constraints, and the relative position and relative speed of the ego vehicle and the game object. Then, according to the preset rules, the decision strategy (i.e., action) of the ego vehicle and the decision strategy (i.e., action) of the game object are obtained in the decision upper limit and decision lower limit. Finally, the decision strategy of the ego vehicle is combined with the decision strategy of the game object to obtain at least one game strategy (i.e., action pair) between the ego vehicle and the game object. As an example, considering the balance between computational complexity and spatial accuracy of the sampling strategy, the acceleration interval can be set to 1m / s 2 For example, a longitudinal sampling strategy space, namely the longitudinal feasible space, can be generated as shown in Table 2, where acc is the absolute value of the acceleration of the vehicle (the vehicle or the game object).

[0212] Table 2. Longitudinal sampling strategy space (longitudinal feasible space)

[0213]

[0214] It should be noted that in Table 2, a total of 8*8=64 longitudinal acceleration offset samples are shown for the vehicle and the game object. The above Table 2 is only for illustration and does not constitute a limitation on this application. In actual applications, the interval and degree of longitudinal acceleration offset sampling can be set based on the actual scenario of straight driving, and this application will not elaborate on this.

[0215] (2) Horizontal feasible space

[0216] The lateral game strategy of the ego vehicle and the game object can be characterized by the size of the lateral position offset, such as Fig.13 As shown. Similarly, first, the decision upper limit and decision lower limit of the game strategy (lateral position offset) are generated, which can be obtained based on the longitudinal dynamics of the vehicle, kinematic constraints, and the relative position and relative speed of the self-vehicle and the game object. Then, according to the preset rules, the decision strategy (i.e., action) of the self-vehicle and the decision strategy (i.e., action) of the game object are obtained in the decision upper limit and decision lower limit. Finally, the decision strategy of the self-vehicle is combined with the decision strategy of the game object to obtain at least one game strategy (i.e., action pair) of the self-vehicle and the game object. As an example, considering the balance between computational complexity and sampling strategy space accuracy, the lateral position offset interval can be set to 1m. For example, a lateral sampling strategy space as shown in Table 3, i.e., a lateral feasible space, can be generated.

[0217] Table 3. Longitudinal sampling strategy space (longitudinal feasible space)

[0218]

[0219] It should also be noted that in Table 3, a total of 7*7=49 lateral position offset samples are shown for the vehicle and the game object. The above Table 3 is only for illustration and does not constitute a limitation on this application. In actual applications, the interval and degree of lateral position offset sampling can be set based on the actual scene of straight driving, and this application will not elaborate on this.

[0220] It should also be noted that each small grid in the lateral position offset grid is expanded to form a grid for longitudinal acceleration offset. Taking Tables 2 and 3 as examples, all other lateral position offsets of the ego vehicle and the game object are sampled, as well as longitudinal acceleration offsets, for a total of 49 lateral offset combinations (7 for the ego vehicle and 7 for the social vehicle, 7*7=49), and a total of 8*8=64 longitudinal acceleration offset samples are collected under each lateral offset sampling combination (see Table 2 for details). Fig.14 ). Finally, 49*64=3136 action pairs (also called action combinations) are formed, each action pair is a game strategy in the sampled game space, that is, m=3136, and the sampled game space includes 3136 game strategies.

[0221] It should also be noted that in the embodiments of the present application, step 502 and step 503 may be executed in any order, and step 502 may be executed first and then step 503, or step 503 may be executed first and then step 502, or step 502 and step 503 may be executed simultaneously, and the present application does not make any specific limitation on this.

[0222] 504. Calculate the strategy costs of the m game strategies under each road topology to obtain n groups of strategy costs, wherein each group of strategy costs includes m strategy costs corresponding to the m game strategies one by one.

[0223] After constructing a sampling game space including m game strategies (i.e., each action pair consisting of the ego vehicle's movable action and the game object's movable action), for each combination of the ego vehicle's lateral and longitudinal actions and the game object's lateral and longitudinal actions, the strategy cost is evaluated under each road topology, thereby obtaining n sets of strategy costs, wherein each set of strategy costs includes m strategy costs corresponding to the m game strategies. For example, still taking Tables 2 and 3 as examples, it can be seen from Tables 2 and 3 that the constructed sampling game space includes a total of 3136 game strategies. Assuming that the game object has two road topologies (i.e., n=2), a total of two sets of strategy costs corresponding to the two road topologies are obtained, and each set of strategy costs includes 3136 game strategies.

[0224] It should be noted that in some embodiments of the present application, the strategy cost (cost) can be evaluated based on the interaction between the future movement of the vehicle and the game object. The smaller the cost, the higher the strategy benefit, and the higher the possibility that the strategy will be used as the optimal strategy. Specifically, a method for calculating the strategy cost of m game strategies under each road topology can be: determine the various factors of the strategy cost of m game strategies under each road topology, such as safety, comfort, passability, road right, lateral deviation, risk area, inter-frame association, etc.; then, calculate the factor cost of each factor in each strategy cost. The method for calculating the factor cost can be to determine the weights of each factor in advance, such as safety weight, comfort weight, passability weight, road right weight, lateral deviation weight, risk area weight, inter-frame association weight, etc. (the values ​​of each weight can be different), and based on each weight, the factor cost of each factor in each strategy cost is calculated. Weighted, the strategy cost of m game strategies under each road topology is obtained, where the value obtained by multiplying each factor by the corresponding weight is called the factor cost, such as safety cost, comfort cost, passability cost, road right cost, lateral offset cost, risk area cost, and inter-frame association cost. Each factor cost can be calculated according to different cost functions, which can be respectively called safety cost function, comfort cost function, passability cost function, road right cost function, lateral offset cost function, risk area cost function, and inter-frame association cost function. In some implementations, the above factor cost can also be normalized and the value range is [0,1]. For ease of understanding, the various factors and factor cost of the strategy cost are explained below, including but not limited to:

[0225] (1) Security

[0226] The safety cost (also called safety cost) can be calculated based on the safety cost function with the relative distance between the self-vehicle and other vehicles (i.e., the game object) as the independent variable, and the safety cost is negatively correlated with the relative distance. For example, the greater the relative distance between the two vehicles, the smaller the safety cost. In some embodiments, the safety cost can also take into account the relative speed and relative angle of the interaction between the self-vehicle and the game object. As an example, Fig.15 is a schematic diagram of a normalized security cost function, where c dist is the security cost, corresponding to Fig.15 The vertical coordinate of dist is the relative distance between the vehicle and the game object, corresponding to Fig.15 The horizontal axis of .

[0227] (2) Comfort

[0228] The comfort cost can be referred to as comfort cost. The comfort cost of a vehicle (the vehicle itself or the game object) can be calculated using a comfort cost function with the absolute value of the acceleration change (i.e., jerk) as the independent variable. As an example, Fig.16 is a schematic diagram of a comfort cost function after homogenization, where c com is the comfort cost, corresponding to Fig.16 The vertical coordinate of Fig.16 The horizontal axis is jerk, which is the change in acceleration of the vehicle or the game object.

[0229] It should be noted that in some embodiments of the present application, the acceleration change of the vehicle (the vehicle or the game object) can be the longitudinal acceleration change, the lateral acceleration change, or the weighted sum of the two. Generally speaking, the smaller the lateral / longitudinal acceleration change of the vehicle or the game object, the better the comfort and the smaller the comfort cost. In some embodiments, the comfort cost can be the comfort cost of the vehicle or the comfort cost of the game object, or the weighted sum of the comfort costs of the two.

[0230] (3) Passability

[0231] The passability cost can be referred to as the passability cost, which can be calculated based on the passability cost function with the speed change of the vehicle or the game object as the independent variable. For example, if a vehicle yields with a large deceleration, it will result in a large speed loss (the difference between the current speed and the future speed, i.e., acceleration) or a long waiting time, and the vehicle's passability cost will increase; for another example, if a vehicle rushes to pass with a large acceleration, it will result in a large speed increase (the difference between the current speed and the future speed, i.e., acceleration) or a short waiting time, and the vehicle's passability cost will decrease.

[0232] It should be noted that in some embodiments of the present application, the passability cost can also be calculated based on a passability cost function with the relative speed ratio of the vehicle or the game object as the independent variable. For example, before executing the action, the absolute value of the vehicle's speed accounts for a larger proportion of the sum of the absolute values ​​of the vehicle's speed and the game object's speed, and the absolute value of the game object's speed accounts for a smaller proportion of the sum of the absolute values ​​of the vehicle's speed and the game object's speed. After executing the action pair, if the vehicle gives way at a larger deceleration, its speed loss increases, and the speed ratio decreases, then the passability cost corresponding to the vehicle's execution of the action is larger; and if, after executing the action pair, the game object rushes in at a larger acceleration, its speed increases, and the speed ratio increases, then the passability cost corresponding to the game object's execution of the action is smaller.

[0233] It should be noted that in some embodiments of the present application, the passability cost can be the corresponding passability cost of the ego vehicle when the ego vehicle performs the action, or it can be the passability cost of the corresponding game object when the game object performs the action, or it can be the weighted sum of the two passability costs, and the present application does not make any specific limitations on this.

[0234] As an example, Fig.17 is a schematic diagram of the passability cost function after normalization, where c pass is the passing cost, corresponding to Fig.17 The vertical coordinate is the absolute value of the vehicle speed, corresponding to Fig.17 The abscissa of the vehicle. The absolute value of the vehicle's midpoint speed is speed0, and the maximum absolute value of the vehicle's speed is speed1.

[0235] (4) Right of way

[0236] The right of way cost can be referred to as the right of way cost. The right of way information corresponding to the vehicle can be determined based on the user portrait of the vehicle or the game object. For example, if the driving behavior of the game object is aggressive and tends to adopt a rushing strategy, it is high right of way; if the driving behavior of the game object is conservative and tends to adopt a yielding strategy, it is low right of way. Among them, high right of way tends to maintain the established motion state or established driving behavior, and low right of way tends to change the established motion state or established driving behavior.

[0237] In an embodiment of the present application, the user portrait can be determined based on the user's gender, age, and completion of historical behavioral actions. In some embodiments, the cloud server can obtain the data required to determine the user portrait and determine the user portrait. If the self-vehicle and / or the game object executes the behavioral action to cause the vehicle with high road right to change its motion state, the corresponding road right cost of the behavioral action is larger and the benefit is smaller. In some embodiments, a higher road right cost can also be determined for the behavioral decision that causes the high road right vehicle to change its motion state to increase the penalty. Through this feedback mechanism, the behavioral action that allows the high road right vehicle to maintain its current motion state has a larger road right benefit, that is, a smaller road right cost.

[0238] As an example, Fig.18 is a schematic diagram of a road right cost function after homogenization, where c road is the road right cost, corresponding to Fig.18 The vertical coordinate of , acc is the absolute value of the acceleration of the vehicle (the vehicle or the game object), corresponding to Fig.18 The horizontal axis of .

[0239] (5) Risk areas

[0240] The risk area cost can be referred to as the risk area cost. For example, if a vehicle (the vehicle itself or the game object) is in a risk area on the road (in this area, the vehicle has a greater driving risk and needs to leave the area as soon as possible), a higher risk area cost needs to be imposed on the vehicle yielding strategy to increase the penalty. By not choosing the vehicle yielding behavior but choosing the vehicle rushing behavior as the decision result, the vehicle can leave the risk area as soon as possible, that is, the decision is made to leave the risk area as soon as possible, so as to ensure that the vehicle leaves the risk area as soon as possible without causing serious impact on traffic.

[0241] (6) Lateral offset

[0242] The lateral offset cost can also be called the offset cost, which can be calculated based on the lateral offset of the vehicle or the game object. Fig.19 is a schematic diagram of the lateral offset cost function after normalization, where c offset is the offset cost, corresponding to Fig.19 The vertical coordinate of the vehicle, offset is the lateral offset of the vehicle, corresponding to Fig.19 The horizontal axis unit is m. The farther the vehicle / game object is offset, the higher the offset cost is in a piecewise manner. After exceeding the soft boundary of the road, the slope of the offset cost increases. When exceeding the hard boundary of the road, the offset cost increases with a greater penalty. That is, the larger the lateral offset of the vehicle, the smaller the lateral offset benefit and the larger the offset cost. Moreover, after the lateral offset of the vehicle is greater than the middle value of the lateral offset, the offset cost increases faster to increase the penalty. After the lateral offset of the vehicle is greater than the upper limit threshold of the lateral offset, for example, the lateral offset cost is a fixed value of 1.2 to increase the penalty.

[0243] (7) Inter-frame association

[0244] The inter-frame correlation cost can also be called the inter-frame correlation cost. If the decision in the previous frame is the object of the game of grabbing, the cost of grabbing in this frame will be appropriately reduced, and the cost of letting go will be appropriately increased, making it easier to output the decision of grabbing, which is consistent with the motion state of the previous frame, and stabilizing the decision state in the time domain. The same is true for letting go. As an example, Fig. 20This is a schematic diagram of the inter-frame correlation cost function after normalization. The intention decision of the ego vehicle in the previous frame K is to grab the object of the game. If the intention decision of the ego vehicle in the current frame K+1 is to grab the object of the game, the corresponding inter-frame correlation cost will be small, such as 0.3 (the default value is 0.5), so it is a reward; if the intention decision of the ego vehicle in the current frame K+1 is to give way to the object of the game, the corresponding inter-frame correlation cost will be large, such as 0.8 (the default value is 0.5), so it is a penalty. At this time, the strategy of choosing to make the intention decision of the ego vehicle in the current frame as the object of the game of grabbing becomes the feasible solution of the current frame. Through the above penalty or reward for the inter-frame correlation cost, it can be ensured that the intention decision of the ego vehicle in the current frame is consistent with the intention decision of the previous frame, so that the motion state of the ego vehicle in the current frame is consistent with the motion state of the previous frame, and the behavior decision of the ego vehicle is stabilized in the time domain.

[0245] In addition, the weights of the above costs can also be different. By weighting and summing the costs, the total cost of an action pair can be obtained. According to the sampled actions of the two vehicles corresponding to each cost, the order of the conflict area (i.e., the area where the two vehicles meet) is deduced. Taking the situations in Tables 2 and 3 as examples, the situation where the ego vehicle arrives first can be classified as follows: Fig.21 The cost of the game object shown in the figure (which can be recorded as c grab ), otherwise it is classified as the give-and-take game object cost(c yield ). The weight distribution can be referred to as: safety weight > road right weight > lateral offset weight > passability weight > comfort weight > inter-frame correlation weight. Considering safety, the scenario where the two vehicles pass the conflict area too close (i.e. extremely dangerous solution, marked as -1) should be excluded, and the scenario where both vehicles brake before the conflict area (i.e. actually infeasible solution, marked as 0) should be excluded. After induction, the game strategies with the following conditions can constitute a feasible domain, such as Fig.21 shown.

[0246] In summary, in the application embodiment, it is necessary to solve the strategy cost of each game strategy in the sampled game space for each of the n road topologies, and obtain a total of n groups of strategy costs, and each group of strategy costs includes m strategy costs, and these m strategy costs correspond to m game strategies one by one. Figure 6 or Figure 7 The corresponding example is used as an example to illustrate. Figure 6 or Figure 7 In the game, the game object has the left-turn road topology and the straight-through road topology, so the left-turn intention and the straight-through intention need to be solved separately, which are explained below.

[0247] For the left turn intention of the game object, the game process is as follows: First, sample the trajectory where the lateral position offset of the vehicle and the game object is 0 (such as Fig. 22 On this basis, the longitudinal acceleration of the vehicle and the game object is sampled (8 longitudinal actions of the vehicle, 8 longitudinal actions of the game object, a total of 64 combinations, such as Fig. 22 The grid on the right shows the trajectory where the lateral position offset of the vehicle is 1 and the lateral position offset of the game object is 0 (as shown in Fig. 22 On this basis, the longitudinal acceleration of the ego vehicle and the game object is sampled (8 longitudinal actions of the ego vehicle and 8 longitudinal actions of the game object, a total of 64, as shown in Fig. 22 As shown in the grid on the right). Similarly, all other lateral position offsets of the ego vehicle and the game object are sampled, as well as the longitudinal acceleration offset, for a total of 49 lateral offset combinations (7 for the ego vehicle and 7 for the game object, 7*7=49), and a total of 8*8=64 longitudinal acceleration offset samples for each lateral offset sampling combination. Finally, 49*64=3136 action pairs are formed. Using the method of calculating the cost of each factor mentioned above, the strategy cost of each action pair can be calculated, for a total of 3136 (i.e., m=3136) strategy costs, which can be expressed as formula (15):

[0248] actionPair left,i ,i∈[1,3136]

[0249] cost left,i ,i∈[1,3136] (15)

[0250] Among them, actionPair left,i is an action pair, cost left,i The corresponding action is the strategic cost of the left turn intention.

[0251] Similarly, for the straight-moving intention of the game object, the same steps as above can be followed to obtain 3136 action pairs and 3136 corresponding strategy costs under the straight-moving intention, which can be expressed as formula (16):

[0252] actionPair straight,i ,i∈[1,3136]

[0253] cost straight,i ,i∈[1,3136] (16)

[0254] Among them, actionPair straight,i is an action pair, cost straight,i The corresponding strategic cost of this action for the intention of going straight.

[0255] It should be noted that in the embodiment of the present application, step 502 and step 504 may be executed in any order. Step 502 may be executed first and then step 504, or step 504 may be executed first and then step 502, or step 502 and step 504 may be executed at the same time. The present application does not make any specific limitation on this.

[0256] 505. According to the n intention probabilities and the n groups of strategy costs, determine the n target strategy costs whose total costs meet the preset conditions, and determine the decision result of the vehicle according to the target game strategy. The n target strategy costs are the strategy costs of the target game strategy under each road topology, and the target game strategy is one of the above m game strategies.

[0257] The intention probabilities of the game objects on each road topology (a total of n intention probabilities) have been calculated in the above step 502. Then, the vehicle determines n target strategy costs whose total costs meet the preset conditions based on the n intention probabilities and the n groups of strategy costs obtained in the above step 504. The n target strategy costs are the strategy costs of the target game strategy under each road topology, and the target game strategy is one of the above m game strategies.

[0258] It should be noted that in some embodiments of the present application, based on n intention probabilities and n groups of strategy costs, the implementation method of determining n target strategy costs whose total costs meet the preset conditions can be: using the n intention probabilities as the weights of the n road topologies respectively, calculating the weighted strategy costs corresponding to each game strategy in the n groups of strategy costs, and obtaining m total costs, then determining the total cost with the minimum value from the m total costs, and finally further determining the n target strategy costs corresponding to the total cost with the minimum value. The specific calculation method can be shown in formula (17):

[0259] TotalCost i =P1Cost 1,i +P2Cost 2,i +…+P n Cost 2,i ,i∈[1,m] (17)

[0260] Select the smallest TotalCost i (ie total cost) corresponding action actionPair i , the final uncertainty interactive game decision result of the vehicle can be obtained, that is, the solution principle of the embodiment of the present application: combining the deterministic interactive game solution with the intention probability to form an uncertain interactive game framework to solve the optimal decision under uncertain intentions.

[0261] To facilitate understanding of the above process, the following is still based on Figure 6 or Figure 7 The corresponding example is used as an example to illustrate. Figure 6 or Figure 7 In the game, the game object has a left-turn road topology and a straight road topology. After executing the above step 504 to obtain the deterministic strategy cost under each intention, the intention probability P(r k =r (i) ), we can obtain the comprehensive strategic cost of each action of the vehicle under the uncertainty of the intention of the game object. The calculation method is shown in formula (18):

[0262] TotalCost i =P1Cost left,i +P2Cost straight,i , i∈[1,3136] (18)

[0263] Select the smallest TotalCost i (ie total cost) corresponding action actionPair i , as the decision result of the final uncertainty interactive game of the ego vehicle. In the embodiment of the present application, P1 = 0.3, P2 = 0.7, that is, the game object has a greater probability (0.7) to continue to go straight. If the influence of P1 and P2 is not considered, the classic method believes that the trajectory of the game object turning left has a clear intersection with the reference trajectory of the ego vehicle, and there is a risk of collision, then the cost left,i , i∈[1,3136] searches for the smallest cost as the optimal solution and outputs the corresponding action, that is, the vehicle moves longitudinally at -2m / s 2 Slow down to give way, shift 1m to the right to avoid, and the final decision result is: avoid to the right laterally and give way longitudinally. But in fact, the game object will keep going straight in the future and will not turn left, so it is not reasonable for the self-vehicle to brake. Not only will the point brake lead to a worse comfort experience, but it will also affect traffic efficiency. Since this application takes into account the uncertainty of intention, the optimal action of the self-vehicle becomes 0m / s 2 The vehicle drives at a constant speed and shifts 1m to the right to avoid the obstacle. The final decision result is: avoid to the right and rush to the front. While ensuring safety, it also ensures passability. In the end, the game object keeps going straight, and the vehicle keeps a stable speed and passes the intersection quickly.

[0264] In summary, the core process architecture of the intelligent driving decision-making method provided in the embodiment of the present application can be as follows: Fig.23As shown, it mainly includes the following parts: S0, identification of objects with uncertain intentions (i.e., game objects) (as described in step 501); S1, identification of road topology affiliation of traffic participants (as described in step 501); S2, generation of feature reference information of target objects (i.e., a specific game object) (as described in step 502); S3, probabilistic reasoning of each intention of the target object (as described in step 502); S4, solution of uncertainty interactive game decision considering the probability of each intention (as described in steps 503-505).

[0265] On the basis of the above corresponding embodiments, in order to better implement the above solutions of the embodiments of the present application, the following also provides related devices for implementing the above solutions. Fig.24 , Fig.24 A structural diagram of a decision-making device provided in an embodiment of the present application, the decision-making device 2400 includes: an acquisition module 2401, an intention reasoning module 2402, a construction module 2403, a calculation module 2404, a determination module 2405 and a decision module 2406, wherein the acquisition module 2401 is used to determine a game object and obtain real-time status information of the game object, the game object is an obstacle with n road topologies, n ≥ 2; the intention reasoning module 2402 is used to determine the intention probability of the game object traveling along each road topology according to the real-time status information of the game object, and obtain n intention probabilities; the construction module 2403 is used to determine the intention probability of the game object traveling along each road topology according to the real-time status information of the vehicle and the real-time status of the game object. state information, construct a sampling game space for the game object, the sampling game space includes m game strategies, m≥1; a calculation module 2404, used to calculate the strategy costs of the m game strategies under each road topology, and obtain n groups of strategy costs, wherein each group of strategy costs includes m strategy costs corresponding to the m game strategies one by one; a determination module 2405, used to determine n target strategy costs whose total costs meet preset conditions according to the n intention probabilities and the n groups of strategy costs, the n target strategy costs are the strategy costs of the target game strategy under each road topology, and the target game strategy is one of the m game strategies; a decision module 2406, used to determine the decision result of the self-vehicle according to the target game strategy.

[0266] In one possible design, the intention reasoning module 2402 is specifically used to: obtain the feature reference information of the game object on each road topology, where the feature reference information is used to describe the baseline state of the game object's intention to travel along each road topology under the current traffic environment; determine the probability of the game object's intention to travel along each road topology based on the real-time state information of the game object and the feature reference information.

[0267] In one possible design, the intention reasoning module 2402 is further specifically used to: obtain a target feature vector based on the real-time status information of the game object and the feature reference information; determine the likelihood probability of the target feature vector based on a probability density distribution data set constructed in advance, the probability density distribution data set including the distribution of the probability density of the feature vector under different feature vector values; determine the intention probability of the game object to travel along each road topology based on the likelihood probability.

[0268] In a possible design, the probability density distribution data set includes: a nonlinear probability density distribution data set, and the nonlinear probability density distribution data set includes the distribution of the probability density of the eigenvector under different speed values ​​and different eigenvector values.

[0269] In one possible design, the intention reasoning module 2402 is specifically used to: obtain the prior intention probability of the obstacle traveling along each road topology; and calculate the intention probability of the game object traveling along each road topology using a Bayesian reasoning algorithm based on the likelihood probability and the prior intention probability.

[0270] In a possible design, the feature reference information is obtained based on each target waypoint on the road topology; or, the feature reference information is obtained based on the road congestion level and / or the aggressiveness of the driving style of the game object.

[0271] In one possible design, the real-time status information of the game object includes at least any one or more of the following: the real-time position information of the game object, the real-time orientation angle of the game object, the real-time speed of the game object, and the real-time acceleration of the game object; the characteristic reference information includes at least any one or more of the following: the reference position information of the game object, the reference orientation angle of the game object, the reference speed of the game object, and the reference acceleration of the game object.

[0272] In a possible design, when there are no other obstacles around the game object except the vehicle, the reference speed of the game object is calculated based on the kinematic constraints of the game object and / or the environmental traffic constraints of the game object.

[0273] In a possible design, the kinematic constraint includes at least one or more of the following: lane topology curvature speed limit, start acceleration constraint, and red light deceleration constraint. The environmental traffic constraint includes at least one or more of the following: road speed limit of the lane where the game object is located, speed constraint of the preceding vehicle in the lane where the game object is located, and speed constraint of obstacles in lanes other than the lane where the game object is located.

[0274] In one possible design, when there is at least one first target obstacle located in front of the game object on the n road topologies and the distance between the game object and the first target obstacle is less than a preset safety distance, the reference speed of the game object is obtained based on the speed of the first target obstacle.

[0275] In a possible design, when there is at least one second target obstacle in the same traveling direction as the game object, and the road topology owned by the at least one second target obstacle does not belong to the n road topologies, the reference speed of the game object is obtained based on the benchmark speed of the at least one second target obstacle, and the benchmark speed is obtained based on a preset method.

[0276] In one possible design, there is at least one third target obstacle in the crossing direction of the game object, and the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than a preset threshold, then the reference speed of the game object is the real-time speed of the game object.

[0277] In one possible design, there is at least one third target obstacle in the crossing direction of the game object, and the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is less than or equal to a preset threshold. Then, the reference speed of the game object is obtained based on the real-time speed of the game object, and the reference speed is less than the real-time speed of the game object.

[0278] In one possible design, the determination module 2405 is specifically used to: use the n intention probabilities as the weights of the n road topologies respectively, calculate the weighted strategy costs corresponding to each game strategy in the n groups of strategy costs, and obtain m total costs; determine the minimum total cost from the m total costs; and determine n target strategy costs corresponding to the minimum total cost.

[0279] In one possible design, the calculation module 2404 is specifically used to: determine various factors of the strategy cost of the m game strategies under each road topology, the various factors of the strategy cost including at least one of safety, comfort, passability, road right, lateral deviation, risk area, and inter-frame association; calculate the factor cost of each factor in each strategy cost; weight the factor cost of each factor in each strategy cost to obtain the strategy costs of the m game strategies under each road topology.

[0280] In one possible design, construction module 2403 is specifically used to: determine the decision upper limit and decision lower limit of the ego vehicle and the game object according to the real-time status information of the ego vehicle and the real-time status information of the game object; obtain the decision strategy of the ego vehicle and the decision strategy of the game object in the decision upper limit and the decision lower limit according to preset rules; combine the decision strategy of the ego vehicle with the decision strategy of the game object to obtain m game strategies of the ego vehicle and the game object, and the m game strategies belong to the sampling game space.

[0281] In one possible design, the acquisition module 2401 is specifically used to: obtain all road topologies within a preset area around the vehicle; obtain obstacle information of at least one obstacle around the vehicle, and calculate the road topology of the at least one obstacle based on the obstacle information and all road topologies; and determine the obstacle to which n road topologies belong from the at least one obstacle as the game object.

[0282] It should be noted that Fig.24 The information interaction, execution process, etc. between the modules in the decision-making device 2400 provided are the same as those in this application. Figure 5 The corresponding method embodiments are based on the same concept. For specific contents, please refer to the description in the method embodiments shown above in this application, which will not be repeated here.

[0283] In addition, the present application also provides a vehicle driving control method, for details, please refer to Fig.25 , the method may include the following steps:

[0284] 2501. Obtain obstacles outside the vehicle.

[0285] First, the obstacle information within a preset area outside the vehicle (such as within a radius of 10m) can be obtained.

[0286] 2502. Determine the decision result of vehicle driving in view of obstacles.

[0287] Afterwards, for the acquired obstacle information, the decision result of the vehicle's driving is determined according to any of the above intelligent driving decision-making methods, that is, the action belonging to the ego vehicle in the action pair formed by the ego vehicle and the game object.

[0288] 2503. Control the movement of the vehicle according to the decision result.

[0289] Finally, the vehicle's movement is controlled according to the decision result.

[0290] Similarly, based on the above corresponding embodiments, in order to better implement the above solutions of the embodiments of the present application, the following also provides related devices for implementing the above solutions. Fig.26 , Fig.26A structural schematic diagram of a vehicle driving control device provided in an embodiment of the present application, the vehicle driving control device 2600 includes: an acquisition module 2601 and a processing module 2602, wherein the acquisition module 2601 is used to acquire obstacles outside the vehicle; the processing module 2602 is used to determine the decision result of the vehicle driving according to any of the above intelligent driving decision-making methods for the acquired obstacles; the processing module 2602 is also used to control the driving of the vehicle according to the decision result.

[0291] It should be noted that Fig.26 The information interaction, execution process, etc. between the modules in the vehicle driving control device 2600 provided are the same as those in this application. Fig.25 The corresponding method embodiments are based on the same concept. For specific contents, please refer to the description in the method embodiments shown above in this application, which will not be repeated here.

[0292] In addition, the present application also provides a vehicle, for details, please refer to Fig. 27 , the vehicle 2700 includes: Fig.26 The vehicle driving control device 2600 and the driving system 2701 , the vehicle driving control device 2600 controls the driving system 2701 .

[0293] The present application also provides a computing device. Fig.28 , Fig.28 Another structural diagram of a computing device provided in an embodiment of the present application, the computing device 2800 may be deployed with Fig.24 , Fig.26 or Fig. 27 The decision-making device 2400, the vehicle driving control device 2600 or each module in the vehicle 2700 described in the corresponding embodiment is used to implement Fig.24 , Fig.26 or Fig. 27In the corresponding embodiment, the functions of the decision-making device 2400, the vehicle driving control device 2600 or the vehicle 2700 are specifically implemented by one or more servers. The computing device 2800 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 2822 and memory 2832, and one or more storage media 2830 (e.g., one or more mass storage devices) storing application programs 2842 or data 2844. Among them, the memory 2832 and the storage medium 2830 may be short-term storage or permanent storage. The program stored in the storage medium 2830 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the computing device 2800. Furthermore, the central processing unit 2822 may be configured to communicate with the storage medium 2830 and execute a series of instruction operations in the storage medium 2830 on the computing device 2800.

[0294] The computing device 2800 may also include one or more power supplies 2826, one or more wired or wireless network interfaces 2850, one or more input and output interfaces 2858, and / or one or more operating systems 2841, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0295] In the embodiment of the present application, the central processor 2822 is used to execute Figure 5 The intelligent driving decision-making method in the corresponding embodiment, or Fig.25 The vehicle driving control method in the corresponding embodiment. For example, the central processor 2822 can be used to: first determine the game object and obtain the real-time state information of the game object, the game object is an obstacle with n road topologies, n≥2; then, according to the real-time state information of the game object, determine the intention probability of the game object traveling along each road topology, obtain n intention probabilities, and construct a sampling game space for the game object according to the real-time state information of the vehicle and the real-time state information of the game object, the sampling game space includes m game strategies, m≥1; then calculate the strategy costs of the m game strategies under each road topology, and obtain n groups of strategy costs, wherein each group of strategy costs includes m strategy costs corresponding to the m game strategies one by one; finally, according to the n intention probabilities and the n groups of strategy costs, determine n target strategy costs whose total costs meet the preset conditions, and determine the decision result of the vehicle according to the target game strategy, the n target strategy costs are the strategy costs of the target game strategy under each road topology, and the target game strategy is one of the above m game strategies.

[0296] It should be noted that the specific manner in which the CPU 2822 performs the above steps is different from that in the present application. Figure 5 or Fig.25 The corresponding method embodiments are based on the same concept, and the technical effects they bring are also the same as those of the above-mentioned embodiments of the present application. For specific contents, please refer to the description in the method embodiments shown above in the present application, and will not be repeated here.

[0297] The present application also provides a computer-readable storage medium in which a program for signal processing is stored. When the program is run on a computer, the computer executes the above-mentioned Figure 5 or Fig.25 The steps described in the illustrated embodiment.

[0298] The computing device provided in the embodiment of the present application may specifically include a chip, and the chip includes: a processing unit and a communication unit. The processing unit may be, for example, a processor, and the communication unit may be, for example, an input / output interface, a pin or a circuit. The processing unit may execute the computer execution instructions stored in the storage unit to enable the chip in the computing device to execute the above Figure 5 or Fig.25 The steps described in the illustrated embodiment.

[0299] Optionally, the storage unit is a storage unit within the chip, such as a register, a cache, etc. The storage unit may also be a storage unit located outside the chip within the wireless access device, such as a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM), etc.

[0300] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0301] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including application-specific integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented with corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, U disk, mobile hard disk, ROM, RAM, disk or optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer or network device, etc.) to execute the methods described in each embodiment of the present application.

[0302] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0303] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, a computer or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a data center that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)), etc.

Claims

1. An intelligent driving decision-making method, characterized in that: include: Determine a game object and obtain real-time status information of the game object, wherein the game object is an obstacle identified by the vehicle and belonging to n road topologies, where n≥2; Determine the intention probability of the game object traveling along each of the road topologies according to the real-time state information of the game object, and obtain n intention probabilities; According to the real-time state information of the vehicle and the real-time state information of the game object, construct a sampling game space for the game object, wherein the sampling game space includes m game strategies, where m≥1; Calculating the strategy costs of the m game strategies under each of the road topologies to obtain n groups of strategy costs, wherein each group of strategy costs includes m strategy costs corresponding one-to-one to the m game strategies; According to the n intention probabilities and the n groups of strategy costs, determine n target strategy costs whose total cost meets a preset condition, the n target strategy costs are strategy costs of a target game strategy under each of the road topologies, and the target game strategy is one of the m game strategies; The decision result of the vehicle is determined according to the target game strategy.

2. The method according to claim 1, characterized in that Determining the intention probability of the game object traveling along each road topology according to the real-time state information of the game object includes: Acquire feature reference information of the game object on each of the road topologies, where the feature reference information is used to describe a baseline state of the game object's intention to travel along each of the road topologies under a current traffic environment; The intention probability of the game object traveling along each of the road topologies is determined according to the real-time state information of the game object and the feature reference information.

3. The method according to claim 2, characterized in that Determining the intention probability of the game object traveling along each road topology according to the real-time state information of the game object and the feature reference information includes: Obtaining a target feature vector according to the real-time state information of the game object and the feature reference information; Determine the likelihood probability of the target feature vector according to a probability density distribution data set constructed in advance, wherein the probability density distribution data set includes the distribution of the probability density of the feature vector under different feature vector values; The intention probability of the game object to travel along each of the road topologies is determined according to the likelihood probability.

4. The method according to claim 3, characterized in that The probability density distribution data set includes: A nonlinear probability density distribution data set, wherein the nonlinear probability density distribution data set includes the distribution of the probability density of the eigenvector under different speed values ​​and different eigenvector values.

5. The method according to claim 3, characterized in that: Determining the intention probability of the game object traveling along each road topology according to the likelihood probability includes: Obtaining a priori probability of the obstacle's intention to travel along each of the road topologies; According to the likelihood probability and the prior intention probability, the intention probability of the game object traveling along each of the road topologies is calculated using a Bayesian inference algorithm.

6. The method according to claim 2, characterized in that The feature reference information is obtained based on the target waypoint on each of the road topologies; or, The feature reference information is obtained based on the road congestion level and / or the aggressiveness of the driving style of the game object.

7. The method according to claim 2, characterized in that: The real-time status information of the game object includes at least one or more of the following: The real-time position information of the game object, the real-time orientation angle of the game object, the real-time speed of the game object, and the real-time acceleration of the game object; The feature reference information includes at least one or more of the following: The reference position information of the game object, the reference orientation angle of the game object, the reference speed of the game object, and the reference acceleration of the game object.

8. The method according to claim 7, characterized in that When there are no other obstacles around the game object except the vehicle, the reference speed of the game object is calculated based on the kinematic constraints of the game object and / or the environmental traffic constraints of the game object.

9. The method according to claim 8, characterized in that The kinematic constraints include at least one or more of the following: Lane topology curvature speed limit, start acceleration constraint, red light deceleration constraint; The environmental traffic constraints include at least one or more of the following: The road speed limit of the lane where the game object is located, the speed constraint of the vehicle ahead in the lane where the game object is located, and the speed constraints of obstacles in lanes other than the lane where the game object is located.

10. The method according to claim 7, characterized in that When there is at least one first target obstacle located in front of the game object on the n road topologies and the distance between the game object and the first target obstacle is less than a preset safety distance, the reference speed of the game object is obtained based on the speed of the first target obstacle.

11. The method according to claim 7, characterized in that When there is at least one second target obstacle in the same traveling direction as the game object, and the road topology owned by the at least one second target obstacle does not belong to the n road topologies, the reference speed of the game object is obtained based on the benchmark speed of the at least one second target obstacle, and the benchmark speed is obtained based on a preset method.

12. The method according to claim 7, characterized in that If there is at least one third target obstacle in the crossing direction of the game object, and the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than a preset threshold, then the reference speed of the game object is the real-time speed of the game object.

13. The method according to claim 7, characterized in that If there is at least one third target obstacle in the crossing direction of the game object, and the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is less than or equal to a preset threshold, then the reference speed of the game object is obtained based on the real-time speed of the game object, and the reference speed is less than the real-time speed of the game object.

14. The method according to any one of claims 1 to 13, characterized in that Determining, according to the n intention probabilities and the n groups of policy costs, n target policy costs whose total costs meet a preset condition comprises: The n intention probabilities are used as the weights of the n road topologies respectively, and the weighted strategy costs corresponding to each game strategy in the n groups of strategy costs are calculated to obtain m total costs; Determine the minimum total cost from the m total costs; Determine n target strategy costs corresponding to the total cost with the minimum value.

15. The method according to any one of claims 1 to 13, characterized in that The calculating of the strategy costs of the m game strategies under each of the road topologies includes: Determine various factors of the strategy cost of the m game strategies under each of the road topologies, wherein the various factors of the strategy cost include at least one of safety, comfort, passability, right of way, lateral deviation, risk area, and inter-frame association; Calculate the factor cost of each factor in each strategy cost; The factor cost of each factor in each strategy cost is weighted to obtain the strategy costs of the m game strategies under each road topology.

16. The method according to any one of claims 1 to 13, characterized in that The step of constructing a sampling game space for the game object according to the real-time state information of the vehicle and the real-time state information of the game object includes: Determine the upper and lower limits of the decision of the self-vehicle and the game object according to the real-time state information of the self-vehicle and the real-time state information of the game object; According to a preset rule, the decision strategy of the vehicle and the decision strategy of the game object are obtained from the decision upper limit and the decision lower limit; The decision strategy of the ego vehicle is combined with the decision strategy of the game object to obtain m game strategies of the ego vehicle and the game object, and the m game strategies belong to the sampled game space.

17. The method according to any one of claims 1 to 13, characterized in that Determining the game object includes: Get all road topologies within the preset area around the vehicle; Obtaining obstacle information of at least one obstacle around the vehicle, and calculating the road topology belonging of the at least one obstacle according to the obstacle information and all the road topologies; An obstacle to which n road topologies belong is determined from the at least one obstacle as the game object.

18. A decision-making device, characterized in that: include: An acquisition module is used to determine a game object and obtain real-time status information of the game object, wherein the game object is an obstacle identified by the vehicle and belonging to n road topologies, where n≥2; An intention reasoning module, used to determine the intention probability of the game object traveling along each of the road topologies according to the real-time state information of the game object, and obtain n intention probabilities; A construction module, used to construct a sampling game space for the game object according to the real-time state information of the vehicle and the real-time state information of the game object, wherein the sampling game space includes m game strategies, where m≥1; A calculation module, used for calculating the strategy costs of the m game strategies under each of the road topologies, and obtaining n groups of strategy costs, wherein each group of strategy costs includes m strategy costs corresponding to the m game strategies one by one; A determination module, configured to determine, according to the n intention probabilities and the n groups of strategy costs, n target strategy costs whose total cost satisfies a preset condition, wherein the n target strategy costs are strategy costs of a target game strategy under each of the road topologies, and the target game strategy is one of the m game strategies; A decision module is used to determine the decision result of the vehicle according to the target game strategy.

19. The device according to claim 18, characterized in that The intention reasoning module is specifically used for: Acquire feature reference information of the game object on each of the road topologies, where the feature reference information is used to describe a baseline state of the game object's intention to travel along each of the road topologies under a current traffic environment; The intention probability of the game object traveling along each of the road topologies is determined according to the real-time state information of the game object and the feature reference information.

20. The device according to claim 19, characterized in that The intention reasoning module is further specifically used for: Obtaining a target feature vector according to the real-time state information of the game object and the feature reference information; Determine the likelihood probability of the target feature vector according to a probability density distribution data set constructed in advance, wherein the probability density distribution data set includes the distribution of the probability density of the feature vector under different feature vector values; The intention probability of the game object to travel along each of the road topologies is determined according to the likelihood probability.

21. The device according to claim 20, characterized in that The probability density distribution data set includes: A nonlinear probability density distribution data set, wherein the nonlinear probability density distribution data set includes the distribution of the probability density of the eigenvector under different speed values ​​and different eigenvector values.

22. The device according to claim 20, characterized in that The intention reasoning module is further specifically used for: Obtaining a priori probability of the obstacle's intention to travel along each of the road topologies; According to the likelihood probability and the prior intention probability, the intention probability of the game object traveling along each of the road topologies is calculated using a Bayesian inference algorithm.

23. The device according to claim 19, characterized in that The feature reference information is obtained based on the target waypoint on each of the road topologies; or, The feature reference information is obtained based on the road congestion level and / or the aggressiveness of the driving style of the game object.

24. The device according to claim 19, characterized in that The real-time status information of the game object includes at least one or more of the following: The real-time position information of the game object, the real-time orientation angle of the game object, the real-time speed of the game object, and the real-time acceleration of the game object; The feature reference information includes at least one or more of the following: The reference position information of the game object, the reference orientation angle of the game object, the reference speed of the game object, and the reference acceleration of the game object.

25. The device according to claim 24, characterized in that When there are no other obstacles around the game object except the vehicle, the reference speed of the game object is calculated based on the kinematic constraints of the game object and / or the environmental traffic constraints of the game object.

26. The device according to claim 25, characterized in that The kinematic constraints include at least one or more of the following: Lane topology curvature speed limit, start acceleration constraint, red light deceleration constraint; The environmental traffic constraints include at least one or more of the following: The road speed limit of the lane where the game object is located, the speed constraint of the vehicle ahead in the lane where the game object is located, and the speed constraints of obstacles in lanes other than the lane where the game object is located.

27. The device according to claim 24, characterized in that When there is at least one first target obstacle located in front of the game object on the n road topologies and the distance between the game object and the first target obstacle is less than a preset safety distance, the reference speed of the game object is obtained based on the speed of the first target obstacle.

28. The device according to claim 24, characterized in that When there is at least one second target obstacle in the same traveling direction as the game object, and the road topology owned by the at least one second target obstacle does not belong to the n road topologies, the reference speed of the game object is obtained based on the benchmark speed of the at least one second target obstacle, and the benchmark speed is obtained based on a preset method.

29. The device according to claim 24, characterized in that If there is at least one third target obstacle in the crossing direction of the game object, and the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is greater than a preset threshold, then the reference speed of the game object is the real-time speed of the game object.

30. The device according to claim 24, characterized in that If there is at least one third target obstacle in the crossing direction of the game object, and the time difference between the game object and the at least one third target obstacle in reaching the predicted collision point is less than or equal to a preset threshold, then the reference speed of the game object is obtained based on the real-time speed of the game object, and the reference speed is less than the real-time speed of the game object.

31. The device according to any one of claims 18 to 30, characterized in that The determination module is specifically used for: The n intention probabilities are used as the weights of the n road topologies respectively, and the weighted strategy costs corresponding to each game strategy in the n groups of strategy costs are calculated to obtain m total costs; Determine the minimum total cost from the m total costs; Determine n target strategy costs corresponding to the total cost with the minimum value.

32. The device according to any one of claims 18 to 30, characterized in that The computing module is specifically used for: Determine various factors of the strategy cost of the m game strategies under each of the road topologies, wherein the various factors of the strategy cost include at least one of safety, comfort, passability, right of way, lateral deviation, risk area, and inter-frame association; Calculate the factor cost of each factor in each strategy cost; The factor cost of each factor in each strategy cost is weighted to obtain the strategy costs of the m game strategies under each road topology.

33. The device according to any one of claims 18 to 30, characterized in that The building blocks are specifically used for: Determine the upper and lower limits of the decision of the self-vehicle and the game object according to the real-time state information of the self-vehicle and the real-time state information of the game object; According to a preset rule, the decision strategy of the vehicle and the decision strategy of the game object are obtained from the decision upper limit and the decision lower limit; The decision strategy of the ego vehicle is combined with the decision strategy of the game object to obtain m game strategies of the ego vehicle and the game object, and the m game strategies belong to the sampled game space.

34. The device according to any one of claims 18 to 30, characterized in that The acquisition module is specifically used for: Get all road topologies within the preset area around the vehicle; Obtaining obstacle information of at least one obstacle around the vehicle, and calculating the road topology belonging of the at least one obstacle according to the obstacle information and all the road topologies; An obstacle to which n road topologies belong is determined from the at least one obstacle as the game object.

35. A vehicle driving control method, characterized in that: include: Obtain obstacles outside the vehicle; With respect to the obstacle, determining a decision result of vehicle driving according to any one of the methods of claims 1-17; The driving of the vehicle is controlled according to the decision result.

36. A vehicle driving control device, characterized in that: include: An acquisition module is used to acquire obstacles outside the vehicle; A processing module, used for determining a decision result of vehicle driving according to any one of the methods of claims 1-17 with respect to the obstacle; The processing module is also used to control the driving of the vehicle according to the decision result.

37. A vehicle, characterized in that: include: The vehicle driving control device and driving system as claimed in claim 36; The vehicle travel control device controls the travel system.

38. A computing device comprising a processor and a memory, wherein the processor is coupled to the memory, wherein: The memory is used to store programs; The processor is used to execute the program in the memory so that the computing device executes the method according to any one of claims 1 to 17, or the computing device executes the method according to claim 35.

39. An intelligent driving system, comprising a sensor system and a processor, characterized in that: The processor is used to execute the method according to any one of claims 1 to 17, or the processor is used to execute the method according to claim 35.

40. A computer-readable storage medium, characterized in that The method comprises a program which, when executed on a computer, causes the computer to execute the method according to any one of claims 1 to 17, or causes the computer to execute the method according to claim 35.

41. A computer program product comprising instructions, characterized in that When it is executed on a computer, it enables the computer to execute the method according to any one of claims 1 to 17, or enables the computer to execute the method according to claim 35.

42. A chip, characterized in that: The chip includes a processor and a data interface, and the processor reads instructions stored in the memory through the data interface to execute the method as described in any one of claims 1-17, or executes the method as described in claim 35.

Images