Intersection steering decision-making method and decision-making device

Abstract

The invention relates to the technical field of automatic driving, and provides an intersection steering decision-making method and decision-making device. According to the method, target obstacle information of a moving object is obtained, the conflict time of the moving object and a target obstacle is determined, and a revenue function is constructed based on the conflict time, the self-vehicle decision probability of the moving object to the target obstacle and the decision probability of the target obstacle. Solving the revenue function to obtain a first target self-vehicle decision probability when the revenue function reaches a balance point, determining a convergence decision by using the first target self-vehicle decision probability, determining that the decision type of the convergence decision is different from the current decision type of the moving object, performing feasibility judgment and enhancement judgment on decision type conversion operation, according to the method, the intersection turning decision is made by combining the game and the rule, the more accurate decision result can be provided by dynamically adapting to the environment change through the game, the calculated amount is simplified by utilizing the constraint of the rule, the decision efficiency is greatly improved on the basis of not obviously reducing the decision quality, and the decision-making efficiency is improved. And thus, the road condition steering decision making efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a method and device for making decisions on intersection turning. Background Art

[0002] With the development of artificial intelligence technology, autonomous driving technology has become more mature. The current difficulties in realizing autonomous driving functions in the industry are mainly concentrated in various complex road scenarios with strong interactions, including turning scenarios, lane change scenarios, merging scenarios, and diversion scenarios.

[0003] Unprotected turning intersections are particularly difficult to achieve autonomous driving due to the diverse sources of conflict and the unavoidable direct interactions. How to reasonably make autonomous driving decisions at unprotected turning intersections is a technical problem that needs to be solved. Summary of the invention

[0004] In view of this, an embodiment of the present application provides a method and a device for making decisions at intersection turns, so as to solve the problem in the prior art that there is no efficient decision-making solution for unprotected turning intersection scenarios.

[0005] In a first aspect of an embodiment of the present application, a method for making a decision on turning at an intersection is provided, the method being used to provide a decision type when a mobile object turns at an unprotected intersection, comprising:

[0006] Obtain target obstacle information of the moving object, and determine a conflict time based on the moving object information and the target obstacle information, where the conflict time is the time when the moving object collides with the target obstacle;

[0007] A benefit function is constructed based on at least the ego-vehicle decision probability of the mobile object with respect to the target obstacle, the decision probability of the target obstacle, and the conflict time;

[0008] Solve the Nash equilibrium point of the profit function to obtain the first target ego vehicle decision probability of the mobile object for the target obstacle and the target decision probability of the target obstacle;

[0009] A convergence decision is determined based on the first target ego-vehicle decision probability, and in response to determining that the convergence decision is different from the current decision type of the mobile object, a feasibility judgment and an enhancement judgment are performed on the decision type conversion operation, and a decision result is determined based on the judgment result.

[0010] A second aspect of an embodiment of the present application provides a device for making a decision on turning at an intersection, comprising:

[0011] an acquisition module configured to acquire target obstacle information of the moving object, and determine a conflict time based on the moving object information and the target obstacle information, wherein the conflict time is a time when the moving object conflicts with the target obstacle;

[0012] A construction module configured to construct a benefit function based on at least the ego-vehicle decision probability of the mobile object with respect to the target obstacle, the decision probability of the target obstacle, and the conflict time;

[0013] A calculation module is configured to solve a Nash equilibrium point for the benefit function to obtain a first target ego-vehicle decision probability of the mobile object to the target obstacle;

[0014] A determination module is configured to determine a convergence decision based on the first target ego-vehicle decision probability, perform a feasibility judgment and an enhancement judgment on a decision type conversion operation in response to determining that a decision type of the convergence decision is different from a current decision type of the mobile object, and determine a decision result based on the judgment result.

[0015] Compared with the prior art, the embodiments of the present application have the following beneficial effects: the embodiments of the present application obtain the target obstacle information of the mobile object, determine the conflict time between the mobile object and the target obstacle, construct a benefit function based on the conflict time and the self-vehicle decision probability of the mobile object on the target obstacle and the decision probability of the target obstacle, solve the benefit function to obtain the first target self-vehicle decision probability when the equilibrium point is reached, determine the convergence decision using the first target self-vehicle decision probability, determine that the decision type of the convergence decision is different from the current decision type of the mobile object, make feasibility judgments and enhancement judgments on the decision type conversion operation, and determine the decision result based on the judgment result, thereby realizing the combination of game and rules for intersection turning decision, which can not only provide more accurate decision results by dynamically adapting to environmental changes through game, but also simplify the calculation amount by using the constraints of rules, greatly improve the decision efficiency without significantly reducing the decision quality, and thus improve the efficiency of road condition turning decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 It is a scenario diagram of an application scenario of an embodiment of the present application.

[0018] Figure 2 It is a flow chart of a method for making decisions at an intersection provided in an embodiment of the present application.

[0019] Figure 3 It is a flowchart of a method for constructing a profit function based on at least the self-vehicle decision probability of a mobile object to a target obstacle, the decision probability of the target obstacle, and the conflict time provided in an embodiment of the present application.

[0020] Figure 4 It is a schematic diagram of a method for determining conflict time provided in an embodiment of the present application.

[0021] Figure 5 It is a decision diagram drawn in a time-space coordinate system using the method provided in an embodiment of the present application.

[0022] Figure 6 It is a schematic diagram of the evolution phase diagram provided in the embodiment of the present application.

[0023] Figure 7 It is a flowchart of a method for determining the feasibility of a decision type conversion operation provided in an embodiment of the present application.

[0024] Figure 8 It is a flowchart of a method for enhancing judgment on decision type conversion operations provided in an embodiment of the present application.

[0025] Fig. 9 It is a schematic diagram of a system for trajectory planning using the intersection turning decision method provided in an embodiment of the present application.

[0026] Fig.10 It is a schematic diagram of an intersection turning decision device provided in an embodiment of the present application.

[0027] Fig.11 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0028] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0029] A method and device for making decisions at an intersection according to an embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0030] Figure 1 : is a scene diagram of an application scenario of an embodiment of the present application. The application scenario may include a mobile object, obstacle 1, obstacle 2, obstacle 3 and obstacle 4. Among them, the mobile object may be a target object that needs to make a turning decision, such as an unmanned vehicle that needs to make a turning decision. The mobile object and each obstacle are active at an unprotected intersection, and the mobile object needs to complete a turning operation at the unprotected intersection. Among them, an unprotected intersection refers to an intersection without traffic light protection.

[0031] Obstacle 1 is a vehicle coming from the same direction and turning outward, which may conflict with the moving object in conflict zone 1. Obstacle 2 is a vehicle coming from the same direction and turning inward, which may conflict with the moving object in conflict zone 2. Obstacle 3 is a vehicle coming from the opposite direction, which may conflict with the moving object in conflict zone 3. Obstacle 4 is a vehicle coming from the opposite direction, which may conflict with the moving object in conflict zone 4. The moving object as the self-vehicle needs to have the ability to deal with the risk of vehicles coming from any direction in an unprotected scenario. The range of vehicles coming at least includes vehicles coming from behind turning inward, vehicles cutting in from the outside, vehicles coming from the opposite side in the opposite direction, and vehicles coming from the opposite direction and crossing in front. In the embodiment of the present application, the obstacle is a target object that will inevitably conflict with the moving object's vehicle running path.

[0032] It should be noted that the specific types, quantities and combinations of the mobile object, obstacle 1, obstacle 2, obstacle 3 and obstacle 4 can be adjusted according to the actual needs of the application scenario, and the embodiments of the present application are not limited to this.

[0033] As mentioned above, unprotected turning intersections are particularly difficult to achieve autonomous driving due to the diverse sources of conflict and the unavoidable direct interactions. How to enable autonomous vehicles to safely and efficiently pass through intersections without traffic light protection involves complex functions of multiple modules such as perception, decision-making, and planning. Related technology summary: Solutions for unprotected intersections mainly include rule-based solutions and data-driven solutions.

[0034] Among them, the rule-based solution is simple and direct, easy to implement, but lacks flexibility and is difficult to cope with complex or generalized traffic conditions. The data-driven solution is implemented through reinforcement learning, deep learning and other means. The autonomous driving system can continuously try and learn in a simulated environment or massive data to find the optimal decision-making strategy. This method can adaptively learn and improve, but it requires a lot of computing resources and a high-quality simulation environment, and the implementation cost is high. Therefore, how to comprehensively consider cost and accuracy and reasonably make autonomous driving decisions at unprotected turning intersections is a technical problem that needs to be solved.

[0035] In view of this, an embodiment of the present application provides a method for making intersection turning decisions, which obtains the target obstacle information of a mobile object, determines the conflict time between the mobile object and the target obstacle, constructs a benefit function based on the conflict time and the self-vehicle decision probability of the mobile object for the target obstacle and the decision probability of the target obstacle, solves the benefit function to obtain the first target self-vehicle decision probability when the equilibrium point is reached, determines a convergence decision using the first target self-vehicle decision probability, determines that the decision type of the convergence decision is different from the current decision type of the mobile object, performs feasibility judgment and enhancement judgment on the decision type conversion operation, and determines the decision result based on the judgment result, thereby realizing intersection turning decision-making by combining game and rules, which can not only provide more accurate decision results by dynamically adapting to environmental changes through game, but also simplify the calculation amount by using the constraints of rules, greatly improves the decision efficiency without significantly reducing the decision quality, and thereby improves the efficiency of road condition turning decisions.

[0036] Figure 2 It is a flow chart of a method for making decisions at an intersection provided in an embodiment of the present application. Figure 2 The intersection turning decision method can be obtained by Figure 1 The mobile object or server is executed. Figure 2 As shown, the method comprises the following steps:

[0037] In step S201, target obstacle information of a moving object is acquired, and a conflict time is determined based on the moving object information and the target obstacle information.

[0038] The conflict time is the time when the moving object collides with the target obstacle.

[0039] In step S202, a benefit function is constructed based on at least the ego-vehicle decision probability of the mobile object with respect to the target obstacle, the decision probability of the target obstacle, and the conflict time.

[0040] In step S203, the Nash equilibrium point is solved for the benefit function to obtain the first target ego-vehicle decision probability of the mobile object for the target obstacle and the target decision probability of the target obstacle.

[0041] In step S204, a convergence decision is determined based on the first target self-vehicle decision probability, and in response to determining that the decision type of the convergence decision is different from the current decision type of the mobile object, a feasibility judgment and an enhancement judgment are performed on the decision type conversion operation, and a decision result is determined based on the judgment result.

[0042] In some embodiments of the present application, the method may be executed by a mobile object or by a server, wherein the server may communicate with the mobile object, obtain information such as the position and speed of the mobile object, and obtain obstacle information, obtain a turning decision of the mobile object at the intersection based on the obtained information, and send the decision to the mobile object.

[0043] Furthermore, the mobile object may be an unmanned vehicle, an unmanned robot, etc. The intersection may be an intersection without traffic light protection. The turning decision of the mobile object at the intersection includes at least a rushing decision type and a yielding decision type.

[0044] In some embodiments of the present application, the target obstacle information of the moving object may be first obtained, and the target obstacle information may include all obstacles that will conflict with the moving object at the current intersection. Further, the target obstacle information may include the position information, speed information, and moving direction information of the target obstacle. For example, the obstacle information may include the speed, center position, shape such as vehicle length and vehicle width of the obstacle, and may also include the obstacle prediction trajectory obtained from the prediction module. At the same time, the information of the moving object itself may also be obtained, including the position information, speed information, and moving direction information of the moving object.

[0045] The collision time between the mobile object and the target obstacle can be determined using the mobile object information and the target obstacle information. The collision time is the time when the mobile object collides with the target obstacle, that is, the time required for the mobile object to move to the position where it collides with the target obstacle, and the time required for the target obstacle to move to the position where it collides with the mobile object.

[0046] In some embodiments, a benefit function can be constructed based on the determined conflict time, the self-vehicle decision probability of the mobile object with respect to the target obstacle, and the decision probability of the target obstacle itself. In this case, the values ​​of the self-vehicle decision probability of the mobile object with respect to the target obstacle and the decision probability of the target obstacle itself can be undetermined values.

[0047] Next, by solving the Nash equilibrium point of the constructed benefit function, we can obtain the probability of the first target ego vehicle decision of the mobile object for the target obstacle and the target decision probability of the target obstacle, which are used as the optimal value of the game between the parties when determining the turning decision of the mobile object at the intersection. Since only the decision for the mobile object is considered, the first target ego vehicle decision probability is mainly used for decision judgment in the future.

[0048] In certain embodiments of the present application, a convergence decision can be determined based on the first target self-vehicle decision probability, and in response to determining that the decision type of the convergence decision is different from the current decision type of the mobile object, a feasibility judgment and an enhancement judgment are performed on the decision type conversion operation, and the decision result is determined based on the judgment result.

[0049] According to the technical solution provided in the embodiment of the present application, by acquiring the target obstacle information of the mobile object, the conflict time between the mobile object and the target obstacle is determined, and a benefit function is constructed based on the conflict time and the self-vehicle decision probability of the mobile object on the target obstacle and the decision probability of the target obstacle. The benefit function is solved to obtain the first target self-vehicle decision probability when the equilibrium point is reached, and the convergence decision is determined using the first target self-vehicle decision probability. It is determined that the decision type of the convergence decision is different from the current decision type of the mobile object, and the feasibility and enhancement judgment of the decision type conversion operation are performed. The decision result is determined based on the judgment result, thereby realizing the combination of game and rules for intersection turning decision-making. It can not only provide more accurate decision results by dynamically adapting to environmental changes through game, but also simplify the calculation amount by using the constraints of rules, and greatly improve the decision efficiency without significantly reducing the decision quality, thereby improving the efficiency of road condition turning decisions.

[0050] In some embodiments of the present application, the target obstacle may include N target obstacles, where N is a positive integer.

[0051] Figure 3 It is a flowchart of a method for constructing a profit function based on at least the self-vehicle decision probability of a mobile object to a target obstacle, the decision probability of the target obstacle, and the conflict time provided in an embodiment of the present application.

[0052] like Figure 3 As shown, the method comprises the following steps:

[0053] In step S301, a payment matrix is ​​constructed based on the conflict time.

[0054] Among them, the element of the mth row and nth column in the payment matrix is ​​the profit value when the mobile object plays against the i-th target obstacle, wherein m is a positive integer greater than or equal to 1 and less than or equal to 2N, n is a positive integer greater than or equal to 1 and less than or equal to 4, and i belongs to the space {1, ..., N}, that is, i is a positive integer greater than or equal to 1 and less than or equal to N. Further, when m is an odd number, i = (m + 1) / 2, and when m is an even number, i = m / 2.

[0055] In step S302, the first profit function of the game between the i-th target obstacle and the mobile object, and the second profit function of the game between the mobile object and the i-th target obstacle are determined based on the decision probability of the i-th target obstacle, the decision probability of the mobile object and the payoff matrix.

[0056] In step S303, the first profit function and the second profit function are determined as profit functions.

[0057] In some embodiments of the present application, a payment matrix may be constructed first, and then the benefit functions of the moving object and each target obstacle may be constructed based on the payment matrix.

[0058] The payoff matrix can be constructed using the conflict time. The element in the mth row and nth column of the payoff matrix is ​​the payoff value when the mobile object and the i-th target obstacle are in a game. In one example, the conflict time includes the first time, the second time, the third time and the fourth time. The first time is the time required for the i-th target obstacle to reach the starting position of the conflict overlap zone with the mobile object, for example, denoted as thwf i,e The second time is the time required for the mobile object to reach the starting position of the conflict overlap area with the i-th target obstacle, for example, denoted as thwf e,i The third time is the time thwb required for the i-th target obstacle to reach the end position of the conflict overlap area with the moving object i,e The fourth time is the time thwb required for the mobile object to reach the end position of the conflict overlap area with the i-th target obstacle e,i .

[0059] Figure 4 Schematic diagram of a method for determining conflict time provided in an embodiment of the present application. Figure 4 As shown in the figure, if the distance between the moving object and the starting point of the collision overlap area with the obstacle is sf e , the distance from the moving object to the end position of the conflict overlap zone with the obstacle is sb e , and the speed of the moving object is v e , then the second time thwf e =sf e / v e , the fourth time thwb e =sb e / v e .

[0060] Similarly, if the distance between the obstacle and the starting point of the collision overlap zone of the moving object is sf i , the distance from the obstacle to the end position of the collision overlap zone with the moving object is sb i , and the speed of the moving object is v i , then the second time thwf i =sf i / v i , the fourth time thwb i =sb i / v i .

[0061] That is to say, thwf represents the time required for the conflicting parties to reach the starting position of the conflict overlap zone. It measures the time required for the conflicting parties to reach the conflict zone and can be used as a passability representation when constructing the payment matrix later. thwb represents the time required for the conflicting parties to reach the terminal position of the conflict overlap zone. It measures the time required for the conflicting parties to resolve the conflict, that is, to pass the conflict zone. It can be used as a security representation when constructing the payment matrix later.

[0062] We can first define the profit function of each traffic participant, i.e., the target obstacle, in the game with the vehicle, i.e., the moving object, as f(d i )=s e,i (thw i , d i )+c e,i (thw i , d i ), where f(d i ) is the profit function, d i is the decision probability, s e,i (thw i , d i ) for safe returns, you can use thwf i To characterize, c e,i (thw i , d i ) is a passing characterization, through thwb i It can be understood that the calculation method of the profit function of the moving object is the same as the calculation method of the profit function of each target obstacle.

[0063] Furthermore, a payment matrix can be constructed, in which each component of the payment matrix corresponds to a revenue function. For example, the payment matrix can include a payment component in which the mobile object's decision type for the i-th target obstacle is to rush and the i-th target obstacle's decision type for the mobile object is also to rush, a payment component in which the mobile object's decision type for the i-th target obstacle is to rush and the i-th target obstacle's decision type for the mobile object is to yield, a payment component in which the mobile object's decision type for the i-th target obstacle is to yield and the i-th target obstacle's decision type for the mobile object is also to yield, and a payment component in which the mobile object's decision type for the i-th target obstacle is to yield and the i-th target obstacle's decision type for the mobile object is to rush.

[0064] That is, the payment matrix may include 4*2N matrix elements. For example, if the number of columns of the payment matrix is ​​4, the matrix element of the (m+1) / 2th row in the 1st column is the negative of the reciprocal of the first time of the i-th target obstacle minus the third time of the i-th target obstacle, that is, (-1 / thwf i,e -thwb i,e). The matrix element in the (m+1) / 2th row of the 2nd column is the negative of the reciprocal of the second time of the moving object relative to the i-th target obstacle minus the fourth time of the moving object relative to the i-th target obstacle, that is, (-1 / thwf e,i -thwb e,i ). The matrix element in the (m+1) / 2th row of the 3rd column is the inverse of the first time of the ith target obstacle plus the third time of the ith target obstacle, that is, (1 / thwf i,e +thwb i,e ). The matrix element of the 4th column and the (m+1) / 2th row is the reciprocal of the second time of the moving object relative to the i-th target obstacle minus the product of the fourth time of the moving object relative to the i-th target obstacle and the preset coefficient, that is, (1 / thwf e,i -λ*thwb e,i ). Wherein, λ is a hyperparameter used to characterize the time efficiency loss of the vehicle. In one example, the value of λ can be greater than or equal to 0 and less than or equal to 1.

[0065] Furthermore, the matrix element in the m / 2th row of the 1st column of the payoff matrix is ​​the reciprocal of the first time of the ith target obstacle minus the product of the third time of the ith target obstacle and the preset coefficient, that is, (1 / thwf i,e -λ*thwb i,e ). The matrix element in the 2nd column and the m / 2nd row is the sum of the reciprocal of the second time of the moving object relative to the i-th target obstacle and the fourth time of the moving object relative to the i-th target obstacle, that is, (1 / thwf e,i +thwb e,i ). The matrix element in the 3rd column and the m / 2th row is the reciprocal of the first time of the ith target obstacle minus the third time of the ith target obstacle, that is, (1 / thwf i,e -thwb i,e ). The matrix element in the 4th column and the m / 2th row is the reciprocal of the second time of the moving object relative to the i-th target obstacle minus the product of the fourth time of the moving object relative to the i-th target obstacle and the preset coefficient, that is, (1 / thwf e,i -λ*thwb e,i ).

[0066] In one example, the value of each element in the payment matrix can be expressed in the manner shown in Table 1:

[0067] Table 1 The values ​​of each element in the payment matrix

[0068] Oe Oi Rush Give way Rush <![CDATA[-1 / thwf i,e -thwb i,e ]]> <![CDATA[-1 / thwf e,i -thwb e,i ]]> <![CDATA[1 / thwf i,e +throw i,e ]]> <![CDATA[1 / thwf e,i -λ*thwb e,i ]]> Give way <![CDATA[1 / thwf i,e -λ*thwb i,e ]]> <![CDATA[1 / thwf e,i +throw e,i ]]> <![CDATA[1 / thwf i,e -thwb i,e ]]> <![CDATA[1 / thwf e,i -λ*thwb e,i ]]>

[0069] Among them, Oe represents the moving object and Oi represents the i-th target obstacle.

[0070] In this way, a payment matrix composed of the time benefit values ​​in the game between the mobile object and each target obstacle can be constructed. The payment matrix can be further used to construct the first benefit function of the game between the mobile object and each target obstacle, and the second benefit function of the game between each target obstacle and the mobile object, wherein the first benefit function and the second benefit function both represent the time benefit. Furthermore, by solving the equilibrium point of the first benefit function and the second benefit function, the target decision probability of the mobile object relative to each target obstacle and the target decision probability of each target obstacle itself can be obtained when the time benefit equilibrium of all parties is optimal.

[0071] The construction of the first benefit function can be realized in the following way: firstly, the expected value of the rushing when the i-th target obstacle and the moving object are in a game, that is, the rushing benefit function is calculated:

[0072] f i,e (d i,e ) = d i,e ×(d e,i ×(1 / thwf e,i -λ*thwb e,i )+(1-d e,i )×(thwb e,i -λ*thwb e,i ))

[0073] At the same time, the expected value of giving way when the i-th target obstacle and the moving object are in a game, that is, the giving way benefit function is calculated:

[0074] f i,e (1-d i,e )=(1-d i,e )×(d e,i ×(1 / thwf e,i -λ*thwb e,i )+(1-d e,i )×(thwf e,i

[0075] -λ*thwb e,i ))

[0076] Then, the first benefit function is obtained by combining the grabbing benefit function and the giving way benefit function when the i-th target obstacle and the moving object are in a game:

[0077] f i,e (d i ) = f i,e (d i,e )+f i,e (1-d i,e ).

[0078] Similarly, the second benefit function can be constructed in the following way: first, the expected value of the mobile object in the game with the i-th target obstacle, that is, the benefit function of the mobile object, is calculated:

[0079] f e,i (d e,i ) = d e,i ×(d i,e ×(1 / thwf i,e -λ*thwb i,e )+(1-d i,e )×(thwb i,e -λ*thwb i,e ))

[0080] At the same time, the expected value of giving way when the mobile object and the i-th target obstacle are in a game is calculated, that is, the giving way benefit function:

[0081] f e,i (1-d e,i )=(1-d e,i )×(d i,e ×(1 / thwf i,e -λ*thwb i,e )+(1-d i,e )×(thwf i,e

[0082] -λ*thwb i,e ))

[0083] Then, the grabbing benefit function and the giving way benefit function of the mobile object in the game with the i-th target obstacle are combined to obtain the second benefit function:

[0084] f e,i (d e ) = f e,i (d e,i )+f e,i (1-d e,i ).

[0085] In certain real-time methods of the present application, the Nash equilibrium point of the benefit function is solved to obtain the first target self-vehicle decision probability of the mobile object with respect to the target obstacle and the target decision probability of the target obstacle. The first-order derivative of the first benefit function between each target obstacle and the mobile object and the second benefit function of the mobile object with respect to each target obstacle can be set to 0, and the first target self-vehicle decision probability of the mobile object with respect to each target obstacle and the target decision probability of each target obstacle are obtained by solving the problem.

[0086] That is to say, through another as well as Solve each pair of benefit functions one by one to obtain the ego vehicle target decision probability d for each target obstaclee,i and the decision probability d of each target obstacle itself i,e .

[0087] In some embodiments of the present application, the predicted trajectory of the moving object in this frame can be determined based on the first target vehicle decision probability. Furthermore, the current position of each target obstacle can be obtained, and the initial decision result of this frame can be determined based on the relative position relationship between the current position and the predicted trajectory of this frame.

[0088] In one example, the predicted trajectory of the mobile object in this frame can be drawn in the time-space coordinate system, and the current position of the mobile object can be marked in the time-space coordinate system, and then the decision type of the self-vehicle in this frame can be determined based on the relative position of the target obstacle and the predicted trajectory of this frame. For example, if the target obstacle is below the curve of the predicted trajectory of this frame, it can be determined that the decision type of the self-vehicle in this frame is to give way, and conversely, if the target obstacle is above the curve of the predicted trajectory of this frame, it can be determined that the decision type of the self-vehicle in this frame is to grab the way. It can be understood that the correspondence between the relative position and the decision type can also be set to other methods as needed, which is not limited here.

[0089] Figure 5 It is a decision diagram drawn in a time-space coordinate system using the method provided in the embodiment of the present application, wherein the horizontal axis t represents time and the vertical axis s represents position. Figure 5 The trajectory in the upper middle figure is the predicted trajectory of the moving object in the previous frame. Figure 5 The trajectory in the lower middle figure is the predicted trajectory of the moving object in this frame.

[0090] Among them, Figure 5 As shown in the figure above, obstacle 1 is located above the predicted trajectory of the moving object for obstacle 1 in the previous frame, so it can be determined that the decision type of the moving object for obstacle 1 in the previous frame is to rush, and obstacle 2 is located below the predicted trajectory of the moving object for obstacle 2 in the previous frame, so it can be determined that the decision type of the moving object for obstacle 2 in the previous frame is to yield.

[0091] Further, such as Figure 5 As shown in the figure below, obstacle 1 is located below the predicted trajectory of the moving object in this frame for obstacle 1, so the decision type of the moving object for obstacle 1 in this frame can be changed to yield, and obstacle 2 is located above the predicted trajectory of the moving object in this frame for obstacle 2, so the decision type of the moving object for obstacle 2 in this frame can be changed to rush.

[0092] In some embodiments of the present application, the convergence decision is determined based on the first target ego-vehicle decision probability, which may be by plotting the first target ego-vehicle decision probability in an evolution phase diagram, and determining the convergence decision according to the quadrant where the first target ego-vehicle decision probability is located in the evolution phase diagram. The decision types corresponding to the convergence decision include overtaking and yielding.

[0093] Figure 6 Schematic diagram of the evolution phase diagram provided in the embodiment of the present application. Figure 6 As shown in the figure, the horizontal axis represents the decision probability of the ego vehicle, and the vertical axis represents the decision probability of the obstacle. The figure is divided into two quadrants, one is the upper left quadrant, and the other is the lower right quadrant. When the decision probability of the ego vehicle falls into the upper left quadrant, as shown in the figure by the decision probability of the ego vehicle 1, the corresponding convergent decision can be a yield decision type. When the decision probability of the ego vehicle falls into the lower right quadrant, as shown in the figure by the decision probability of the ego vehicle 2, the corresponding convergent decision can be a grab decision type.

[0094] When the converged decision determined according to the first target ego-vehicle decision probability is different from the current decision type of the mobile object, it indicates that the decision type needs to be converted. At this time, the feasibility of the decision type conversion operation can be judged first, and when the judgment result is feasible, the enhanced judgment of the decision type conversion can be performed to ensure the safety and continuity of the mobile object's steering.

[0095] Figure 7 1 is a flow chart of a method for determining the feasibility of a decision type conversion operation provided by an embodiment of the present application. Figure 7 As shown, the method comprises the following steps:

[0096] In step S701, a first predicted trajectory of the moving object converging with the first target decision motion is determined based on the first target ego-vehicle decision probability, and the first predicted trajectory is processed by reverse decision trajectory to obtain a second predicted trajectory.

[0097] In step S702, a distance increment required for the mobile object to perform decision type conversion is calculated according to the first predicted trajectory and the second predicted trajectory.

[0098] In step S703, the distance increment is converted into an acceleration increment, and the acceleration increment is evenly divided into M time steps to obtain the distance increment required for each time step.

[0099] Among them, the M time steps are the time steps corresponding to the M frames before the current frame.

[0100] In step S704, a third predicted trajectory of the moving object is determined according to the distance increment required for each time step in the M time steps.

[0101] In step S705 , in response to determining that the third predicted trajectory satisfies a preset constraint boundary of the moving object, it is determined that the decision type conversion operation is feasible.

[0102] In some embodiments, when the feasibility of the decision type conversion operation is judged, the first predicted trajectory of the mobile object according to the first target convergence decision movement can be determined based on the first target ego vehicle decision probability. The first predicted trajectory can be recorded as T{t0, t1, ..., t M}, where t0, t1 and t M They represent different time steps.

[0103] In one example, the convergence decision can be determined first according to the first target ego vehicle decision probability. For example, the evolution decision function can be used to draw an evolution phase diagram based on the obtained first target ego vehicle decision probability, and then the convergence decision result can be determined according to the position of the first target ego vehicle decision probability in the evolution phase diagram. The convergence decision result can be expressed as D e,i Indicates that its value is 0 or 1.

[0104] If D e,i =0 means that the current decision type is to yield, and the IDM (Intelligent Driver Model) model can be used to determine the first predicted trajectory. e,i =1 represents that the current decision type is to rush in. At this time, the OBVP (Optimal Boundary Value Problem) model can be used to determine the first predicted trajectory.

[0105] Furthermore, the first predicted trajectory can be processed by reverse decision trajectory to obtain a second predicted trajectory after the mobile object changes the decision type, for example, from overtaking to yielding, or from yielding to overtaking. The second predicted trajectory is a reverse decision prediction trajectory, which can be recorded as T rev {t0, t1, ..., t M The reverse decision trajectory processing of the first predicted trajectory can be implemented by using any reverse decision trajectory generation algorithm (GenerateReverseDecisionTrajectory), which is not limited here.

[0106] After determining the first predicted trajectory and the second predicted trajectory, the distance increment required for the mobile object to perform the decision type conversion can be calculated according to the first predicted trajectory and the second predicted trajectory. In one example, the difference between the first predicted trajectory and the second predicted trajectory can be integrated to obtain the distance increment required for the mobile object to perform the decision type conversion.

[0107] That is, after determining the decision probability of the first target ego vehicle, the GenerateDecisionTrajectory algorithm can be called first, and the decision probability of the first target ego vehicle d is input into the algorithm. e,i , to calculate the first predicted trajectory T{t0, t1, ..., t M}. The workflow of the algorithm is: calling the evolutionary decision function EvolutionaryDecision(d e,i ) Draw the evolution phase diagram and get the first target convergence decision D e,i If D e,i =0, T{t0, t1, ..., t M =IDM(O e ); if D e,i =0, T{t0, t1, ..., t M =OBVP(O e ); return T{t0, t1, ..., t M}.

[0108] D e,i and T{t0, t1, ..., t M} Input the decision type conversion difficulty determination algorithm (DecisionTransformDifficulty), in which the GenerateReverseDecisionTrajectory algorithm can be called to obtain the second predicted trajectory T rev {t0, t1, ..., t M}, and then input the first predicted trajectory and the second predicted trajectory into the static transformation distance increment calculation algorithm (StaticTransformNeedS) to calculate the distance increment required for the mobile object to perform decision type conversion. The workflow of the algorithm StaticTransformNeedS is as follows: Get T{t0, t1, ..., t M} and T rev {t0, t1, ..., t M}; Determine the required distance increment as, where S(t) is the moving distance of the moving object in the first predicted trajectory at time step t, S rev (t) is the moving distance of the moving object in the second predicted trajectory at time step t Finally, the value of TransformNeeds is returned, and the value of TransformNeeds is also the output value of DecisionTransformDifficulty.

[0109] In some implementations, after determining the distance increment required for the mobile object to perform the decision type conversion, the distance increment can be converted into an acceleration increment, and the acceleration increment can be evenly divided into M time steps to obtain the distance increment required for each time step. Further, the third predicted trajectory of the mobile object is determined based on the distance increment required for each time step in the M time steps. If it is determined that the third predicted trajectory satisfies the preset constraint boundary of the mobile object, it is determined that the decision type conversion operation is feasible.

[0110] In other words, the distance change required for the decision type conversion can be converted into acceleration change, and the acceleration change can be evenly distributed to the previous M time steps. After averaging, the distance increment required for each time step can be obtained, and the new prediction trajectory of the current frame can be tracked using the distance increment. If the new prediction trajectory of the current frame can meet the boundary constraints and has a solution, it can be considered that the decision type conversion is feasible. Otherwise, it can be considered that the decision type conversion is not feasible and the decision type conversion is not performed.

[0111] Furthermore, after determining that the decision type conversion is feasible, an enhanced judgment can be made on the decision type conversion operation.

[0112] In some embodiments of the present application, the enhanced judgment on the decision type conversion operation may include solving the maximum acceleration value required for the conversion as a conversion difficulty coefficient for measuring the difficulty of the conversion. In other embodiments of the present application, the enhanced judgment on the decision type conversion operation may also include judging whether it is feasible not to perform the decision type conversion operation, and if it is feasible, not performing the decision type conversion to make the moving object travel smoothly.

[0113] Figure 8 is a flow chart of a method for enhancing judgment of decision type conversion operation provided by an embodiment of the present application. Figure 8 As shown, the method comprises the following steps:

[0114] In step S801, the intelligent driver IDM model is used to determine the yielding acceleration, and the two-point boundary value optimal control problem OBVP model is used to determine the grabbing acceleration.

[0115] In step S802, the maximum required incremental acceleration is determined based on the yielding acceleration and the grabbing acceleration.

[0116] In step S803, in response to determining that the maximum required incremental acceleration satisfies a preset constraint boundary of the moving object, the first target ego-vehicle decision probability is updated at least based on the maximum required incremental acceleration to obtain a second target ego-vehicle decision probability.

[0117] In step S804, a decision result is determined according to the second target vehicle decision probability.

[0118] In some implementations, when feasibility judgment is made on the decision type conversion operation, the IDM model can be used to determine the acceleration of giving way first, and the OBVP model can be used to determine the acceleration of overtaking. The specific implementation method of solving the acceleration using the IDM model or the OBVP model is determined by the structure of the IDM model and the OBVP model itself, and is not limited here. In one example, the acceleration of giving way determined by using the IDM model can be recorded as a, and the acceleration of overtaking determined by using the OBVP model can be recorded as a -1 , the maximum required incremental acceleration can be recorded as Δa.

[0119] Furthermore, the maximum required incremental acceleration may be determined based on the determined yielding acceleration and grabbing acceleration.

[0120] After determining the maximum required incremental acceleration, it can be determined whether the maximum required incremental acceleration meets the preset constraint boundary of the moving object. If not, the decision type conversion is not performed. Among them, the preset constraint boundary of the moving object can be set according to actual needs and is not limited here.

[0121] On the other hand, if the maximum required incremental acceleration satisfies the preset constraint boundary of the moving object, the decision type conversion can be considered feasible. At this point, the decision can be further enhanced to measure whether it is necessary to obey the decision conversion.

[0122] When performing enhancement, the first target ego vehicle decision probability may be updated based on at least the maximum required incremental acceleration to obtain a second target ego vehicle decision probability. For example, if the second target ego vehicle decision probability is recorded as but

[0123] Wherein, determining the maximum required incremental acceleration based on the yielding acceleration and the overtaking acceleration may be, in response to determining that the decision type is converted from overtaking to yielding, determining the maximum required incremental acceleration as the difference between the yielding acceleration and the overtaking acceleration. Alternatively, in response to determining that the decision type is converted from yielding to overtaking, determining the maximum required incremental acceleration as the difference between the overtaking acceleration and the yielding acceleration.

[0124] That is to say, if the decision type is converted from overtaking to yielding, in order to keep the trajectory smooth, it is necessary to determine whether the overtaking can be maintained when enhancing the judgment of the decision type conversion. At this time, in the evolution phase diagram, it is necessary to determine whether the first target ego vehicle decision probability can be moved to the right along the x-axis to increase the value of the ego vehicle decision probability and enter the overtaking quadrant. Therefore, the maximum required incremental acceleration can be calculated using the following formula: Δa = a -1 -a, since a is a positive number, a -1 is a complex number, and the Δa calculated here is a negative number. Further, using the formula Updated e,i , the obtained second target vehicle decision probability The value is greater than d e,i The value of is larger.

[0125] On the contrary, if the decision type is converted from yielding to rushing, in order to keep the trajectory smooth, it is necessary to judge whether the yielding can be maintained when enhancing the judgment of the decision type conversion. At this time, in the evolution phase diagram, it is necessary to judge whether the first target ego vehicle decision probability can be moved to the left along the x-axis to reduce the value of the ego vehicle decision probability and enter the yielding quadrant. Therefore, the maximum required incremental acceleration can be calculated using the following formula: Δa = aa -1 , since a is a positive number, a -1 is a complex number, and the Δa calculated here is a positive number. Further, using the formula Updated e,i , the obtained second target vehicle decision probability The value is greater than d e,i The value of is smaller.

[0126] In some implementations, the decision result can be determined based on the second target ego vehicle decision probability, that is, the final decision is re-evolved to convergence based on the second target ego vehicle decision probability to obtain a new converged decision, and the new converged decision is used as the final decision result. The decision result is determined by the value of in the corresponding quadrant of the evolution phase diagram. For example, if the decision probability of the second target vehicle is If the value of falls in the upper left quadrant of the evolution phase diagram, the decision result is to yield. On the contrary, if the decision probability of the second target vehicle is If the value of falls in the lower right quadrant of the evolution phase diagram, the decision result is to rush into the next line.

[0127] On the other hand, the final decision result can also be determined directly based on the comparison between the evolved decision probability of the second target ego vehicle and the decision probability of the corresponding target obstacle. The decision probability d of the target obstacle relative to the moving object i,e to determine the decision result. Then the moving object has a stronger right to overtake, and the decision result can be determined to be overtaking. Then the target obstacle has a stronger right to overtake, and the decision result can be determined to be overtaking.

[0128] Fig. 9 Schematic diagram of a system for trajectory planning using the intersection turning decision method provided by the embodiment of the present application. Fig. 9As shown, the system includes a current frame environment information preprocessing module, a modeling and solving module, a trajectory prediction module and a trajectory planning module. Among them, in the current frame environment information preprocessing module, the environment information and the vehicle state can be obtained, the required obstacles are initially screened to obtain the target obstacles, and the parameters required for the cost design are preprocessed. In the modeling and solving module, an optimization model can be established, including constructing a cost function based on expectations and constructing a benefit function for each agent, and then setting a solving rule to solve the expected decision of each agent to obtain a convergent decision. On the one hand, the convergent decision is provided to the trajectory prediction model to predict the trajectory, and on the other hand, together with the predicted trajectory, the decision conversion difficulty coefficient is calculated in the decision conversion difficulty coefficient solving unit, and the decision is corrected according to the decision conversion difficulty coefficient. The corrected decision can be directly sent to the output decision module for output. On the other hand, the corrected decision can also be input into the trajectory planning module, and the solving model is called based on the corrected decision, and the decision trajectory is obtained by the solving model. Finally, the decision trajectory is recorded.

[0129] The technical solution of the embodiment of the present application introduces timing information and historical decision information on the basis of the original game decision, thereby achieving efficient decision-making for mobile objects such as small unmanned vehicles in complex unprotected turning scenarios.

[0130] The technical solution of the embodiment of the present application can avoid the factorial complexity of multi-agent interaction, reduce the dimension of the game calculation process to a linear complexity problem, reduce the calculation complexity, and thus can realize the interaction problem of multiple obstacles in complex scenes with less computing power, thereby reducing the decision-making cost.

[0131] Furthermore, the technical solution of the embodiment of the present application can better balance the stability and rationality of the vehicle's decision-making, especially for decision-making jump scenarios, it can ensure the data security of decision conversion and improve the decision-making effect.

[0132] At the same time, the embodiment of the present application provides a solution based on a mixture of game and rules to deal with the problem of passing through unprotected turning intersections. It can combine the advantages of the game solution to dynamically adapt to environmental changes and the efficient calculation of the planning solution, efficiently generate stable decisions for unprotected turning intersections, and improve the decision-making efficiency of unmanned mobile objects when turning on road conditions.

[0133] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described one by one here.

[0134] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0135] Fig.10 Schematic diagram of a road intersection turning decision device provided in an embodiment of the present application. Fig.10 As shown, the intersection turning decision device includes:

[0136] The acquisition module 1001 is configured to acquire target obstacle information of the moving object, and determine the conflict time based on the moving object information and the target obstacle information, where the conflict time is the time when the moving object conflicts with the target obstacle.

[0137] The construction module 1002 is configured to construct a benefit function based on at least the ego-vehicle decision probability of the mobile object to the target obstacle, the decision probability of the target obstacle, and the conflict time.

[0138] The calculation module 1003 is configured to solve the Nash equilibrium point of the benefit function to obtain the first target ego-vehicle decision probability of the mobile object to the target obstacle.

[0139] The determination module 1004 is configured to determine a convergence decision based on the first target ego-vehicle decision probability, perform a feasibility judgment and an enhancement judgment on the decision type conversion operation in response to determining that the decision type of the convergence decision is different from the current decision type of the mobile object, and determine a decision result based on the judgment result.

[0140] According to the technical solution provided in the embodiment of the present application, by acquiring the target obstacle information of the mobile object, the conflict time between the mobile object and the target obstacle is determined, and a benefit function is constructed based on the conflict time and the self-vehicle decision probability of the mobile object on the target obstacle and the decision probability of the target obstacle. The benefit function is solved to obtain the first target self-vehicle decision probability when the equilibrium point is reached, and the convergence decision is determined using the first target self-vehicle decision probability. It is determined that the decision type of the convergence decision is different from the current decision type of the mobile object, and the feasibility and enhancement judgment of the decision type conversion operation are performed. The decision result is determined based on the judgment result, thereby realizing the combination of game and rules for intersection turning decision-making. It can not only provide more accurate decision results by dynamically adapting to environmental changes through game, but also simplify the calculation amount by using the constraints of rules, and greatly improve the decision efficiency without significantly reducing the decision quality, thereby improving the efficiency of road condition turning decisions.

[0141] In some embodiments, the target obstacle includes N target obstacles, where N is a positive integer; a profit function is constructed based at least on the self-vehicle decision probability of the mobile object for the target obstacle, the decision probability of the target obstacle, and the conflict time, including: constructing a payment matrix based on the conflict time, the element in the mth row and nth column of the payment matrix is ​​the profit value when the mobile object plays a game with the i-th target obstacle, wherein m is a positive integer greater than or equal to 1 and less than or equal to 2N, n is a positive integer greater than or equal to 1 and less than or equal to 4, and i belongs to the space {1,...,N}; based on the decision probability of the i-th target obstacle, the decision probability of the mobile object, and the payment matrix, a first profit function of the game between the i-th target obstacle and the mobile object and a second profit function of the game between the mobile object and the i-th target obstacle are determined respectively; the first profit function and the second profit function are determined as the profit function.

[0142] In some embodiments, the conflict time includes a first time, a second time, a third time and a fourth time; wherein the first time is the time required for the target obstacle to reach the starting position of the conflict overlap zone with the moving object; the second time is the time required for the moving object to reach the starting position of the conflict overlap zone with the target obstacle; the third time is the time required for the target obstacle to reach the end position of the conflict overlap zone with the moving object; and the fourth time is the time required for the moving object to reach the end position of the conflict overlap zone with the target obstacle.

[0143] In some embodiments, in response to determining that n is 1, m is an odd value, i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the negative of the reciprocal of the first time of the ith target obstacle minus the third time of the ith target obstacle; in response to determining that n is 2, m is an odd value, i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the negative of the reciprocal of the second time of the mobile object relative to the ith target obstacle minus the fourth time of the mobile object relative to the ith target obstacle; in response to determining that n is 3, m is an odd value, i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the reciprocal of the first time of the ith target obstacle plus the third time of the ith target obstacle; in response to determining that n is 4, m is an odd value, i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the reciprocal of the second time of the mobile object relative to the ith target obstacle minus the fourth time of the mobile object relative to the ith target obstacle. In response to determining that n is 1, m is an even value, i is equal to m divided by 2, the value of the element in the mth row and nth column in the payment matrix is ​​the reciprocal of the first time of the ith target obstacle minus the product of the third time of the ith target obstacle and the preset coefficient; in response to determining that n is 2, m is an even value, i is equal to m divided by 2, the value of the element in the mth row and nth column in the payment matrix is ​​the sum of the reciprocal of the second time of the mobile object relative to the ith target obstacle and the fourth time of the mobile object relative to the ith target obstacle; in response to determining that n is 3, m is an even value, i is equal to m divided by 2, the value of the element in the mth row and nth column in the payment matrix is ​​the reciprocal of the first time of the ith target obstacle minus the third time of the ith target obstacle; in response to determining that n is 4, m is an even value, i is equal to m divided by 2, the value of the element in the mth row and nth column in the payment matrix is ​​the reciprocal of the second time of the mobile object relative to the ith target obstacle minus the product of the fourth time of the mobile object relative to the ith target obstacle and the preset coefficient.

[0144] In some embodiments, the Nash equilibrium point of the benefit function is solved to obtain the first target self-vehicle decision probability of the mobile object with respect to the target obstacle. The steps are as follows: the first-order derivatives of the first benefit function between each target obstacle and the mobile object and the second benefit function of the mobile object with respect to each target obstacle are set to 0, and the first target self-vehicle decision probability of the mobile object with respect to each target obstacle is obtained.

[0145] In some embodiments, the feasibility of the decision type conversion operation is judged, including: determining a first predicted trajectory of the mobile object according to the first target convergence decision movement based on the first target ego-vehicle decision probability, performing inverse decision trajectory processing on the first predicted trajectory to obtain a second predicted trajectory; calculating the distance increment required for the mobile object to perform the decision type conversion according to the first predicted trajectory and the second predicted trajectory; converting the distance increment into an acceleration increment, and evenly dividing the acceleration increment into M time steps to obtain the distance increment required for each time step, wherein the M time steps are time steps corresponding to M frames before the current frame; determining a third predicted trajectory of the mobile object according to the distance increment required for each time step in the M time steps; and determining that the decision type conversion operation is feasible in response to determining that the third predicted trajectory satisfies a preset constraint boundary of the mobile object.

[0146] In some embodiments, determining a first predicted trajectory of a mobile object moving according to a first target convergence decision includes: determining a decision type of a convergence decision; in response to determining that the decision type of the convergence decision is to give way, using an intelligent driver IDM model to determine the first predicted trajectory; or in response to determining that the decision type of the convergence decision is to cut in, using a two-point boundary value optimal control problem OBVP model to determine the first predicted trajectory.

[0147] In some embodiments, enhanced judgment is performed on the decision type conversion operation, including: using the intelligent driver IDM model to determine the yielding acceleration, and using the two-point boundary value optimal control problem OBVP model to determine the overtaking acceleration; determining the maximum required incremental acceleration based on the yielding acceleration and the overtaking acceleration; in response to determining that the maximum required incremental acceleration satisfies the preset constraint boundary of the moving object, updating the first target ego-vehicle decision probability based on at least the maximum required incremental acceleration to obtain a second target ego-vehicle decision probability; determining the decision result according to the second target ego-vehicle decision probability.

[0148] In some embodiments, the maximum required incremental acceleration is determined based on the yielding acceleration and the overtaking acceleration, including: in response to determining that the decision type is converted from overtaking to yielding, determining the maximum required incremental acceleration as the difference between the yielding acceleration and the overtaking acceleration; or in response to determining that the decision type is converted from yielding to overtaking, determining the maximum required incremental acceleration as the difference between the overtaking acceleration and the yielding acceleration; updating the first target self-vehicle decision probability based on at least the maximum required incremental acceleration to obtain the second target self-vehicle decision probability, including: normalizing the maximum required incremental acceleration to obtain the normalized maximum required incremental acceleration; subtracting the normalized maximum required incremental acceleration from the first target self-vehicle decision probability to obtain the second target self-vehicle decision probability.

[0149] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0150] Fig.11 Schematic diagram of an electronic device provided in an embodiment of the present application. Fig.11 As shown, the electronic device 11 of this embodiment includes: a processor 1101, a memory 1102, and a computer program 1103 stored in the memory 1102 and executable on the processor 1101. When the processor 1101 executes the computer program 1103, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 1101 executes the computer program 1103, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0151] The electronic device 11 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 11 may include, but is not limited to, a processor 1101 and a memory 1102. Those skilled in the art will appreciate that Fig.11 The electronic device 11 is merely an example and does not limit the electronic device 11 , and may include more or fewer components than those shown in the figure, or different components.

[0152] Processor 1101 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0153] The memory 1102 may be an internal storage unit of the electronic device 11, for example, a hard disk or memory of the electronic device 11. The memory 1102 may also be an external storage device of the electronic device 11, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (FlashCard), etc. equipped on the electronic device 11. The memory 1102 may also include both an internal storage unit of the electronic device 11 and an external storage device. The memory 1102 is used to store computer programs and other programs and data required by the electronic device.

[0154] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0155] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, which may be in source code form, object code form, executable file or some intermediate form. Computer-readable media may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0156] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for making decisions at intersection turns, characterized in that: The method is used to provide a decision type when a mobile object turns at an unprotected intersection; The method comprises: Acquire target obstacle information of the mobile object, and determine a conflict time based on the mobile object information and the target obstacle information, wherein the conflict time is the time when the mobile object conflicts with the target obstacle; constructing a benefit function based at least on the ego-vehicle decision probability of the mobile object with respect to the target obstacle, the decision probability of the target obstacle, and the conflict time; Solving the Nash equilibrium point for the profit function to obtain the first target ego-vehicle decision probability of the mobile object for the target obstacle; A convergence decision is determined based on the first target ego-vehicle decision probability, and in response to determining that a decision type of the convergence decision is different from a current decision type of the mobile object, a feasibility judgment and an enhancement judgment are performed on a decision type conversion operation, and a decision result is determined based on the judgment result.

2. The method according to claim 1, characterized in that: The target obstacles include N target obstacles, where N is a positive integer; The constructing of a benefit function based at least on the ego-vehicle decision probability of the mobile object with respect to the target obstacle, the decision probability of the target obstacle, and the conflict time comprises: A payment matrix is ​​constructed based on the conflict time, wherein the element in the mth row and the nth column of the payment matrix is ​​the profit value when the mobile object negotiates with the i-th target obstacle, wherein m is a positive integer greater than or equal to 1 and less than or equal to 2N, n is a positive integer greater than or equal to 1 and less than or equal to 4, and i belongs to the space {1, ..., N}; Determine a first profit function of the game between the i-th target obstacle and the mobile object, and a second profit function of the game between the mobile object and the i-th target obstacle based on the decision probability of the i-th target obstacle, the decision probability of the mobile object, and the payment matrix; The first profit function and the second profit function are determined as the profit function.

3. The method according to claim 1 or 2, characterized in that: The conflicting time includes a first time, a second time, a third time and a fourth time; The first time is the time required for the target obstacle to reach the starting position of the conflict overlap zone with the moving object; The second time is the time required for the mobile object to reach the starting position of the conflicting overlapping area with the target obstacle; The third time is the time required for the target obstacle to reach the end position of the conflicting overlapping area with the moving object; The fourth time is the time required for the mobile object to reach the end position of the conflicting and overlapping area with the target obstacle.

4. The method according to claim 3, characterized in that In response to determining that n is 1, m is an odd value, and i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payoff matrix is ​​the negative of the reciprocal of the first time of the i-th target obstacle minus the third time of the i-th target obstacle; In response to determining that n is 2, m is an odd value, and i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the negative of the reciprocal of the second time of the mobile object relative to the i-th target obstacle minus the fourth time of the mobile object relative to the i-th target obstacle; In response to determining that n is 3, m is an odd value, and i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the reciprocal of the first time of the i-th target obstacle plus the third time of the i-th target obstacle; In response to determining that n is 4, m is an odd value, and i is equal to the sum of m plus 1 divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the reciprocal of the second time of the moving object relative to the i-th target obstacle minus the product of the fourth time of the moving object relative to the i-th target obstacle and a preset coefficient; In response to determining that n is 1, m is an even value, and i is equal to m divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the reciprocal of the first time of the i-th target obstacle minus the product of the third time of the i-th target obstacle and a preset coefficient; In response to determining that n is 2, m is an even value, and i is equal to m divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the sum of the reciprocal of the second time of the moving object relative to the i-th target obstacle and the fourth time of the moving object relative to the i-th target obstacle; In response to determining that n is 3, m is an even value, and i is equal to m divided by 2, the value of the element in the mth row and nth column of the payment matrix is ​​the reciprocal of the first time of the i-th target obstacle minus the third time of the i-th target obstacle; In response to determining that n is 4, m is an even value, and i is equal to m divided by 2, the value of the element in the mth row and nth column in the payment matrix is ​​the reciprocal of the second time of the moving object relative to the i-th target obstacle minus the product of the fourth time of the moving object relative to the i-th target obstacle and a preset coefficient.

5. The method according to claim 2, characterized in that: The step of solving the Nash equilibrium point of the benefit function to obtain the first target self-vehicle decision probability of the mobile object with respect to the target obstacle is to set the first-order derivative of the first benefit function between each target obstacle and the mobile object and the second benefit function of the mobile object with respect to each target obstacle to 0, and solve to obtain the first target self-vehicle decision probability of the mobile object with respect to each target obstacle.

6. The method according to claim 1, characterized in that The feasibility judgment of the decision type conversion operation includes: Determine a first predicted trajectory of the mobile object's convergence decision motion according to the first target based on the first target ego-vehicle decision probability, and perform inverse decision trajectory processing on the first predicted trajectory to obtain a second predicted trajectory; Calculating the distance increment required for the mobile object to perform decision type conversion according to the first predicted trajectory and the second predicted trajectory; Convert the distance increment into an acceleration increment, and evenly divide the acceleration increment into M time steps to obtain the distance increment required for each time step, wherein the M time steps are time steps corresponding to M frames before the current frame; Determining a third predicted trajectory of the mobile object according to a distance increment required for each time step in the M time steps; In response to determining that the third predicted trajectory satisfies a preset constraint boundary of the moving object, it is determined that the decision type conversion operation is feasible.

7. The method according to claim 6, characterized in that Determining a first predicted trajectory of the moving object according to the first target convergence decision movement includes: determining a decision type of the convergence decision; In response to determining that the decision type of the convergence decision is yielding, determining the first predicted trajectory using an intelligent driver IDM model; or In response to determining that the decision type of the convergence decision is rushing, the first predicted trajectory is determined using a two-point boundary value optimal control problem (OBVP) model.

8. The method according to claim 6, characterized in that Enhanced judgment of decision type conversion operations, including: The intelligent driver IDM model is used to determine the acceleration of giving way, and the two-point boundary value optimal control problem OBVP model is used to determine the acceleration of grabbing. Determine the maximum required incremental acceleration based on the yielding acceleration and the rushing acceleration; In response to determining that the maximum required incremental acceleration satisfies a preset constraint boundary of the mobile object, updating the first target ego-vehicle decision probability based at least on the maximum required incremental acceleration to obtain a second target ego-vehicle decision probability; A decision result is determined according to the second target vehicle decision probability.

9. The method according to claim 8, characterized in that The determining of the maximum required incremental acceleration based on the yielding acceleration and the grabbing acceleration includes: In response to determining that the decision type is converted from overtaking to yielding, determining the maximum required incremental acceleration as the difference between the yielding acceleration and the overtaking acceleration; or In response to determining that the decision type is converted from yielding to stealing, determining the maximum required incremental acceleration as a difference between the stealing acceleration and the yielding acceleration; The updating of the first target ego-vehicle decision probability based at least on the maximum required incremental acceleration to obtain a second target ego-vehicle decision probability includes: Normalizing the maximum required incremental acceleration to obtain a normalized maximum required incremental acceleration; The normalized maximum required incremental acceleration is subtracted from the first target ego-vehicle decision probability to obtain a second target ego-vehicle decision probability.

10. A device for making decisions at intersections, characterized in that: include: an acquisition module, configured to acquire target obstacle information of a moving object, and determine a conflict time based on the moving object information and the target obstacle information, wherein the conflict time is a time when the moving object conflicts with the target obstacle; A construction module configured to construct a benefit function based at least on the ego-vehicle decision probability of the mobile object with respect to the target obstacle, the decision probability of the target obstacle, and the conflict time; A calculation module is configured to solve the Nash equilibrium point for the profit function to obtain a first target ego-vehicle decision probability of the mobile object with respect to the target obstacle; A determination module is configured to determine a convergence decision based on the first target self-vehicle decision probability, perform a feasibility judgment and an enhancement judgment on a decision type conversion operation in response to determining that a decision type of the convergence decision is different from a current decision type of the mobile object, and determine a decision result based on the judgment result.

Images