An automatic driving behavior decision-making method and system based on a multiple attention mechanism and a medium

Abstract

The application relates to an automatic driving behavior decision method and system based on a multiple attention mechanism and a medium, the method comprising: J1. A vehicle travels on a road, real-time static map data information of the road is acquired based on a vehicle-mounted high-precision map, and data information of other obstacle targets and state data information of the vehicle are acquired in real time based on a vehicle-mounted sensing system; J2. Based on the state data information of the vehicle, the data information of the other obstacle targets and the static map data information of the road, the relationship between the vehicle and the other obstacle targets and the map is acquired, the vehicle, the other obstacle targets and the map elements are respectively identified, local features of each are extracted, and an attention mechanism is used to extract interactive features between the vehicle and the other obstacle targets and the map elements. The application can not only obtain more responsible and rich decision results, but also meet the decision expression requirements of complex scenes.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to an autonomous driving behavior decision-making method, system, and medium based on a multi-attention mechanism. Background Technology

[0002] Autonomous driving software systems mainly consist of environmental perception, high-precision positioning, routing and navigation, behavior decision-making, trajectory planning, and control modules. Among these, the behavior decision-making module acts as the brain of autonomous driving, playing a crucial role in the entire system. It receives information from environmental perception about other road users and the surrounding static environment, as well as the vehicle's position information from high-precision positioning and navigation information from the routing and navigation module. In complex scenarios, it makes safe and comfortable decisions and sends these driving behaviors to the downstream trajectory planning module, guiding the vehicle to travel along the planned trajectory.

[0003] In the prior art, a patent (application number: 202111325837.X) discloses a decision-making and planning method and apparatus for autonomous vehicles, including: acquiring parameter information of traffic participants; obtaining the potential field of traffic participants based on the parameter information of traffic participants; constructing a decision sequence tree, which includes multiple decision sequences; constructing a corresponding fusion potential field for each feasible region in the decision sequence based on the potential field of traffic participants; calculating the planning trajectory corresponding to all decision sequences in the time domain based on the fusion potential field and a preset system state equation; evaluating the planning trajectories corresponding to all decision sequences to obtain multiple evaluation results; and sorting the multiple evaluation results to obtain the sorted evaluation results. However, this scheme is limited to the parameter information of traffic participants and does not consider the rich prior information of high-precision maps. The various elements of the map, as well as the topological relationships and navigation information of the map, all have a constraining and guiding effect on autonomous driving behavior decisions. In addition, this scheme only considers the current state of the vehicle, obstacles in the current lane, and obstacles in the target lane, without considering the influence of other traffic participants on the vehicle's behavior. Therefore, its scheme is only suitable for simple, low-interaction driving scenarios where there is little interaction between the vehicle and the environment.

[0004] In the prior art, a patent (application number: 202011560297.9) discloses an autonomous vehicle trajectory prediction method based on a local attention mechanism. This method uses the historical trajectories of surrounding vehicles as input, fully considering the impact of interactions between the autonomous vehicle and neighboring vehicles on the future trajectory of the autonomous vehicle. It constructs the spatial interaction between the autonomous vehicle and neighboring vehicles based on the road geometry and vehicle geometry, then estimates the vehicles with high correlation to the future trajectory of the autonomous vehicle through the local attention mechanism, calculates the correlation between these vehicles and the autonomous vehicle, and uses a weighted sum of correlations to construct the temporal interaction. Finally, it integrates the temporal and spatial interactions between the autonomous vehicle and surrounding vehicles at the current moment, inputting the integrated interaction features into a decoder of a fully connected layer to obtain the trajectory distribution and coordinates of the autonomous vehicle over a future period. However, this scheme only considers the interaction between the autonomous vehicle and other obstacles, failing to utilize the rich information contained in high-precision maps and ignoring the connection between the autonomous vehicle and map information (lanes, intersections, pedestrian crossings, and their topological relationships). This information plays a crucial role in the decision-making of autonomous driving. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides an autonomous driving behavior decision-making method, system and medium based on a multi-attention mechanism, which can not only obtain more complex and richer decision results, but also meet the decision expression needs of complex scenarios.

[0006] To achieve the above and other related objectives, the present invention provides the following technical solution:

[0007] An autonomous driving behavior decision-making method based on a multi-attention mechanism, the method comprising:

[0008] J1. When a vehicle is driving on the road, it acquires static map data of the road in real time based on the vehicle-mounted high-precision map, and acquires data of other obstacles and vehicle status data in real time based on the vehicle-mounted perception system.

[0009] J2. Based on the vehicle's state data and the data of other obstacle targets, obtain the relationship between the vehicle and other obstacle targets, identify the vehicle and other obstacle targets respectively, extract the local features of the vehicle and other obstacle targets, and then use an attention mechanism to extract the interaction features between the vehicle and other obstacle targets. Based on the vehicle's state data and the static map data of the road, obtain the interaction relationship between the vehicle and the map, identify the vehicle and different map elements respectively, extract the local features of the vehicle and the map, and then use an attention mechanism to extract the interaction features between the vehicle and the map elements.

[0010] J3. Input the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion, and output the first driving behavior data information of the vehicle;

[0011] J4. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, the driving behavior of the vehicle is analyzed using lateral decision-making and longitudinal decision-making algorithms, and the second driving behavior data of the vehicle is output.

[0012] J5. Based on the second driving behavior data of the vehicle and the first driving behavior data of the vehicle, construct a weighted analysis function for vehicle decision-making and output the vehicle's decision result data.

[0013] Furthermore, the weight analysis function for constructing vehicle decisions is as follows:

[0014] , ,

[0015] Where Q1 is the weighting function for the vehicle's first driving behavior, Q2 is the weighting function for the vehicle's second driving behavior, m and n are the total number of samples, and S 1i For the first driving behavior data of the i-th vehicle, S 2j For the second driving behavior data of the j-th vehicle, α i Let β be the parameter coefficient for the first driving behavior of the i-th vehicle. j Let be the parameter coefficient of the second driving behavior of the j-th vehicle.

[0016] Furthermore, the constraint condition for constructing the weight analysis function for vehicle decision-making is that the value range of Q1 is (0,1) and the value range of Q2 is (0,1).

[0017] Furthermore, in step J3, the step of inputting the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion includes:

[0018] J31. Based on the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets, establish a binary classification training dataset T.

[0019] T={(X1,Y1),(X2,Y2),...,(X i ,Y i ),...,(X n ,Y n )},

[0020] Among them, X iY represents the interaction features between the i-th vehicle and map elements. i Let n represent the interaction features between the i-th vehicle and other obstacle targets, where n is the total number of samples.

[0021] J32. Based on the training dataset T for the binary classification, establish a feature fusion function f(t).

[0022] ,

[0023] t∈T, G i (t): t⇒{-1, +1}

[0024] Where t is an element of the training dataset for binary classification, δ i These are constant coefficients.

[0025] Furthermore, in step G4, the step of analyzing the vehicle's driving behavior using lateral and longitudinal decision-making algorithms includes:

[0026] G41. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, a piecewise cubic polynomial spline function M is used.

[0027] ,

[0028] Where i = 0, 1, 2, ..., n, represents the number of segments to be divided into n+1 segments, and s i Let a be the mileage of the first point of the i-th segment of the trajectory. x,i b x,i c x,i d x,i a y,i b y,i c y,i and d y,i Using constant parameters of the spline function, the vehicle trajectory is fitted to obtain the vehicle trajectory curve data information;

[0029] G42. Based on the vehicle's trajectory curve data, establish the curvature function k at any point on the vehicle's trajectory curve.

[0030] ,

[0031] in, , , and are the first and second derivatives of x and y with respect to s, respectively;

[0032] G43. Based on the curvature function k at any point on the trajectory curve of the vehicle, analyze the lateral driving behavior of the vehicle and output the lateral driving behavior data information of the vehicle.

[0033] G44. Based on the vehicle's trajectory curve data, perform statistical analysis on the vehicle's trajectory speed curve and acceleration curve, determine whether the vehicle's longitudinal driving behavior is constant speed, acceleration, deceleration, stopping, slowing down, or emergency braking, and output the vehicle's longitudinal driving behavior data.

[0034] G45. Based on the vehicle's lateral driving behavior data information and the vehicle's longitudinal driving behavior data information, couple them together to output the vehicle's second driving behavior data information.

[0035] Furthermore, in step G43, the analysis of the vehicle's lateral driving behavior involves first calculating the angle θ between the start and end points of the future trajectory. If θ is greater than 30°, then the curvature of all points on the future trajectory is traversed. If there exists a point with a curvature k greater than 1 / R... 0 R 0 If the minimum turning radius is given, then the vehicle's lateral driving behavior is a right turn; if there exists a point where the curvature k is less than -1 / R... 0 If the curvature k at a certain point is greater than 1 / R, then the vehicle's lateral driving behavior is a left turn; if both conditions are met, then the vehicle's lateral driving behavior is considered an S-curve; if there exists a point where the curvature k is greater than 1 / R... 1 R 1 Let be the minimum turning radius for a vehicle to make a U-turn; then the vehicle's lateral driving behavior is a right U-turn. If there exists a point where the curvature k is less than -1 / R... 1 If so, the vehicle's lateral driving behavior is a left U-turn.

[0036] Furthermore, the static data information of the road includes the map lane edges and center lines, intersections, pedestrian crossings, and the topological relationships between different map elements; the data information of other obstacle targets includes the target type, size, position coordinates at different times, direction, and speed; and the state data information of the vehicle includes the vehicle's position coordinates, direction, and speed at different times.

[0037] Furthermore, in step J2, the feature function of the interaction features between the vehicle and other obstacle targets is Attention(Q,K,V).

[0038] ,

[0039] Where Q, K, and V are the Query, Key, and Value in the attention mechanism, respectively, dK is the dimension of K, and softmax is the normalized exponential function.

[0040] To achieve the above and other related objectives, the present invention also provides an autonomous driving behavior decision-making system based on a multi-attention mechanism, including a computer device programmed or configured to perform the steps of any of the autonomous driving behavior decision-making methods based on a multi-attention mechanism.

[0041] To achieve the above and other related objectives, the present invention also provides a computer-readable storage medium storing a computer program programmed or configured to execute any of the described autonomous driving behavior decision-making methods based on the multi-attention mechanism.

[0042] The present invention has the following positive effects:

[0043] 1. This invention divides the behavioral decision-making results of autonomous driving into two dimensions: lateral decision-making and vertical decision-making. By combining the two dimensions, more complex and richer decision-making results can be obtained, thereby meeting the decision expression needs of complex scenarios.

[0044] 2. This invention utilizes information about the vehicle's future driving trajectory, combined with a high-precision map and its topological relationships, to generate true values ​​of the vehicle's driving decision-making behavior, facilitating the training and verification of the decision-making model.

[0045] 3. This invention employs multiple attention mechanisms to extract the interaction relationships between the vehicle and other obstacle targets, the interaction relationship between the vehicle and the map, and the relationship between these relationships and the final driving behavior. This enables autonomous driving decisions to better consider the interaction between the vehicle and the surrounding dynamic and static environment, resulting in more intelligent and reliable final decision-making outcomes. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0047] Figure 2 This is a network structure diagram of the autonomous driving behavior decision-making method of the present invention;

[0048] Figure 3 This is a schematic diagram illustrating the vectorized representation of map elements according to the present invention. Detailed Implementation

[0049] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0050] Example 1: As Figure 1 or Figure 2 or Figure 3 As shown, an autonomous driving behavior decision-making method based on a multi-attention mechanism includes:

[0051] J1. When a vehicle is driving on the road, it acquires static map data of the road in real time based on the vehicle-mounted high-precision map, and acquires data of other obstacles and vehicle status data in real time based on the vehicle-mounted perception system.

[0052] J2. Based on the vehicle's state data and the data of other obstacle targets, obtain the relationship between the vehicle and other obstacle targets, identify the vehicle and other obstacle targets respectively, extract the local features of the vehicle and other obstacle targets, and then use an attention mechanism to extract the interaction features between the vehicle and other obstacle targets. Based on the vehicle's state data and the static map data of the road, obtain the interaction relationship between the vehicle and the map, identify the vehicle and different map elements respectively, extract the local features of the vehicle and the map, and then use an attention mechanism to extract the interaction features between the vehicle and the map elements.

[0053] J3. Input the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion, and output the first driving behavior data information of the vehicle;

[0054] J4. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, the driving behavior of the vehicle is analyzed using lateral decision-making and longitudinal decision-making algorithms, and the second driving behavior data of the vehicle is output.

[0055] J5. Based on the second driving behavior data of the vehicle and the first driving behavior data of the vehicle, construct a weighted analysis function for vehicle decision-making and output the vehicle's decision result data.

[0056] In this embodiment, the weight analysis function for constructing vehicle decisions is:

[0057] , ,

[0058] Where Q1 is the weighting function for the vehicle's first driving behavior, Q2 is the weighting function for the vehicle's second driving behavior, m and n are the total number of samples, and S 1i For the first driving behavior data of the i-th vehicle, S 2j For the second driving behavior data of the j-th vehicle, α i Let β be the parameter coefficient for the first driving behavior of the i-th vehicle. j Let be the parameter coefficient of the second driving behavior of the j-th vehicle.

[0059] In this embodiment, the constraint condition for constructing the weight analysis function for vehicle decision-making is that the value range of Q1 is (0,1) and the value range of Q2 is (0,1).

[0060] In this embodiment, step J3, the step of inputting the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion, includes:

[0061] J31. Based on the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets, establish a binary classification training dataset T.

[0062] T={(X1,Y1),(X2,Y2),...,(X i ,Y i ),...,(X n ,Y n )},

[0063] Among them, X i Y represents the interaction features between the i-th vehicle and map elements. i Let n represent the interaction features between the i-th vehicle and other obstacle targets, where n is the total number of samples.

[0064] J32. Based on the training dataset T for the binary classification, establish a feature fusion function f(t).

[0065] ,

[0066] t∈T, G i (t): t⇒{-1, +1}

[0067] Where t is an element of the training dataset for binary classification, δ i These are constant coefficients.

[0068] Example 2: Based on the autonomous driving behavior decision-making method based on the multi-attention mechanism in Example 1, the present invention will be further explained and described below.

[0069] like Figure 1 As shown, an autonomous driving behavior decision-making method based on a multi-attention mechanism includes:

[0070] J1. When a vehicle is driving on the road, it acquires static map data of the road in real time based on the vehicle-mounted high-precision map, and acquires data of other obstacles and vehicle status data in real time based on the vehicle-mounted perception system.

[0071] J2. Based on the vehicle's state data and the data of other obstacle targets, obtain the relationship between the vehicle and other obstacle targets, identify the vehicle and other obstacle targets respectively, extract the local features of the vehicle and other obstacle targets, and then use an attention mechanism to extract the interaction features between the vehicle and other obstacle targets. Based on the vehicle's state data and the static map data of the road, obtain the interaction relationship between the vehicle and the map, identify the vehicle and different map elements respectively, extract the local features of the vehicle and the map, and then use an attention mechanism to extract the interaction features between the vehicle and the map elements.

[0072] J3. Input the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion, and output the first driving behavior data information of the vehicle;

[0073] J4. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, the driving behavior of the vehicle is analyzed using lateral decision-making and longitudinal decision-making algorithms, and the second driving behavior data of the vehicle is output.

[0074] J5. Based on the second driving behavior data of the vehicle and the first driving behavior data of the vehicle, construct a weighted analysis function for vehicle decision-making and output the vehicle's decision result data.

[0075] In this embodiment, step G4, the step of analyzing the vehicle's driving behavior using lateral and longitudinal decision-making algorithms, includes:

[0076] G41. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, a piecewise cubic polynomial spline function M is used.

[0077] ,

[0078] Where i = 0, 1, 2, ..., n, represents the number of segments to be divided into n+1 segments, and s i Let a be the mileage of the first point of the i-th segment of the trajectory. x,i b x,i c x,i d x,i a y,i b y,i c y,i and d y,i Using constant parameters of the spline function, the vehicle trajectory is fitted to obtain the vehicle trajectory curve data information;

[0079] G42. Based on the vehicle's trajectory curve data, establish the curvature function k at any point on the vehicle's trajectory curve.

[0080] ,

[0081] in, , , and are the first and second derivatives of x and y with respect to s, respectively;

[0082] G43. Based on the curvature function k at any point on the trajectory curve of the vehicle, analyze the lateral driving behavior of the vehicle and output the lateral driving behavior data information of the vehicle.

[0083] G44. Based on the vehicle's trajectory curve data, perform statistical analysis on the vehicle's trajectory speed curve and acceleration curve, determine whether the vehicle's longitudinal driving behavior is constant speed, acceleration, deceleration, stopping, slowing down, or emergency braking, and output the vehicle's longitudinal driving behavior data.

[0084] G45. Based on the vehicle's lateral driving behavior data information and the vehicle's longitudinal driving behavior data information, couple them together to output the vehicle's second driving behavior data information.

[0085] In this embodiment, in step G43, the analysis of the vehicle's lateral driving behavior involves first calculating the angle θ between the start and end points of the future trajectory. If θ is greater than 30°, then the curvature of all points on the future trajectory is traversed. If there exists a point with a curvature k greater than 1 / R... 0 R 0 If the minimum turning radius is given, then the vehicle's lateral driving behavior is a right turn; if there exists a point where the curvature k is less than -1 / R... 0 If the curvature k at a certain point is greater than 1 / R, then the vehicle's lateral driving behavior is a left turn; if both conditions are met, then the vehicle's lateral driving behavior is considered an S-curve; if there exists a point where the curvature k is greater than 1 / R... 1 R 1 Let be the minimum turning radius for a vehicle to make a U-turn; then the vehicle's lateral driving behavior is a right U-turn. If there exists a point where the curvature k is less than -1 / R... 1 If so, the vehicle's lateral driving behavior is a left U-turn.

[0086] In this embodiment, the static data information of the road includes the map lane edges and center lines, intersections, pedestrian crossings, and the topological relationships between different map elements. The data information of other obstacle targets includes the target type, size, position coordinates at different times, direction, and speed. The state data information of the vehicle includes the vehicle's position coordinates, direction, and speed at different times.

[0087] In this embodiment, in step J2, the feature function of the interaction features between the vehicle and other obstacle targets is Attention(Q,K,V).

[0088] ,

[0089] Where Q, K, and V are the Query, Key, and Value in the attention mechanism, respectively, dK is the dimension of K, and softmax is the normalized exponential function.

[0090] The present invention provides an autonomous driving behavior decision-making system based on a multi-attention mechanism, including a computer device that is programmed or configured to perform the steps of any of the autonomous driving behavior decision-making methods based on the multi-attention mechanism described above.

[0091] The present invention provides a computer-readable storage medium storing a computer program programmed or configured to perform any of the described autonomous driving behavior decision-making methods based on a multi-attention mechanism.

[0092] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0093] In summary, this invention not only yields more comprehensive and richer decision results, but also meets the decision expression needs of complex scenarios.

[0094] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An autonomous driving behavior decision-making method based on a multi-attention mechanism, characterized in that, The method includes: J1. When a vehicle is driving on the road, it acquires static map data of the road in real time based on the vehicle-mounted high-precision map, and acquires data of other obstacles and vehicle status data in real time based on the vehicle-mounted perception system. J2. Based on the vehicle's state data and the data of other obstacle targets, obtain the interaction relationship between the vehicle and other obstacle targets, identify the vehicle and other obstacle targets respectively, extract the local features of the vehicle and other obstacle targets, and then use an attention mechanism to extract the interaction features between the vehicle and other obstacle targets. Based on the vehicle's state data and the static map data of the road, obtain the interaction relationship between the vehicle and the map, identify the vehicle and different map elements respectively, extract the local features of the vehicle and the map, and then use an attention mechanism to extract the interaction features between the vehicle and the map elements. J3. Input the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion, and output the first driving behavior data information of the vehicle; J4. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, the driving behavior of the vehicle is analyzed using lateral decision-making and longitudinal decision-making algorithms, and the second driving behavior data of the vehicle is output. J5. Based on the second driving behavior data information and the first driving behavior data information of the vehicle, construct a weighted analysis function for vehicle decision-making and output the vehicle's decision result data information; The weighted analysis function for constructing vehicle decisions is: ， ， Where Q1 is the weighting function for the vehicle's first driving behavior, Q2 is the weighting function for the vehicle's second driving behavior, m and n are the total number of samples, and S 1i For the first driving behavior data of the i-th vehicle, S 2j For the second driving behavior data of the j-th vehicle, α i Let β be the parameter coefficient for the first driving behavior of the i-th vehicle. j Let be the parameter coefficient of the second driving behavior of the j-th vehicle.

2. The autonomous driving behavior decision-making method based on a multi-attention mechanism according to claim 1, characterized in that, The constraints of the weight analysis function for constructing vehicle decision-making are that the value range of Q1 is (0,1) and the value range of Q2 is (0,1).

3. The autonomous driving behavior decision-making method based on a multi-attention mechanism according to claim 1, characterized in that, In step J3, the step of inputting the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets into the AdaBoost algorithm for feature fusion includes: J31. Based on the interaction features between the vehicle and map elements and the interaction features between the vehicle and other obstacle targets, establish a binary classification training dataset T. T={(X1,Y1),(X2,Y2),...,(X i ,Y i ),...,(X n ,Y n )}, Among them, X i Y represents the interaction features between the i-th vehicle and map elements. i Let n represent the interaction features between the i-th vehicle and other obstacle targets, where n is the total number of samples. J32. Based on the training dataset T for the binary classification, establish a feature fusion function f(t). ， t∈T,G i (t):t⇒{-1,+1} Where t is an element of the training dataset for binary classification, δ i These are constant coefficients.

4. The autonomous driving behavior decision-making method based on a multi-attention mechanism according to claim 1, characterized in that, In step J4, the step of analyzing the vehicle's driving behavior using lateral and longitudinal decision-making algorithms includes: J41. Based on the static map data of the road, the data of other obstacle targets, and the state data of the vehicle, a piecewise cubic polynomial spline function M is used. ， Where i = 0, 1, 2, ..., n, represents the number of segments to be divided into n+1 segments, and s i Let a be the mileage of the first point of the i-th segment of the trajectory. x,i b x,i c x,i d x,i a y,i b y,i c y,i and d y,i Using constant parameters of the spline function, the vehicle trajectory is fitted to obtain the vehicle trajectory curve data information; J42. Based on the vehicle's trajectory curve data, establish the curvature function k at any point on the vehicle's trajectory curve. ， in, , , and are the first and second derivatives of x and y with respect to s, respectively; J43. Based on the curvature function k at any point on the trajectory curve of the vehicle, analyze the lateral driving behavior of the vehicle and output the lateral driving behavior data information of the vehicle. J44. Based on the vehicle's trajectory curve data, perform statistical analysis on the vehicle's trajectory speed curve and acceleration curve, determine whether the vehicle's longitudinal driving behavior is constant speed, acceleration, deceleration, stopping, slowing down, or emergency braking, and output the vehicle's longitudinal driving behavior data. J45. Based on the vehicle's lateral driving behavior data information and the vehicle's longitudinal driving behavior data information, couple them together to output the vehicle's second driving behavior data information.

5. The autonomous driving behavior decision-making method based on a multi-attention mechanism according to claim 4, characterized in that, In step J43, the analysis of the vehicle's lateral driving behavior involves first calculating the angle θ between the start and end points of the future trajectory. If θ is greater than 30°, then the curvature of all points on the future trajectory is traversed. If there exists a point with a curvature k greater than 1 / R... 0 R 0 If the minimum turning radius is given, then the vehicle's lateral driving behavior is a right turn; if there exists a point where the curvature k is less than -1 / R... 0 If the curvature k at a certain point is greater than 1 / R, then the vehicle's lateral driving behavior is a left turn; if both conditions are met, then the vehicle's lateral driving behavior is considered an S-curve; if there exists a point where the curvature k is greater than 1 / R... 1 R 1 Let be the minimum turning radius for a vehicle to make a U-turn; then the vehicle's lateral driving behavior is a right U-turn. If there exists a point where the curvature k is less than -1 / R... 1 If so, the vehicle's lateral driving behavior is a left U-turn.

6. The autonomous driving behavior decision-making method based on a multi-attention mechanism according to claim 1, characterized in that: The static data information of the road includes the map lane edges and center lines, intersections, pedestrian crossings, and the topological relationships between different map elements. The data information of other obstacle targets includes the target type, size, position coordinates at different times, direction, and speed. The state data information of the vehicle includes the vehicle's position coordinates, direction, and speed at different times.

7. The autonomous driving behavior decision-making method based on a multi-attention mechanism according to claim 1, characterized in that, In step J2, the feature function for the interaction features between the vehicle and other obstacle targets is Attention(Q, K, V). ， Where Q, K, and V are the Query, Key, and Value in the attention mechanism, respectively, and d K K represents the dimension size, and softmax is the normalization exponential function.

8. An autonomous driving behavior decision-making system based on a multi-attention mechanism, comprising a computer device, characterized in that, The computer device is programmed or configured to perform the steps of the autonomous driving behavior decision-making method based on the multi-attention mechanism as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is programmed or configured to perform the autonomous driving behavior decision-making method based on the multi-attention mechanism as described in any one of claims 1 to 7.

Images