Automatic driving lane selection control method and device and vehicle

Abstract

The invention discloses an automatic driving lane selection control method and device and a vehicle. The method comprises the steps that information of a road in front of the vehicle is collected and positioned; if it is determined that the vehicle travels to the node associated road according to the information of the road in front of the vehicle and the positioning information, the driving state parameters and lane pointer data of the current vehicle are obtained, and the driving state parameters include the steering wheel angle, the vehicle angular velocity and the vehicle linear velocity; the lane pointer data comprises a current transverse distance between the vehicle and a lane center line of a nearest lane, a lane topology and a virtual lane link sequence in front of the vehicle; inputting the driving state parameter, the lane pointer data, the node associated road, a preset wheelbase of the vehicle and a steering transmission ratio of the vehicle into a trained comprehensive cost determination model to obtain a comprehensive cost; and selecting a target lane with the minimum cost from the lane topology link alternative lanes according to the comprehensive cost, and controlling the operation of the current vehicle.

Description

Technical Field

[0001] The present invention relates to the field of automotive electronics technology, and in particular to an automatic driving lane selection control method, device, and vehicle. Background Art

[0002] When autonomous driving needs to enter a merging or diverging intersection without navigation, in order to ensure vehicle safety, comfort and reliability, it is necessary to select the optimal lane and perform lateral and vertical control based on the perceived environmental information and the vehicle's status.

[0003] Traditional intersection navigation methods utilize traffic lights and Vulnerable Road User (VRU) predictions to achieve longitudinal control at intersections. However, in practice, this longitudinal control method fails to ensure smooth intersection navigation, especially at complex intersections, and is not universally applicable. Current control methods still face technical challenges in ensuring the safety and reliability of intelligent driving technology at urban intersections. Summary of the Invention

[0004] The embodiments of the present application provide an automatic driving lane selection control method, device and vehicle, which can select a reasonable opposite lane based on multiple parameters such as the vehicle's own driving status and external environmental factors, thereby achieving comfortable and reliable passage through intersections.

[0005] In a first aspect, the present invention provides the following technical solutions through an embodiment of the present invention:

[0006] An automatic driving lane selection control method includes: collecting road information and positioning information in front of the vehicle;

[0007] If the vehicle is determined to be traveling on a node-associated road based on the vehicle's forward road information and positioning information, current vehicle driving state parameters and lane pointer data are obtained, wherein the driving state parameters include a steering wheel angle, a vehicle angular velocity, and a vehicle linear velocity. The lane pointer data includes the vehicle's current lateral distance from the centerline of the nearest lane, lane topology, and a virtual lane link sequence in front of the vehicle. The virtual lane link sequence includes a connection relationship between different lanes in front of the vehicle and a sequence of feasible paths. The driving state parameters, lane pointer data, the node-associated road, a preset vehicle wheelbase, and a vehicle steering ratio are input into a trained comprehensive cost determination model to obtain a comprehensive cost. The comprehensive cost determination model is obtained by inputting multiple training samples into a neural network for training, wherein each training sample includes historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbases, historical vehicle steering ratios, and corresponding historical standard comprehensive costs. The comprehensive cost reflects the costs of all lanes in the linked candidate lanes in the lane topology. Based on the comprehensive cost, a target lane with the lowest cost is selected from the linked candidate lanes in the lane topology to control the operation of the current vehicle.

[0008] Preferably, before collecting the road information and positioning information ahead of the vehicle, the method further includes: obtaining training samples, the training samples including multiple sets of driving state parameters and lane pointer data of the vehicle under each road structure when the vehicle travels to different node-associated roads, and the comprehensive cost of the alternative lanes of the vehicle's lane topology link under each set of driving state parameters and lane pointer data; based on the multiple training samples, training a pre-constructed neural network to obtain a trained comprehensive cost determination model.

[0009] Preferably, the comprehensive cost is obtained through the following steps: determining the intersection lane selection cost based on the node-associated road, driving status parameters and lane pointer data of the vehicle, the preset vehicle wheelbase and the vehicle's steering transmission ratio, wherein the intersection lane selection cost includes a lateral sub-cost, a steering wheel angle sub-cost and a consistency sub-cost; and obtaining the comprehensive cost based on the lateral sub-cost, the steering wheel angle sub-cost and the consistency sub-cost.

[0010] Preferably, determining the lateral sub-cost includes: substituting the steering wheel angle, the ego-vehicle angular velocity, the ego-vehicle linear velocity, the wheelbase, and the steering gear ratio into a preset motion model formula to obtain a predicted lateral distance of the current vehicle after a preset time period; determining the maximum curvature and the minimum curvature on the lane link using a three-point method based on the virtual lane link sequence, and then obtaining a curvature difference based on the maximum curvature and the minimum curvature; and obtaining a lateral sub-cost based on the current lateral distance, the predicted lateral distance, and the curvature difference.

[0011] Preferably, the determination of the steering wheel angle sub-cost and the consistency sub-cost includes: determining the preview point of the current vehicle based on the node-associated road; determining the predicted steering wheel angle required for the current vehicle to track the preview point on the center line of the lane using a pure tracking method based on the preview point; obtaining the steering wheel angle sub-cost based on the deviation between the predicted steering wheel angle and the steering wheel angle; and obtaining the consistency sub-cost based on the difference between the predicted steering wheel angle at the current moment and the predicted steering wheel angle at the previous moment.

[0012] Preferably, determining the preview point of the current vehicle based on the node-associated road includes: traversing each lane in the virtual lane link sequence to obtain the projection distance of the vehicle's current position on each lane in the virtual lane link sequence; adding the set normal viewing distance to the projection distance to obtain the target point at the normal viewing distance; adding the set near viewing distance to the projection distance to obtain the target point at the near viewing distance; and determining the preview point of the current vehicle based on the target point at the normal viewing distance, the target point at the near viewing distance, and the node-associated road.

[0013] Preferably, if the node-associated road is an intersection scene, determining the preview point of the current vehicle based on the target point at normal sight distance, the target point at near sight distance and the node-associated road includes: preliminarily setting the preview point as the end point of the center line of the current lane, and calculating the distance difference between the preview point and the projection distance; if the distance difference is less than the set near sight distance, determining the target point at the near sight distance as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance, determining the target point at normal sight distance as the preview point of the current vehicle.

[0014] Preferably, if the node-associated road is a diversion scenario, the preview point of the current vehicle is determined based on the target point at normal sight distance, the target point at near sight distance and the node-associated road, including: if the current vehicle meets the diversion condition, obtaining the diversion starting point; preliminarily setting the preview point to the corresponding point on the center line of the diversion lane, and determining the distance difference between the preview point and the diversion starting point; if the distance difference is less than the set near sight distance, determining the target point at the near sight distance as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance, determining the target point at normal sight distance as the preview point of the current vehicle.

[0015] In a second aspect, the present invention provides the following technical solution through an embodiment of the present invention:

[0016] An automatic driving lane selection control device, comprising:

[0017] A collection module is used to collect road information and positioning information in front of the vehicle;

[0018] an acquisition module, configured to, if it is determined based on the vehicle's forward road information and positioning information that the vehicle is traveling on a node-associated road, obtain current vehicle driving state parameters and lane pointer data, wherein the driving state parameters include a steering wheel angle, a vehicle angular velocity, and a vehicle linear velocity; and the lane pointer data includes a current lateral distance from the vehicle to the centerline of the nearest lane, lane topology, and a virtual lane connection sequence in front of the vehicle, wherein the virtual lane connection sequence includes a connection relationship between different lanes in front of the vehicle and a feasible path sequence;

[0019] a cost determination module that inputs the driving state parameters, the lane pointer data, the node-associated roads, a preset vehicle wheelbase, and a vehicle steering ratio into a trained comprehensive cost determination model to obtain a comprehensive cost, wherein the comprehensive cost determination model is trained by inputting multiple training samples into a neural network, each training sample including historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbases, historical vehicle steering ratios, and corresponding historical standard comprehensive costs, the comprehensive cost being used to reflect the costs of all lanes in the linked candidate lanes of the lane topology;

[0020] The control module is used to select a target lane with the minimum cost from the alternative lanes linked by the lane topology according to the comprehensive cost, and control the operation of the current vehicle.

[0021] In a third aspect, the present invention provides the following technical solution through an embodiment of the present invention:

[0022] A vehicle comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods described in the first aspect when executing the program.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0024] An embodiment of the present invention provides an autonomous driving lane selection control method. When a vehicle approaches a node-associated road, the method obtains the vehicle's driving state parameters and lane pointer data. The driving state parameters include steering wheel angle, ego vehicle angular velocity, and ego vehicle linear velocity. The lane pointer data includes the vehicle's current lateral distance from the centerline of the nearest lane, lane topology, and the virtual lane link sequence ahead of the vehicle. The method then inputs the driving state parameters, lane pointer data, node-associated road, vehicle wheelbase, and steering gear ratio into a trained neural network model to obtain a comprehensive cost. Based on the comprehensive cost, the method selects the target lane with the lowest cost from the lane topology linked candidate lanes to control the vehicle's operation. This method identifies the road topology connectivity of the road ahead, rationalizes the lane selection cost, and optimizes the cost weights through a neural network to complete lane selection in the scenario. The method dynamically adjusts the predicted position based on the vehicle's current position and lane sequence, incorporating different traffic scenarios, to ensure the consistency and stability of path planning. This method enables intersection-passing capabilities for assisted driving without lane-level navigation, achieving comfortable and reliable results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 This is a flowchart of an automatic driving lane selection control method according to an embodiment of the present invention;

[0027] Figure 2 This is an example diagram of lane selection at an intersection in an embodiment of the present invention;

[0028] Figure 3 Schematic diagram of training a neural network model in different scenarios according to an embodiment of the present invention;

[0029] Figure 4 A schematic diagram of the structure of a neural network model in an embodiment of the present invention;

[0030] Figure 5Schematic diagram of tracking a preview point on a lane centerline according to an embodiment of the present invention;

[0031] Figure 6 This is a schematic diagram of the structure of an automatic driving lane selection control device according to an embodiment of the present invention;

[0032] Figure 7 Schematic diagram of the structure of a vehicle in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The embodiments of the present application provide an automatic driving lane selection control method, device and vehicle, which can select a reasonable opposite lane based on multiple parameters such as the vehicle's own driving status and external environmental factors, thereby achieving comfortable and reliable passage through intersections.

[0034] The technical solution of the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0035] A method for controlling lane selection of an automatic driving system includes: collecting information about the road ahead of the vehicle and positioning information; if it is determined that the vehicle is traveling to a road associated with a node based on the information about the road ahead of the vehicle and the positioning information, obtaining the driving state parameters and lane pointer data of the current vehicle, wherein the driving state parameters include the steering wheel angle, the angular velocity of the vehicle and the linear velocity of the vehicle, and the lane pointer data includes the current lateral distance between the vehicle and the center line of the nearest lane, the lane topology and the virtual lane connection sequence in front of the vehicle, the virtual lane connection sequence including the connection relationship between different lanes in front of the vehicle and a feasible path sequence; combining the driving state parameters, lane pointer data and node association data into a virtual lane connection sequence; The connecting road, the preset vehicle wheelbase and the vehicle's steering transmission ratio are input into the trained comprehensive cost determination model to obtain the comprehensive cost, wherein the comprehensive cost determination model includes inputting multiple training samples into the neural network for training, each training sample includes historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbases, historical vehicle steering transmission ratios and corresponding historical standard comprehensive costs, and the comprehensive cost is used to reflect the costs of all lanes in the linked alternative lanes of the lane topology; according to the comprehensive cost, the target lane with the smallest cost is selected from the lane topology linked alternative lanes to control the operation of the current vehicle.

[0036] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0037] In the first aspect, an embodiment of the present invention provides an automatic driving lane selection control method, specifically, Figure 1 As shown, the method includes the following steps S101 to S104:

[0038] Step S101, collecting the road information and positioning information ahead of the vehicle;

[0039] Step S102: If it is determined based on the vehicle's forward road information and positioning information that the vehicle is traveling on a node-associated road, current vehicle driving state parameters and lane pointer data are obtained, wherein the driving state parameters include the steering wheel angle, the vehicle's angular velocity, and the vehicle's linear velocity. The lane pointer data includes the vehicle's current lateral distance from the centerline of the nearest lane, the lane topology, and a virtual lane connection sequence in front of the vehicle. The virtual lane connection sequence includes the connection relationship between different lanes in front of the vehicle and a feasible path sequence.

[0040] Step S103, input the driving state parameters, the lane pointer data, the node-associated roads, the preset vehicle wheelbase and the vehicle steering transmission ratio into the trained comprehensive cost determination model to obtain the comprehensive cost, wherein the comprehensive cost determination model includes inputting multiple training samples into the neural network for training, each of the training samples includes historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbase, historical vehicle steering transmission ratio and corresponding historical standard comprehensive cost, and the comprehensive cost is used to reflect the cost of all lanes in the linked alternative lanes of the lane topology.

[0041] It should be noted that the lane centering cruise assist function is not available when entering the following Figure 2 In the scenario shown, discontinuous lane lines often degrade the driving experience and bring safety risks. With the improvement of perception performance, lane topology (TOPO) connections will be constructed between the interrupted lane lines, which provides the corresponding foundation for Lane Centering Assist (LCC) to smoothly pass this scenario.

[0042] In a specific embodiment, the road information in front of the vehicle can be collected in real time by radar, camera and other equipment installed on the vehicle, and the vehicle's positioning information can be obtained by GPS, Beidou positioning module and other equipment installed on the vehicle to determine the vehicle's position on the map.

[0043] Based on the road information and positioning information in front of the vehicle, determine whether the vehicle is traveling to the node-associated road. For example, if the road in front of the vehicle is an intersection, the camera and / or positioning information recognizes that the distance between the vehicle and the intersection meets the lane selection control conditions. At this time, start to obtain the current vehicle's driving status parameters and vehicle pointer data.

[0044] Among them, node-associated roads may include: intersection scenarios and diversion scenarios. Diversion scenarios may represent road conditions where a main road and other branch roads are divided. Intersection scenarios may represent scenarios where lane lines are discontinuous, such as intersections, traffic light intersections, T-shaped intersections, and the like.

[0045] In specific implementations, obtaining the current lateral distance current_lat may include obtaining the vehicle's position in the local coordinate system and the lane information of the nearest lane, and calculating the distance from the vehicle to the centerline of the lane by projection, which is the lateral deviation current_lat between the vehicle's current position and the lane centerline. The lane information of the nearest lane may include the distance from the vehicle to the left and right lane lines, respectively.

[0046] Alternatively, the current lane position of the vehicle can be obtained through a virtual indicator in the vehicle's intelligent driving system, and then the distance between the vehicle's current position and the center line of the lane can be calculated.

[0047] Acquisition of lane topology and virtual lane link sequence can include: using a forward-looking camera to capture images of the road in front of the vehicle, and detecting targets such as lane lines, traffic signs, and other vehicles through image recognition algorithms; or using lidar to obtain three-dimensional point cloud data of the vehicle's surroundings in real time; or, obtaining the vehicle's position in the map through the vehicle's positioning system, and then querying the lane information in front of the vehicle from the high-precision map to obtain the lane topology and virtual lane link sequence.

[0048] It should be noted that lane topology can accurately determine the connection method between lanes, the direction of lanes at intersections, etc.; the virtual lane link sequence is a technology used in the field of autonomous driving to dynamically construct vehicle driving paths. It can combine sensor data with high-precision maps through algorithms to virtually generate the connection relationship and feasible path sequence of different lanes in front of the vehicle.

[0049] The combined cost represents the cost of all lanes in the linked candidate lanes in the lane topology. A lower lane cost indicates higher feasibility and is therefore chosen as the target lane. It should be noted that a lower cost indicates greater reliability, comfort, consistency, and stability when the vehicle is operating in that target lane.

[0050] In one embodiment, before collecting the road information and positioning information ahead of the vehicle, the method may further include: obtaining training samples, the training samples including multiple sets of driving state parameters and lane pointer data of the vehicle under each road structure when the vehicle travels to different node-associated roads, and the comprehensive cost of the alternative lanes of the vehicle's lane topology link under each set of driving state parameters and lane pointer data; based on multiple training samples, training a pre-constructed neural network to obtain a trained comprehensive cost determination model.

[0051] In a specific embodiment, in order to improve the generalization performance of the comprehensive cost determination model, a variety of traffic scenarios are designed (e.g. Figure 3 Based on different vehicle driving conditions, the system collects vehicle driving state parameters and lane pointer data for neural network training. The trained comprehensive cost determination model can automatically switch the weight coefficients of the model operation according to the scenario.

[0052] Specifically, obtaining training samples may include: controlling the vehicle to operate in different road scenarios using different driving state parameters, selecting target paths for the vehicle in different road scenarios and under different driving state parameters, and obtaining a standard comprehensive cost.

[0053] The construction process of the comprehensive cost determination model may include: constructing a hierarchical model, using expert scoring to determine the relative importance of each parameter, and then calculating the weight coefficient through consistency testing to construct a comprehensive cost determination model. The comprehensive cost determination model can be a 4-layer BP neural network, including an input layer, a first hidden layer, a second hidden layer, and an output layer, such as Figure 4 As shown in the figure, the first layer Layer1 is the input (driving state parameters and lane pointer data, etc.), the second layer Layer2 is the first hidden layer (including current lateral deviation, predicted lateral deviation, curvature difference, etc.), the third layer Layer3 is the second hidden layer (including lateral sub-cost, consistency sub-cost, steering wheel angle sub-cost), and the fourth layer Layer4 is the comprehensive cost.

[0054] The weight coefficients from Layer 1 to Layer 2 can be directly obtained through model prediction calculation, and the weight coefficients from Layer 2 to Layer 3 and Layer 3 to Layer 4 are obtained through model training.

[0055] The comprehensive cost is obtained through the following steps: determining the intersection lane selection cost based on the node-associated road at which the vehicle is located, driving status parameters and lane pointer data, the preset vehicle wheelbase and the vehicle's steering transmission ratio, wherein the intersection lane selection cost includes a lateral sub-cost, a steering wheel angle sub-cost and a consistency sub-cost; obtaining the comprehensive cost based on the lateral sub-cost, the steering wheel angle sub-cost and the consistency sub-cost.

[0056] In a specific embodiment, determining the lateral sub-cost may include: introducing the steering wheel angle, vehicle angular velocity, vehicle linear velocity, wheelbase, and steering gear ratio into the motion model to obtain the predicted lateral distance of the current vehicle after a preset time; based on the virtual lane link sequence, using the three-point method to determine the maximum curvature and minimum curvature on the lane link, and then obtaining the curvature difference based on the maximum curvature and minimum curvature; and obtaining the lateral sub-cost based on the current lateral distance, the predicted lateral distance, and the curvature difference.

[0057] Optionally, the preset time length may be 2 to 5 seconds, for example, 3.6 seconds. In one embodiment, the calculation of the current vehicle's position 3.6 seconds in the future requires the use of a motion model to estimate the position (x, y) of the vehicle at time t. Assume that the position (x, y) of the vehicle at time t can be calculated using the following motion model function:

[0058] x=v×t

[0059]

[0060] Where k is the curvature, v is the linear velocity of the vehicle, and t is 3.6 seconds.

[0061] In a specific embodiment, the curvature k can be estimated as follows: first, determine whether the vehicle's linear speed (speed) is greater than a preset linear speed. If so, indicating that the vehicle speed is relatively high, the curvature K = vehicle's angular speed (yaw_rate_rps) / vehicle's linear speed (speed). If speed is less than or equal to the preset linear speed, the curvature is calculated based on the steering angle, i.e., curvature K = tan(front wheel steering angle δ) / wheelbase (base) = tan(steering wheel angle × steering gear ratio) / wheelbase (base).

[0062] Optionally, the preset linear speed may be between 2.5 m / s and 3.3 m / s, for example, 3 m / s.

[0063] The vehicle's position (x, y) in the local coordinate system at 3.6 seconds and the lane information of the nearest lane are obtained through the above method. The distance from the vehicle to the centerline of the lane is calculated by projection, which is the predicted lateral deviation predict_lat between the vehicle's position and the lane at the next 3.6 seconds.

[0064] Of course, as another optional embodiment, the steering wheel angle, vehicle angular velocity, vehicle linear velocity, wheelbase, steering ratio, and motion model coefficients may be brought into the motion model to obtain the predicted lateral distance of the current vehicle after 3.6 seconds.

[0065] Next, based on the extracted virtual lane link sequence, the point set on the virtual lane centerline is obtained, and then the maximum curvature and minimum curvature on the lane link are determined using the three-point method.

[0066] In a specific embodiment, it is determined whether the number of points on the center line of the virtual lane obtained is greater than 2. If so, the curvature calculation begins. Three consecutive points p0(x1, y1), p1(x2, y2), and p2(x3, y3) are selected from the aforementioned point set, and the sign and magnitude of the curvature are calculated using the following cross product formula:

[0067]

[0068] Where dist0 is the distance between p0 and p1, and dist1 is the distance between p1 and p2. Here, the product of the two distances is used as the denominator to prevent division by zero. cross_val is obtained as follows:

[0069] Based on p0(x1, y1), p1(x2, y2), and p2(x3, y3), we calculate two vectors: v1 = [x2 - x1, y2 - y1] and v2 = [x3 - x2, y3 - y2], with cross_val = v1 × v2. The sign of cross_val determines the sign of the curvature (indicating the direction of the curvature), and its absolute value, combined with the denominator (the product of the two distances), quantifies the curvature.

[0070] The three-point method is used to calculate all curvature values for all points in the point set, and then the maximum curvature and the minimum curvature are determined from all the curvature values. The minimum curvature is then subtracted from the maximum curvature to obtain the curvature difference, that is, curvature difference curvature_diff = maximum curvature max_curvature - minimum curvature min_curvature.

[0071] Next, the current lateral distance current_lat, the predicted lateral distance predict_lat, and the curvature difference curvature_diff are normalized and then multiplied by a weight coefficient (determined according to the comprehensive cost determination model) to determine the lateral sub-cost.

[0072] In a specific embodiment, the determination of the steering wheel angle sub-cost and the consistency sub-cost may include: step S201, determining the preview point of the current vehicle based on the node-associated road; step S202, determining the predicted steering wheel angle required for the current vehicle to track the preview point on the center line of the lane using a pure tracking method based on the preview point; step S203, obtaining the steering wheel angle sub-cost based on the deviation between the predicted steering wheel angle and the steering wheel angle; step S204, obtaining the consistency sub-cost based on the difference between the predicted steering wheel angle at the current moment and the predicted steering wheel angle at the previous moment.

[0073] In one embodiment, step S201, determining a preview point of the current vehicle based on a node-associated road, may include: traversing each lane in a virtual lane link sequence to obtain a projection distance of the vehicle's current position on each lane in the virtual lane link sequence; adding a set normal viewing distance to the projection distance to obtain a target point at the normal viewing distance; adding a set near viewing distance to the projection distance to obtain a target point at the near viewing distance; and determining the preview point of the current vehicle based on the target point at the normal viewing distance, the target point at the near viewing distance, and the node-associated road.

[0074] For example, the projection distance is added to the set normal viewing distance, and the target point at the normal viewing distance is obtained as follows: projection distance (0, lateral deviation) + set normal viewing distance ds, and the target point at the normal viewing distance is obtained as (ds, lateral deviation).

[0075] Specifically, traverse each lane in the lane sequence, obtain the projection distance ego_s of the vehicle's current position ego_pos_ on the lane sequence, and calculate the target point target at normal sight distance and the target point at near sight distance.

[0076] The normal sight distance ds can be set to the larger of a first set distance (e.g., 15 meters) and a first calculated distance x1, which is equal to the product of a first preset tracking time (e.g., 3.6 seconds) and the vehicle's linear velocity v. The near sight distance near_ds can be set to the larger of a second set distance (e.g., 10 meters) and a second calculated distance x2, which is equal to the product of a second preset tracking time (e.g., 1 second) and the vehicle's linear velocity v.

[0077] Add the projection distance to the set normal sight distance ds, that is, the current position of the vehicle plus the set normal sight distance (the total distance cannot exceed the total length of the lane), to obtain the target point target under the normal sight distance. target is the corresponding lane point under the normal sight distance. Add the projection distance to the set near sight distance near_ds to obtain the target point near_target under the near sight distance. near_target is the corresponding lane point under the near sight distance.

[0078] In one embodiment, if the node-associated road is an intersection scene, determining the preview point of the current vehicle based on the target point at normal sight distance, the target point at near sight distance, and the node-associated road can include: preliminarily setting the preview point to the end point of the current lane centerline, and calculating the distance difference between the preview point and the projection distance; if the distance difference is less than the set near sight distance near_ds, determining the target point at the near sight distance as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance near_ds, determining the target point at the normal sight distance as the preview point of the current vehicle.

[0079] Specifically, if the lane in front of the vehicle is a virtual lane and the scene in front is an intersection, the preview point is set to the end point of the centerline of the lane, and the distance difference between the preview point and the current position of the vehicle is calculated. If the distance difference is less than the set near-sight distance near_ds, the preview point is switched to the target point near_target at the near-sight distance to avoid overly sensitive corner calculation; if the distance difference is greater than or equal to the set near-sight distance near_ds, the preview point is switched to the target point target at the normal sight distance.

[0080] In another embodiment, if the node-associated road is a diversion scenario, the preview point of the current vehicle is determined based on the target point at normal sight distance, the target point at near sight distance, and the node-associated road, which may include: if the current vehicle meets the diversion condition, obtaining the diversion starting point; preliminarily setting the preview point to the corresponding point on the center line of the diversion lane, and determining the distance difference between the preview point and the diversion starting point; if the distance difference is less than the set near sight distance near_ds, determining the target point at the near sight distance as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance near_ds, determining the target point at the normal sight distance as the preview point of the current vehicle.

[0081] In this specific embodiment, the lane sequence is traversed to find a divergence point, i.e., where the current vehicle has multiple subsequent lanes. A determination is made as to whether the current vehicle meets the divergence conditions. If so, the divergence starting point is obtained. It should be noted that the divergence starting point represents the point at which the divergence road begins to diverge, and from this point, there are two or more connected topological lanes.

[0082] Specifically, to determine whether the current vehicle meets the diversion conditions, a comprehensive analysis can be conducted from multiple dimensions such as traffic rules, vehicle status, road environment, driving dynamics, etc. to determine whether the vehicle meets the diversion conditions.

[0083] The distance between the divergence starting point and the vehicle's current position is the divergence starting distance. A point is located on the centerline of the divergence lane based on the divergence starting distance as the preview point. The distance difference between the preview point and the divergence starting point is calculated. If the difference is less than the set near sight distance near_ds, the preview point is switched to the target point near_target under the near sight distance. If the difference is greater than or equal to the set near sight distance near_ds, the preview point is switched to the target point target under the normal sight distance.

[0084] Here, the position of the preview point is dynamically adjusted based on the vehicle's current position and the virtual lane connection sequence, combined with different node-related road conditions (intersection scenarios, diversion scenarios), to ensure the consistency and stability of subsequent path planning and improve the comfort and reliability of vehicle driving.

[0085] Next, step S202 is executed to determine the predicted steering wheel angle required for the current vehicle to track the preview point on the lane centerline using a pure tracking method based on the preview point.

[0086] In a specific embodiment, the steering wheel angle required to track a preview point on the lane centerline is calculated using a pure tracking method, where the input is a preview point and the output is the steering wheel angle. It should be noted that the principle of the pure tracking method is that the vehicle can reach the preview point by turning the steering wheel at a fixed angle (regardless of the direction in which the preview point is reached).

[0087] like Figure 5 As shown, based on the preview point, a pure tracking method is used to determine the predicted steering wheel angle required for the current vehicle to track the preview point on the centerline of the lane. This may include: under the premise that the front wheel angle remains unchanged, the vehicle performs a circular motion with the rotation center as the circle, the radius of the circle is R, that is, the distance from the reference point (the center of the vehicle's rear axle) to the rotation center is R. At this time, the rotation center, the reference point, and the preview point determined based on the above method form an isosceles triangle. The connection between the reference point and the preview point is the base of the isosceles triangle, and the length is l. d (like Figure 5 As shown in Figure 2, the angle between the base of the isosceles triangle and the longitudinal axis of the vehicle is denoted as α. Since the longitudinal axis of the vehicle is always tangent to the arc, the angle between the base of the isosceles triangle is 2α.

[0088] The acquisition of the rotation center may include: first defining the center characteristics of an ideal circular motion according to the front wheel steering angle and wheelbase of the vehicle, and then determining the position of the rotation center with reference to the curvature of the path (or the preview point).

[0089] According to the sine theorem, we can get:

[0090]

[0091] According to the bicycle model, the turning radius R and the front wheel angle have the following relationship:

[0092]

[0093] From this we can get the front wheel angle:

[0094]

[0095] Where L is the wheelbase. The above formula can also be rewritten in terms of lateral error as follows. This calculation form is suitable for the ego vehicle coordinate system, that is, all coordinates are expressed in the ego vehicle local coordinate system:

[0096]

[0097] Among them, L is the wheelbase, ye is the lateral deviation of the preview point, is the Euclidean distance between the preview point and the reference point. According to the pure tracking method, the steering wheel angle required to track the preview point on the lane centerline is implemented as follows:

[0098] Step S203, obtaining a steering wheel angle sub-cost based on the deviation between the predicted steering wheel angle and the steering wheel angle, including: multiplying the deviation between the predicted steering wheel angle and the steering wheel angle by a weight coefficient (determined according to a comprehensive cost determination model) to determine the steering wheel angle sub-cost.

[0099] Step S204, obtains the consistency sub-cost based on the difference between the predicted steering wheel angle at the current moment and the predicted steering wheel angle at the previous moment, including: multiplying the difference between the predicted steering wheel angle at the current moment and the predicted steering wheel angle at the previous moment by a weight coefficient (determined according to the comprehensive cost determination model) to determine the consistency sub-cost.

[0100] The consistency sub-cost can be used to calculate the consistency cost of the current lane sequence and evaluate the rationality and consistency of the vehicle path to ensure the smoothness and stability of vehicle driving.

[0101] In one embodiment, the consistency sub-cost is obtained by calculating the difference (normalized angle difference) between the predicted steering wheel at the previous moment and the predicted steering wheel at the current moment, and then multiplying it by the weight coefficient and the angle conversion coefficient (57.3, converting radians to degrees).

[0102] Step S104 : selecting a target lane with the minimum cost from the candidate lanes linked by the lane topology according to the comprehensive cost, and controlling the operation of the current vehicle.

[0103] In a specific embodiment, all candidate lanes are traversed, and for each candidate, a comprehensive cost is calculated using a comprehensive cost determination model. The lane with the lowest comprehensive cost is then selected as the target lane. By combining this comprehensive cost determination model with a real-time control algorithm, the vehicle can autonomously select the optimal lane in complex traffic environments, achieving safe and efficient driving control.

[0104] Therefore, compared with the traditional lane screening based on lateral distance, this application proposes a lane selection method that adopts multi-objective (lateral sub-cost, steering wheel angle sub-cost, consistency sub-cost) weight coefficients, and designs cost calculation methods under different objectives; at the same time, in order to solve the problem of weighted calculation of multi-objective cost weights, this solution proposes to adopt a neural network method, combined with a multi-objective cost model, to design a 4-layer neural network architecture. Through data collection, different weight models can be trained based on different driving habits to meet the needs of users with different driving habits. The proposed method increases the weight of the comfort factor in lane selection through weight setting, greatly improving the comfort of lane selection. At the same time, by suppressing large steering wheel movements, it improves the safety of lane selection, and has good practicality in engineering.

[0105] The lane selection control method proposed in this application uses perception to identify lane topology connections, rationally defines lane selection costs, and optimizes cost weights through a neural network to complete lane selection in this scenario, thereby achieving comfortable and reliable intersection navigation capabilities for lane centering assisted driving without lane-level navigation.

[0106] In summary, the autonomous driving lane selection control method provided by the embodiments of the present invention identifies the road topology connectivity of the road ahead, rationally defines lane selection costs, and optimizes cost weights through a neural network to complete lane selection in this scenario. This method dynamically adjusts the predicted position based on the vehicle's current position and lane sequence, incorporating different traffic scenarios, to ensure the consistency and stability of path planning. This enables assisted driving functions to achieve intersection navigation without lane-level navigation, achieving a comfortable and reliable experience.

[0107] In the second aspect, based on the same inventive concept, this embodiment provides an automatic driving lane selection control device, such as Figure 6 As shown, including:

[0108] The acquisition module 401 is used to acquire the road information and positioning information in front of the vehicle;

[0109] Acquisition module 402 is configured to, if it is determined based on the vehicle's forward road information and positioning information that the vehicle is traveling on a node-associated road, obtain current vehicle driving state parameters and lane pointer data, wherein the driving state parameters include a steering wheel angle, a vehicle angular velocity, and a vehicle linear velocity; and the lane pointer data includes a current lateral distance from the vehicle to the centerline of the nearest lane, lane topology, and a virtual lane connection sequence in front of the vehicle, wherein the virtual lane connection sequence includes a connection relationship between different lanes in front of the vehicle and a feasible path sequence.

[0110] a cost determination module 403 for inputting the driving state parameters, the lane pointer data, the node-associated roads, the preset vehicle wheelbase, and the vehicle steering gear ratio into a trained comprehensive cost determination model to obtain a comprehensive cost, wherein the comprehensive cost determination model is obtained by inputting multiple training samples into a neural network for training, each training sample including historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbases, historical vehicle steering gear ratios, and corresponding historical standard comprehensive costs, the comprehensive cost being used to reflect the costs of all lanes in the linked candidate lanes of the lane topology;

[0111] The control module 404 is configured to select a target lane with the minimum cost from the candidate lanes linked by the lane topology according to the comprehensive cost, and control the operation of the current vehicle.

[0112] As an optional embodiment, the device also includes: a training module for obtaining training samples, the training samples including multiple sets of driving state parameters and lane pointer data of the vehicle under each road structure when the vehicle travels to different node-associated roads, and the comprehensive cost of the alternative lanes of the vehicle's lane topology link under each set of driving state parameters and lane pointer data; based on multiple training samples, a pre-constructed neural network is trained to obtain a trained comprehensive cost determination model.

[0113] As an optional embodiment, the comprehensive cost is obtained by the following steps:

[0114] The intersection lane selection cost is determined based on the node-associated road at the vehicle, driving status parameters and lane pointer data, the preset vehicle wheelbase and the vehicle's steering transmission ratio. The intersection lane selection cost includes a lateral sub-cost, a steering wheel angle sub-cost and a consistency sub-cost; based on the lateral sub-cost, the steering wheel angle sub-cost and the consistency sub-cost, a comprehensive cost is obtained.

[0115] As an optional embodiment, the determination of the horizontal sub-cost includes:

[0116] The steering wheel angle, ego vehicle angular velocity, ego vehicle linear velocity, wheelbase, and steering gear ratio are fed into a preset motion model to obtain the predicted lateral distance of the current vehicle after a preset duration. Based on the virtual lane link sequence, the three-point method is used to determine the maximum and minimum curvatures on the lane link. The curvature difference is then calculated based on the maximum and minimum curvatures. The lateral sub-cost is then determined based on the current lateral distance, the predicted lateral distance, and the curvature difference.

[0117] As an optional embodiment, the determination of the steering wheel angle sub-cost and the consistency sub-cost includes: determining the preview point of the current vehicle based on the node-associated road; determining the predicted steering wheel angle required for the current vehicle to track the preview point on the center line of the lane using a pure tracking method based on the preview point; obtaining the steering wheel angle sub-cost based on the deviation between the predicted steering wheel angle and the steering wheel angle; and obtaining the consistency sub-cost based on the difference between the predicted steering wheel angle at the current moment and the predicted steering wheel angle at the previous moment.

[0118] As an optional embodiment, determining the preview point of the current vehicle based on the node-associated road includes: traversing each lane in the virtual lane link sequence to obtain the projection distance of the vehicle's current position on each lane in the virtual lane link sequence; adding the projection distance to the set normal viewing distance to obtain the target point at the normal viewing distance; adding the projection distance to the set near viewing distance to obtain the target point at the near viewing distance; and determining the preview point of the current vehicle based on the target point at the normal viewing distance, the target point at the near viewing distance, and the node-associated road.

[0119] As an optional embodiment, if the node-associated road is an intersection scene, the preview point of the current vehicle is determined based on the target point at normal sight distance, the target point at near sight distance and the node-associated road, including: preliminarily setting the preview point to the end point of the current lane centerline, and calculating the distance difference between the preview point and the projection distance; if the distance difference is less than the set near sight distance, the target point at the near sight distance is determined as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance, the target point at normal sight distance is determined as the preview point of the current vehicle.

[0120] As an optional embodiment, if the node-associated road is a diversion scenario, the preview point of the current vehicle is determined based on the target point at normal sight distance, the target point at near sight distance, and the node-associated road, including: if the current vehicle meets the diversion conditions, the diversion starting point is obtained; the preview point is initially set to the corresponding point on the center line of the diversion lane, and the distance difference between the preview point and the diversion starting point is determined; if the distance difference is less than the set near sight distance, the target point at the near sight distance is determined as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance, the target point at normal sight distance is determined as the preview point of the current vehicle.

[0121] The above modules can be implemented by software codes, in which case the above modules can be stored in the memory of the control device. The above modules can also be implemented by hardware such as integrated circuit chips.

[0122] An embodiment of the present invention provides an automatic driving lane selection control device, the implementation principle and technical effects of which are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding content in the aforementioned method embodiment.

[0123] In a third aspect, based on the same inventive concept, this embodiment provides a vehicle 500, such as Figure 7 As shown, it includes: a memory 501, a processor 502 and a computer program 503 stored in the memory and capable of running on the processor. When the processor 501 executes the program, the steps of the automatic driving lane selection control method described in the first aspect are implemented.

[0124] Since the vehicle described in this embodiment is the vehicle used to implement the autonomous driving lane selection control method in the embodiments of this application, those skilled in the art will be able to understand the specific implementation of the vehicle in this embodiment and its various variations based on the autonomous driving lane selection control method described in the embodiments of this application. Therefore, how the vehicle implements the method in the embodiments of this application will not be described in detail here. As long as those skilled in the art can implement the vehicle used in the autonomous driving lane selection control method in the embodiments of this application, it falls within the scope of protection of this application.

[0125] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A module that specifies functions in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction module, which is implemented in the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0129] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0130] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for controlling lane selection in an automatic driving vehicle, characterized in that: include: Collect the road information and positioning information ahead of the vehicle; If it is determined based on the vehicle's forward road information and positioning information that the vehicle is traveling on a node-associated road, obtaining the vehicle's current driving state parameters and lane pointer data, wherein the driving state parameters include the steering wheel angle, the vehicle's angular velocity, and the vehicle's linear velocity; the lane pointer data includes the vehicle's current lateral distance from the centerline of the nearest lane, the lane topology, and a virtual lane connection sequence in front of the vehicle, wherein the virtual lane connection sequence includes the connection relationship between different lanes in front of the vehicle and a feasible path sequence; Inputting the driving state parameters, the lane pointer data, the node-associated roads, the preset vehicle wheelbase, and the vehicle steering transmission ratio into a trained comprehensive cost determination model to obtain a comprehensive cost, wherein the comprehensive cost determination model includes inputting multiple training samples into a neural network for training, each training sample including historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbases, historical vehicle steering transmission ratios, and corresponding historical standard comprehensive costs, the comprehensive cost being used to reflect the costs of all lanes in the linked alternative lanes of the lane topology; According to the comprehensive cost, a target lane with the minimum cost is selected from the alternative lanes linked by the lane topology to control the operation of the current vehicle.

2. The method according to claim 1, wherein Before collecting the vehicle's front road information and positioning information, the method further includes: Acquire training samples, the training samples including multiple sets of driving state parameters and lane pointer data for each road structure when the vehicle travels on different node-associated roads, and comprehensive costs of alternative lanes linked to the vehicle's lane topology for each set of driving state parameters and lane pointer data; Based on the multiple training samples, the pre-built neural network is trained to obtain a trained comprehensive cost determination model.

3. The method according to claim 1, wherein The comprehensive cost is obtained by the following steps: Determine the intersection lane selection cost based on the node-associated road at which the vehicle is located, driving state parameters and lane pointer data, a preset vehicle wheelbase, and a vehicle steering gear ratio, wherein the intersection lane selection cost includes a lateral sub-cost, a steering wheel angle sub-cost, and a consistency sub-cost; A comprehensive cost is obtained according to the lateral sub-cost, the steering wheel angle sub-cost, and the consistency sub-cost.

4. The method according to claim 3, wherein The determination of the horizontal sub-cost includes: Substituting the steering wheel angle, the vehicle angular velocity, the vehicle linear velocity, the wheelbase, and the steering gear ratio into a preset motion model to obtain a predicted lateral distance of the current vehicle after a preset time period; According to the virtual lane link sequence, a maximum curvature and a minimum curvature on the lane link are determined using a three-point method, and then a curvature difference is obtained based on the maximum curvature and the minimum curvature; A horizontal sub-cost is obtained according to the current horizontal distance, the predicted horizontal distance, and the curvature difference.

5. The method according to claim 3, wherein The determination of the steering wheel turning sub-cost and the consistency sub-cost includes: Determine the preview point of the current vehicle based on the node-associated road; Based on the preview point, a pure tracking method is used to determine the predicted steering wheel angle required for the current vehicle to track the preview point on the lane centerline; Obtaining a steering wheel angle sub-cost according to a deviation between the predicted steering wheel angle and the steering wheel angle; The consistency sub-cost is obtained based on the difference between the predicted steering wheel angle at the current moment and the predicted steering wheel angle at the previous moment.

6. The method according to claim 5, wherein Determining the preview point of the current vehicle based on the node-associated road includes: Traverse each lane in the virtual lane link sequence and obtain the projection distance of the vehicle's current position on each lane in the virtual lane link sequence; Add the projection distance to the set normal viewing distance to obtain the target point at the normal viewing distance; add the projection distance to the set near viewing distance to obtain the target point at the near viewing distance; The preview point of the current vehicle is determined according to the target point at normal sight distance, the target point at near sight distance and the node-associated road.

7. The method according to claim 6, wherein If the node-associated road is an intersection scene, determining the preview point of the current vehicle based on the target point at normal sight distance, the target point at near sight distance, and the node-associated road includes: Preliminarily setting the preview point as the end point of the current lane centerline, and calculating the distance difference between the preview point and the projection distance; If the distance difference is less than the set near sight distance, the target point at the near sight distance is determined as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance, the target point at the normal sight distance is determined as the preview point of the current vehicle.

8. The method according to claim 6, wherein If the node-associated road is a diversion scenario, determining the preview point of the current vehicle based on the target point at normal sight distance, the target point at near sight distance, and the node-associated road includes: If the current vehicle meets the diversion conditions, obtain the diversion starting point; Preliminarily setting the preview point to a corresponding point on the center line of the diversion lane, and determining the distance difference between the preview point and the diversion starting point; If the distance difference is less than the set near sight distance, the target point at the near sight distance is determined as the preview point of the current vehicle; if the distance difference is greater than or equal to the set near sight distance, the target point at the normal sight distance is determined as the preview point of the current vehicle.

9. An automatic driving lane selection control device, characterized in that: include: A collection module is used to collect road information and positioning information in front of the vehicle; an acquisition module, configured to, if it is determined based on the vehicle's forward road information and positioning information that the vehicle is traveling on a node-associated road, obtain current vehicle driving state parameters and lane pointer data, wherein the driving state parameters include a steering wheel angle, a vehicle angular velocity, and a vehicle linear velocity; and the lane pointer data includes a current lateral distance from the vehicle to the centerline of the nearest lane, lane topology, and a virtual lane connection sequence in front of the vehicle, wherein the virtual lane connection sequence includes a connection relationship between different lanes in front of the vehicle and a feasible path sequence; a cost determination module, configured to input the driving state parameters, the lane pointer data, the node-associated roads, a preset vehicle wheelbase, and a vehicle steering transmission ratio into a trained comprehensive cost determination model to obtain a comprehensive cost, wherein the comprehensive cost determination model is trained by inputting multiple training samples into a neural network, each training sample including historical driving state parameters, historical lane pointer data, historical node-associated roads, historical vehicle wheelbases, historical vehicle steering transmission ratios, and corresponding historical standard comprehensive costs, the comprehensive cost being used to reflect the costs of all lanes in the linked candidate lanes of the lane topology; The control module is used to select a target lane with the minimum cost from the alternative lanes linked by the lane topology according to the comprehensive cost, and control the operation of the current vehicle.

10. A vehicle, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the method according to any one of claims 1 to 8 are implemented when the processor executes the program.

Images