A method, system and computer-readable storage medium for main vehicle motion planning

Abstract

This application relates to the technical field of vehicle route prediction, and more particularly to a method, system, and computer-readable storage medium for motion planning of a primary vehicle. The method comprises: obtaining the historical trajectory of the primary vehicle, the historical trajectories of other vehicles, and the road network structure, inputting these into a deep neural network to obtain the predicted trajectory of the primary vehicle from the perspective of other vehicles; and generating a driving path by combining the predicted trajectory of the primary vehicle from the perspective of other vehicles with the primary vehicle's travel routes. This application has the technical effect of improving the rationality of the primary vehicle's driving path planning, thereby increasing the accuracy of the primary vehicle's driving path.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle route prediction, and in particular to a method and system for main vehicle motion planning and a computer-readable storage medium. Background Art

[0002] At present, with the in-depth research on the technology of intelligent driving cars, people are gradually paying attention to how to navigate cars more efficiently and safely in complex intersection scenarios, that is, how to safely perform vehicle lane changes, overtaking, deceleration and other related operations on complex roads. This requires intelligent driving cars to be able to predict their own trajectories and those of surrounding vehicles, so as to avoid such risks in time before the surrounding environment changes.

[0003] Existing vehicle trajectory prediction solutions primarily use deep neural networks, such as recurrent neural networks (RNNs) and graph neural networks (GNNs), along with long short-term memory (LSTM) networks, to predict future vehicle trajectories. The LSTM network extracts temporal features from the vehicle's historical trajectory, fits the nonlinear relationship between the historical and future trajectories, and finally completes the prediction through backpropagation of the loss function. Many models utilize the seq2seq encoder-decoder architecture used in natural language processing. The LSTM encoder extracts the temporal features of the vehicle trajectory and encodes them into a feature context vector. The LSTM decoder decodes the context vector to generate future trajectory coordinates. For example, the paper "VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation," included in the Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, describes the use of VectorNet for vehicle trajectory prediction.

[0004] In the process of implementing this application, the inventors found that there are at least the following problems in the existing technology: when predicting vehicle trajectories, due to the complex traffic conditions in reality, the prediction accuracy of the current prediction method is low and cannot be applied to complex road conditions. Summary of the Invention

[0005] To this end, embodiments of the present application provide a vehicle motion planning method, system, and computer-readable storage medium, which can address the technical problem of low prediction accuracy of existing vehicle trajectory prediction methods. The specific technical solutions are as follows:

[0006] In a first aspect, an embodiment of the present application provides a method for motion planning of a host vehicle, the method comprising:

[0007] Obtain the historical trajectory of the main vehicle, the historical trajectory of other vehicles, and the road network structure, input them into the deep neural network, and obtain the trajectory of the main vehicle predicted from the perspective of other vehicles;

[0008] The driving path is generated by combining the main vehicle trajectory predicted from the perspective of other vehicles and the main vehicle's travel route.

[0009] Preferably, the deep neural network is a modified vectornet, which is a network structure that adds time and space alternating aggregation to vectornet, and uses the previous prediction output as the input of the next prediction to decode the trajectory predicted by the modified vectornet.

[0010] Preferably, the driving path generated by combining the trajectory of the main vehicle predicted from the perspective of other vehicles and the travel route of the main vehicle includes:

[0011] Determine whether the host vehicle changes lanes;

[0012] If changing lanes, obtain the lane change position of the current lane change;

[0013] planning at least one vehicle driving path based on the selected lane change location;

[0014] The desired driving path is selected based on the deviation between at least one vehicle driving path and the host vehicle trajectory predicted from the perspective of other vehicles.

[0015] Preferably, the obtaining of the current lane change position:

[0016] A lane change window is set between each two adjacent vehicles on the road to be changed;

[0017] The desired lane change window is selected as the lane change position based on the main vehicle trajectory predicted from the perspective of other vehicles.

[0018] Preferably, the desired lane change window is selected as the lane change position based on the host vehicle trajectory predicted from the perspective of other vehicles as follows:

[0019] Filter out optional windows based on the current driving situation of the main vehicle;

[0020] Based on the main vehicle's trajectory predicted from the perspective of other vehicles, the nearest lane change window is selected from the optional window as the lane change position.

[0021] Preferably, the selecting of optional windows according to the current driving condition of the host vehicle includes:

[0022] The current driving status of the host vehicle is obtained. If the host vehicle is currently accelerating, the time point at which the lane change operation occurs is preset. The lane change position of the host vehicle at that time point under the current acceleration and the positions of each lane change window are calculated to filter out the optional windows.

[0023] Preferably, the driving path generated by combining the trajectory of the main vehicle predicted from the perspective of other vehicles and the travel route of the main vehicle also includes:

[0024] If you do not change lanes and detect an obstacle ahead;

[0025] Obtain the intersection point between the main vehicle and the obstacle based on the main vehicle trajectory predicted from the perspective of other vehicles and the obstacle trajectory;

[0026] The avoidance relationship between the main vehicle and the obstacle is determined based on the intersection point, and the main vehicle trajectory predicted from the perspective of other vehicles is corrected as the driving path based on the avoidance relationship between the main vehicle and the obstacle.

[0027] Preferably, the desired driving path is selected based on the deviation between at least one vehicle driving path and the host vehicle trajectory predicted from the perspective of other vehicles, including:

[0028] The desired driving path is selected based on the preset human driving data and the deviation between the vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles.

[0029] Preferably, the deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles is calculated by respectively calculating the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in a direction perpendicular to the main vehicle's driving direction, and / or the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation.

[0030] In a second aspect, an embodiment of the present application provides a host vehicle motion planning system, the system comprising:

[0031] The path prediction module is used to obtain the historical trajectory of the main vehicle, the historical trajectory of other vehicles, and the road network structure, and input them into the deep neural network to obtain the predicted trajectory of the main vehicle from the perspective of other vehicles;

[0032] The path planning module generates a driving path by combining the main vehicle's trajectory predicted from the perspective of other vehicles and the main vehicle's travel route.

[0033] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the aforementioned main vehicle motion planning methods are implemented.

[0034] In summary, compared with the prior art, the technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0035] 1. A deep neural network predicts the trajectory of the main vehicle from the perspective of other vehicles. This prediction is used to assist in adapting driving rules for different driving situations, thereby deriving a more reasonable and accurate driving path for the main vehicle.

[0036] 2. A deep neural network obtains the main vehicle's trajectory predicted from the perspective of other vehicles. Combined with the current road conditions, a lane change position is selected based on the current road conditions. The deviation of the vehicle driving path planned by the lane change position is compared with the main vehicle's trajectory predicted from the perspective of other vehicles. The vehicle driving path with the smallest deviation from the main vehicle's trajectory predicted from the perspective of other vehicles is selected. The current actual main vehicle driving situation, road conditions, and the prediction results of the deep neural network are integrated, and the real-time traffic flow status is combined with the prediction results of the neural network to accurately predict the vehicle trajectory. This is suitable for complex road conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of a method for motion planning of a main vehicle provided in one embodiment of the present application.

[0038] Figure 2 This is one of the flow charts of a main vehicle motion planning method provided by another embodiment of the present application.

[0039] Figure 3 It is a space-first vectornet model of a main vehicle motion planning method provided in one embodiment of the present application.

[0040] Figure 4 This is the second flow chart of a main vehicle motion planning method provided in another embodiment of the present application.

[0041] Figure 5 This is the third flow chart of a main vehicle motion planning method provided in another embodiment of the present application.

[0042] Figure 6 This is a fourth flow chart of a method for motion planning of a main vehicle provided in another embodiment of the present application;

[0043] Figure 7 This is the fifth flow chart of a main vehicle motion planning method provided in another embodiment of the present application. DETAILED DESCRIPTION

[0044] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed, but as long as they are within the scope of the claims of the present application, they are protected by the patent law.

[0045] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0046] In addition, the term "and / or" in this application is simply a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this application, unless otherwise specified, generally indicates that the related objects are in an "or" relationship.

[0047] In this application, the terms "first", "second", etc. are used to distinguish identical or similar items with substantially the same effects and functions. It should be understood that there is no logical or temporal dependency between "first", "second", and "nth", nor is there any limitation on the quantity and execution order.

[0048] The embodiments of the present application are described in further detail below with reference to the accompanying drawings.

[0049] Reference Figure 1 In one embodiment of the present application, a method for motion planning of a vehicle is provided. The main steps of the method are described as follows:

[0050] S1: Obtain the historical trajectory of the main vehicle, the historical trajectory of other vehicles, and the road network structure, input them into the deep neural network, and obtain the trajectory of the main vehicle predicted from the perspective of other vehicles;

[0051] S2: Generate a driving path based on the predicted trajectory of the main vehicle from the perspective of other vehicles and the main vehicle's travel route. Specifically, in this embodiment, the main vehicle is the vehicle for which trajectory prediction is being performed, the other vehicles are other vehicles on the main vehicle's route, and the road network structure is the road network structure on which the main vehicle is traveling when trajectory prediction is performed.

[0052] In this embodiment, the deep neural network can be a spatiotemporal graph attention network, trajrank, TPNet, vectornet, etc. By inputting the historical trajectory of the main vehicle and the historical trajectories of other vehicles and the road network structure into the deep neural network, the main vehicle trajectory predicted from the perspective of other vehicles can be obtained.

[0053] The main vehicle's travel route is a modified vehicle driving route based on different driving conditions. For example, if the main vehicle is in a lane-changing state without a human driver, the driving route is planned based on the main vehicle's travel route and the main vehicle's trajectory predicted from the perspective of other vehicles as follows: 1. If the main vehicle's travel route requires a lane change, a lane change position is selected; 2. At least one vehicle driving route is planned based on the lane change position; 3. Based on the deviation between the vehicle driving path and the main vehicle's trajectory predicted from the perspective of other vehicles, the vehicle driving path with the smallest deviation is selected as the driving route. In the case of no human driver, the main vehicle's driving (motion) module converts the driving route into a control instruction for the main vehicle's driving to control the main vehicle's driving. In other embodiments of the present application, different main vehicle travel routes are provided for different main vehicle driving conditions, which will not be described in detail here.

[0054] Through the setting of this application, the deep neural network obtains the main vehicle trajectory predicted from the perspective of other vehicles, and the main vehicle trajectory predicted from the perspective of other vehicles is used to assist driving rules in different driving situations, thereby obtaining the main vehicle's driving path, making the main vehicle's driving path more reasonable and improving the accuracy of the main vehicle's driving path.

[0055] Reference Figure 2 Furthermore, in another embodiment, step S2 includes:

[0056] S21: Determine whether the host vehicle changes lanes;

[0057] S22: If changing lanes, obtaining the current lane change position;

[0058] S23: Planning at least one vehicle driving path according to the selected lane change position;

[0059] S24: Selecting a desired driving path based on the deviation between the at least one vehicle driving path and the host vehicle trajectory predicted from the perspective of other vehicles.

[0060] Specifically, in this embodiment, if the host vehicle's travel route requires a lane change, the lane change position is the position to which the host vehicle needs to change. When there are at least two other vehicles in the lane to which the lane is changed, the lane change position is the position between the two adjacent vehicles. When there are no vehicles in the lane to which the lane is changed, the lane change position is the desired position in that lane. When in the lane to which the lane is changed, obtaining the current lane change position may include: 1. selecting an appropriate lane change position based on human driving data. The human driving data is pre-collected data on the choices made by people in different driving situations. For example, in the case of a lane change, a typical lane change range is used. From the available lane change positions within the range, a position farther from the host vehicle is selected as the lane change position. In other embodiments, other lane change logics may be used. 2. All possible lane change positions are marked, and the position to be changed is determined based on the current speed of the host vehicle, and this position is used as the lane change position. 3. Based on the predicted trajectory of the host vehicle from the perspective of other vehicles, the position closest to the predicted trajectory of the host vehicle from the perspective of other vehicles is selected as the lane change position. In other embodiments of the present application, other methods may be used to select the lane change position, such as according to the driver's driving habits, etc., which will not be elaborated here.

[0061] After selecting the lane change position, multiple vehicle driving paths are planned for the main vehicle to reach the lane change position, starting from the current position of the main vehicle and ending at the lane change position. The principle of planning the vehicle driving path is to adopt multiple interpretable rules, such as traffic laws and regulations, normal vehicle driving logic, and other vehicle driving conditions. For example, if the direction of vehicle driving is stipulated in traffic laws and regulations, the reverse driving scheme will be avoided when planning the vehicle driving path, which will not be elaborated here.

[0062] After obtaining at least one vehicle driving path, the vehicle driving path is compared with the main vehicle trajectory predicted from the perspective of other vehicles to obtain the deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles. The vehicle driving path with the smaller deviation is used as the driving path.

[0063] Through the settings of this application, the main vehicle trajectory predicted from the perspective of other vehicles is obtained by the deep neural network. Combined with the current road conditions, the lane change position is selected in the current road conditions. The vehicle driving path planned by the lane change position is compared with the main vehicle trajectory predicted from the perspective of other vehicles. The vehicle driving path with the smallest deviation from the main vehicle trajectory predicted from the perspective of other vehicles is selected. The current actual main vehicle driving situation, road conditions, and the prediction results of the deep neural network are integrated, and the real-time traffic flow status is combined with the prediction results of the neural network to accurately predict the vehicle trajectory, which is suitable for complex road conditions.

[0064] Reference Figure 2 , further, step S2 also includes:

[0065] S25: If the lane is not changed and an obstacle is detected ahead;

[0066] S26: Obtaining the intersection point between the main vehicle and the obstacle based on the main vehicle trajectory predicted from the perspective of other vehicles and the obstacle trajectory;

[0067] S27: Determine the avoidance relationship between the host vehicle and the obstacle based on the intersection point, and correct the host vehicle trajectory predicted from the perspective of other vehicles as the driving path based on the avoidance relationship between the host vehicle and the obstacle.

[0068] Specifically, taking unmanned driving as an example, the main vehicle's travel route is: while the main vehicle is driving, the obstacle prediction module predicts obstacles around the main vehicle, such as other vehicles with a slower driving speed than the main vehicle, other vehicles temporarily parked, pedestrians, etc., which means that the main vehicle needs to avoid them or make relevant decisions to change the current driving state. I will not go into details here.

[0069] The avoidance relationship is that the main vehicle and the obstacle should avoid each other. For example, if the obstacle is a vehicle that is about to turn left into the lane driven by the main vehicle, and it is driving straight according to the turn avoidance rule, it should avoid the obstacle for the main vehicle, and the main vehicle should maintain its normal driving path. If the obstacle is another vehicle in front of the main vehicle's driving path, and the speed of the other vehicle is lower than that of the main vehicle, then the main vehicle should avoid the other vehicle. If the main vehicle can change lanes, it will choose to change lanes. Then, the lane change is planned according to the position of the other vehicle in the main vehicle's trajectory predicted from the perspective of other vehicles, and the driving path is generated according to the planned path. In other cases, the corresponding avoidance relationship is set according to the form, and the rules for modifying the main vehicle's trajectory predicted from the perspective of other vehicles to generate a driving path corresponding to the avoidance relationship are not elaborated here.

[0070] An example in this embodiment is as follows:

[0071] 1. First, decide whether to change lanes

[0072] a. Convert the predicted trajectory of the main vehicle from the perspective of other vehicles into a 0 / 1 binary variable indicating whether to change lanes;

[0073] b. This binary prediction is weighted and added to the value map (the value map is the core basis for determining the path in classic path planning). In this application, the value map represents the long-term benefits of the main vehicle under various different decisions.

[0074] 2. If the host vehicle wants to change lanes, its travel route is:

[0075] a. By dividing the lane-changing traffic flow into several lane-changing windows, for example, if there are three cars in the future lane, there will be four lane-changing windows.

[0076] b. Sort by the distance between the midpoint of each lane change window and the entry point of the main vehicle's lane change trajectory predicted from the perspective of other vehicles, and delete the lane change windows whose distance is greater than a certain threshold.

[0077] c. For the remaining lane change windows, assign different preferences based on the distance between the midpoint of the lane change window and the entry point of the main vehicle's trajectory predicted from the perspective of other vehicles (the smaller the distance, the greater the preference reward). The lane change window with the largest preference reward is selected as the desired lane change window. Then, based on the position of the desired lane change window, the main vehicle's trajectory predicted from the perspective of other vehicles is adjusted to form the driving route.

[0078] 3. If you want to change lanes and have selected a desired lane change window that matches the lane change window predicted by the host vehicle trajectory model:

[0079] The trajectory of the main vehicle predicted from the perspective of other vehicles can be used as the driving path.

[0080] 4. If the vehicle does not change lanes and the autonomous vehicle's obstacle prediction module detects a possible obstacle ahead, the vehicle's route is:

[0081] a. Obtain the intersection point based on the main vehicle's trajectory and the obstacle's predicted trajectory predicted from the perspective of other vehicles;

[0082] b. Determine the avoidance relationship between the host vehicle and the obstacle based on the intersection point;

[0083] c. Based on the avoidance relationship, the main vehicle trajectory predicted from the perspective of other vehicles is corrected as the driving route.

[0084] Optionally, in another embodiment, the deep neural network is vectornet.

[0085] Compared with the previous image + Conv encoding method, VectorNet greatly reduces the training parameters of the neural network and has better results in actual vehicle trajectory prediction.

[0086] Furthermore, VectorNet based on obstacle training is used to predict the trajectory of the main vehicle, and the predicted trajectory of the main vehicle from the perspective of other vehicles is obtained.

[0087] Furthermore, VectorNet is an improved VectorNet, which includes a network structure that adds time and space alternating aggregation to VectorNet, and uses the previous prediction output as the input of the next prediction to decode the trajectory predicted by the improved VectorNet. Obstacle training is performed on the improved VectorNet.

[0088] Information is aggregated alternately in the time dimension and the space dimension to characterize the extension of information in the time dimension and the interaction between different obstacles in the space dimension, and the decoding of the trajectory is strengthened to make it more in line with physical intuition.

[0089] The specific network structure is as follows:

[0090] 1: Use a sliding window to divide the time interval in the time dimension, for example, 1-100 minutes, starting with a value interval of 5 minutes. The value interval is 10 minutes, and the sliding window is 1-10 minutes, 5-15 minutes, 10-20 minutes... and so on.

[0091] 2: Use a shared weight network layer (such as an attention model) on each sliding window to aggregate information.

[0092] 3: Use a network (such as a transformer model) to aggregate the features of different obstacles (including the main vehicle) in space to characterize the interactions between obstacles in space.

[0093] 4: Aggregate the spatially aggregated obstacle features temporally (for example, using the LSTM network).

[0094] 5: The decoding part uses the "per step" idea, for example, generating the first-second position, and then using the first-second position prediction as the input for the second-second position prediction to obtain a complete predicted trajectory.

[0095] Through the settings of this implementation, the prediction accuracy of VectorNet is improved.

[0096] Furthermore, in this embodiment, acceleration, turn signals, length and width of the car are added as features of VectorNet. Figure 3 This is the space-first vectornet model in this implementation.

[0097] Reference Figure 4 , optionally, in another embodiment, S22 includes:

[0098] S221: A lane change window is set between each two adjacent vehicles on the road to be changed;

[0099] S222: Selecting a desired lane change window as a lane change position based on the host vehicle's trajectory predicted from the perspective of other vehicles.

[0100] Specifically, in a lane-changing scenario, the main vehicle sets a lane-changing window between every two vehicles on the road to which it is about to change lanes. The nearest lane-changing window is selected based on the main vehicle's trajectory predicted from the perspective of other vehicles. The predicted future trajectory result is the one that is closest to the current driving conditions of the main vehicle and surrounding vehicles. The nearest lane-changing window is selected based on the main vehicle's trajectory predicted from the perspective of other vehicles, minimizing the impact of the main vehicle's lane change on the originally predicted driving trajectory of the main vehicle and reducing the impact on the driving trajectories of other vehicles. The smaller the variable, the higher the accuracy.

[0101] Reference Figure 5 , optionally, in another embodiment, S222 includes:

[0102] S2221: Filtering optional windows based on the current driving situation of the main vehicle;

[0103] S2222: Based on the predicted trajectory of the host vehicle from the perspective of other vehicles, the nearest lane change window is selected from the optional windows as the lane change position.

[0104] Specifically, the current driving situation of the host vehicle includes factors such as its relative position to other vehicles and the speed difference between them. Based on this situation, inappropriate lane change windows can be filtered out. For example, lane change windows located behind the host vehicle's direction of travel or when the host vehicle is accelerating relative to other vehicles can be filtered out. Lane change windows whose distance from the host vehicle is less than the distance calculated based on the speed difference between the host vehicle and other vehicles can be filtered out. By filtering available windows based on the current driving situation of the host vehicle, redundant calculations can be reduced.

[0105] Reference Figure 6 Optionally, in another embodiment, S2221 is S2221': obtain the current driving condition of the main vehicle. If the main vehicle is currently in an accelerating state, preset the time point at which the lane change operation occurs, calculate the lane change position of the main vehicle at the time point under the current acceleration and the positions of each lane change window, and filter out the optional windows.

[0106] Specifically, in this embodiment, the current driving state of the host vehicle is an accelerating state, indicating that the current speed of the host vehicle will change. A preset time for the host vehicle to change lanes is set. In this embodiment, the preset time for the host vehicle to change lanes is the time after which the host vehicle will change lanes. The preset method can be: 1. Set based on safe driving conditions. For example, after the driver triggers a lane change request, a determination is made based on safe driving conditions to determine whether the surrounding conditions satisfy the lane change request. The time until the moment when a lane change condition is determined to exist or a preset time after the determination of the lane change condition is set as the time for the host vehicle to change lanes; 2. Set based on historical experience to a time after a preset period of time after the lane change request is triggered. In other embodiments, other methods for calculating the preset time for the host vehicle to change lanes may be used, which are not detailed here.

[0107] Calculate the lane change position of the host vehicle at that time point under the current acceleration and the position of each lane change window. When the host vehicle accelerates, the relative position of the host vehicle and other vehicles will change. If there are lane change windows behind the host vehicle's forward direction after acceleration, these lane change windows can be omitted and not calculated. Optional windows are obtained to reduce the amount of calculation.

[0108] Reference Figure 7 Optionally, in another embodiment, S24 is S24': selecting a desired driving path based on human driving data and a deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles.

[0109] Specifically, in this embodiment, a scoring method is used to select the desired driving path. In other embodiments of the present application, a comprehensive evaluation method, mathematical analysis, etc. can be used to select the desired driving path in combination with human driving data and the deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles. These will not be elaborated here.

[0110] In this embodiment, human driving data is pre-collected driving choices of people in different driving situations, that is, driving data that meets human expectations. In this embodiment, it is reflected in the combination of driving safety data and user comfort data of different driving methods. In other embodiments of the present application, it can also be other driving data content that meets human expectations.

[0111] The following is an example of this implementation:

[0112] The driving safety data of each vehicle's driving path is scored. For example, the full score for the driving safety data of each vehicle's driving path is set to 10. For each lane change on the vehicle's driving path, one point is deducted; each time acceleration is required, 0.5 points are deducted; and each additional vehicle that needs to be overtaken is deducted 0.5 points. The final score is calculated as the driving safety data score.

[0113] The user comfort data of each vehicle driving path is scored. For example, the user comfort data of each vehicle driving path is initially scored to 10 points. When the user changes lanes to the right, the normal comfort driving state is to turn the vehicle right. One point will be deducted for each left turn required on each vehicle driving path. The normal comfort state is no acceleration. One point will be deducted for each additional acceleration.

[0114] The driving condition of the main vehicle corresponding to each vehicle driving path is scored. For example, the driving condition of the main vehicle corresponding to each vehicle driving path is initially scored to ten points. One point is deducted for each time acceleration is required, each time lane change is required, and each time deceleration is required. In other words, each situation on each vehicle driving path that requires changing the driving status of the main vehicle is scored.

[0115] The deviation between the vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles is scored. For example, the deviation between the vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles is initially scored out of ten points. Multiple deviation ranges are preset. If the deviation is within the first deviation range, one point is deducted; if the deviation is within the second deviation range, two points are deducted, and so on. Points are deducted for vehicle driving paths within different deviation ranges.

[0116] After scoring the human driving data and the deviation between the vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles, the driving path is selected based on the total score, with the option with the highest total score being given priority.

[0117] Optionally, in another embodiment, the method for calculating the deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles is: respectively calculating the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the direction perpendicular to the main vehicle's driving direction, and / or the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation.

[0118] Specifically, the difference between the driving path of each vehicle and the main vehicle trajectory predicted from the perspective of other vehicles can be calculated by calculating the distance between each vehicle's driving path and the main vehicle trajectory predicted from the perspective of other vehicles in a direction perpendicular to the main vehicle's driving direction, or the distance between each vehicle's driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation, or the difference between each vehicle's driving path and the main vehicle trajectory predicted from the perspective of other vehicles in a direction perpendicular to the main vehicle's driving direction and the distance between each vehicle's driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation.

[0119] Taking the distance between each vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles in the perpendicular direction of the main vehicle's travel direction (lateral deviation) and the distance between each vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles in the same direction of the main vehicle's travel direction (longitudinal deviation) as an example, the deviation is calculated as follows: lateral deviation + alpha * longitudinal deviation, where alpha is the parameter of longitudinal influence relative to lateral influence.

[0120] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0121] In one embodiment of the present application, a host vehicle motion planning system is provided, which corresponds one-to-one to the host vehicle motion planning method in the above embodiment. The host vehicle motion planning system includes:

[0122] The path prediction module is used to obtain the historical trajectory of the main vehicle, the historical trajectory of other vehicles, and the road network structure, and input them into the deep neural network to obtain the predicted trajectory of the main vehicle from the perspective of other vehicles;

[0123] The path planning module is used to generate a driving path by combining the main vehicle's trajectory predicted from the perspective of other vehicles and the main vehicle's travel route;

[0124] Furthermore, in another embodiment, the deep neural network is a modified vectornet, which is a network structure that adds time-space alternating aggregation to vectornet, and uses the previous prediction output as the input of the next prediction to decode the trajectory predicted by the modified vectornet, and obstacle training is used for the modified vectornet.

[0125] Furthermore, in another embodiment, the path planning module is also used to determine whether the main vehicle changes lanes; if so, obtain the lane change position of the current lane change; plan at least one vehicle driving path based on the selected lane change position; and select the required driving path based on the deviation between the at least one vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles.

[0126] Furthermore, in another embodiment, the first selection module is further configured to set a lane change window between each two adjacent vehicles on the road to be changed; and select the required lane change window as the lane change position based on the main vehicle's trajectory predicted from the perspective of other vehicles.

[0127] Furthermore, in another embodiment, selecting a desired lane change window as the lane change position based on the host vehicle's trajectory predicted from the perspective of other vehicles includes: filtering optional windows based on the current driving conditions of the host vehicle; and selecting the nearest lane change window from the optional windows as the lane change position based on the host vehicle's trajectory predicted from the perspective of other vehicles.

[0128] Furthermore, in another embodiment, the path planning module is also used to, if the lane does not change and an obstacle is sensed ahead, obtain the intersection point of the main vehicle and the obstacle based on the main vehicle trajectory predicted from the perspective of other vehicles and the predicted obstacle trajectory; determine the avoidance relationship between the main vehicle and the obstacle based on the intersection point, and correct the main vehicle trajectory predicted from the perspective of other vehicles into a driving path based on the avoidance relationship between the main vehicle and the obstacle.

[0129] Furthermore, in another embodiment, the optional windows are filtered out according to the current driving condition of the main vehicle. To obtain the current driving condition of the main vehicle, if the current main vehicle is in an accelerating state, the time point at which the lane change operation occurs is preset, and the lane change position of the main vehicle reaching the time point under the current acceleration and the positions of each lane change window are calculated to filter out the optional windows.

[0130] Furthermore, in another embodiment, the second selection module is further configured to select a desired driving path based on human driving data and a deviation between the vehicle driving path and the host vehicle trajectory predicted from the perspectives of other vehicles.

[0131] Furthermore, in another embodiment, the method for calculating the deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles is: respectively calculating the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the perpendicular direction of the main vehicle's driving direction, and / or the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation.

[0132] Each module of the aforementioned host vehicle motion planning system can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor within a computer device in hardware form, or stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0133] In one embodiment of the present application, a computer device is provided, which may be a server. The computer device includes a processor, memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Volatile or non-volatile storage devices include, but are not limited to: magnetic disks, optical disks, EEPROM (Electrically Erasable Programmable Read Only Memory), EPROM (Erasable Programmable Read Only Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory), magnetic storage, flash memory, and PROM (Programmable Read-Only Memory). The memory of the computer device provides an environment for the operation of the operating system and computer programs stored therein. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by a processor, the steps of the main vehicle motion planning method described in the above embodiment are implemented.

[0134] In one embodiment of the present application, a computer-readable storage medium is provided, storing a computer program. When executed by a processor, the computer program implements the steps of the host vehicle motion planning method described in the above embodiment. The computer-readable storage medium includes a ROM (Read-Only Memory), a RAM (Random-Access Memory), a CD-ROM (Compact Disc Read-Only Memory), a magnetic disk, a floppy disk, and the like.

[0135] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, the division of the above-mentioned functional units and modules is only used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system described in this application is divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A method for motion planning of a main vehicle, characterized in that: The method comprises: Obtain the historical trajectory of the main vehicle, the historical trajectory of other vehicles, and the road network structure, input them into the deep neural network, and obtain the trajectory of the main vehicle predicted from the perspective of other vehicles; The main vehicle's trajectory is combined with the main vehicle's travel route predicted from the perspective of other vehicles to generate the main vehicle's driving path; The driving path generated by combining the trajectory of the main vehicle predicted from the perspective of other vehicles and the travel route of the main vehicle includes: Determine whether the host vehicle changes lanes; If changing lanes, obtain the lane change position of the current lane change; planning at least one vehicle driving path based on the selected lane change location; Selecting a desired driving path based on the deviation between at least one vehicle driving path and the host vehicle trajectory predicted from the perspective of other vehicles; The desired driving path is selected based on the deviation between at least one vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles, including: The desired driving path is selected based on preset human driving data and the deviation between the vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles; The deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles is calculated by calculating the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in a direction perpendicular to the main vehicle's driving direction, and / or the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation.

2. The method for motion planning of a host vehicle according to claim 1, characterized in that: The deep neural network is a modified vectornet, which is a network structure that adds time and space alternating aggregation to vectornet, and uses the previous prediction output as the input of the next prediction to decode the trajectory predicted by the modified vectornet.

3. The method for motion planning of a host vehicle according to claim 1, characterized in that: The obtaining of the current lane change position includes: A lane change window is set between each two adjacent vehicles on the road to be changed; The desired lane change window is selected as the lane change position based on the main vehicle trajectory predicted from the perspective of other vehicles.

4. The method for motion planning of a host vehicle according to claim 3, characterized in that: The selecting of the required lane change window as the lane change position according to the host vehicle trajectory predicted from the perspective of other vehicles includes: Filter out optional windows based on the current driving situation of the main vehicle; Based on the main vehicle's trajectory predicted from the perspective of other vehicles, the nearest lane change window is selected from the optional window as the lane change position.

5. The method for host vehicle motion planning according to claim 4, characterized in that: The optional windows selected according to the current driving condition of the host vehicle include: The current driving status of the host vehicle is obtained. If the host vehicle is currently accelerating, the time point at which the lane change operation occurs is preset. The lane change position of the host vehicle at that time point under the current acceleration and the positions of each lane change window are calculated to filter out the optional windows.

6. The method for motion planning of a host vehicle according to claim 1, characterized in that: The driving path generated by combining the main vehicle trajectory predicted from the perspectives of other vehicles and the main vehicle's travel route also includes: If you do not change lanes and detect an obstacle ahead; Obtain the intersection point between the main vehicle and the obstacle based on the main vehicle trajectory predicted from the perspective of other vehicles and the obstacle trajectory; The avoidance relationship between the main vehicle and the obstacle is determined based on the intersection point, and the main vehicle trajectory predicted from the perspective of other vehicles is corrected as the driving path based on the avoidance relationship between the main vehicle and the obstacle.

7. A main vehicle motion planning system, characterized in that: The system comprises: The path prediction module is used to obtain the historical trajectory of the main vehicle, the historical trajectory of other vehicles, and the road network structure, and input them into the deep neural network to obtain the predicted trajectory of the main vehicle from the perspective of other vehicles; The path planning module is used to generate a driving path by combining the main vehicle's trajectory predicted from the perspective of other vehicles and the main vehicle's travel route; The driving path generated by combining the trajectory of the main vehicle predicted from the perspective of other vehicles and the travel route of the main vehicle includes: Determine whether the host vehicle changes lanes; If changing lanes, obtain the lane change position of the current lane change; planning at least one vehicle driving path based on the selected lane change location; Selecting a desired driving path based on the deviation between at least one vehicle driving path and the host vehicle trajectory predicted from the perspective of other vehicles; The desired driving path is selected based on the deviation between at least one vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles, including: The desired driving path is selected based on preset human driving data and the deviation between the vehicle's driving path and the main vehicle's trajectory predicted from the perspective of other vehicles; The deviation between the vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles is calculated by calculating the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in a direction perpendicular to the main vehicle's driving direction, and / or the distance between each vehicle driving path and the main vehicle trajectory predicted from the perspective of other vehicles in the same direction as the main vehicle's driving direction to obtain the deviation.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the main vehicle motion planning method according to any one of claims 1 to 6 are implemented.

Images