Multi-modal trajectory prediction method based on target driving, vehicle, equipment and medium

Abstract

The invention discloses a multi-modal trajectory prediction method based on target driving, a vehicle, equipment and a medium, and the method comprises the steps: obtaining a high-precision automatic driving data set which comprises trajectory data of road participants and a vectorized high-precision map, obtaining trajectory features and map features, and fusing the features through LaneGCN and a space attention layer; predicting Ke candidate targets # imgabs0 # for each road participant, and calculating a confidence score; screening out a candidate target with the highest confidence score as a prediction target, regarding a region within a certain range of the prediction target as a GOI region, and coding the GOI region to obtain a final feature; predicting Ka prediction trajectories and confidence degree scores corresponding to the trajectories for each road participant by using the final features; and after training is completed, selecting the prediction trajectory with the highest confidence score as a final trajectory prediction result. According to the method, complex and uncertain traffic conditions can be better processed, and the accuracy of track prediction is improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent driving technology, and specifically relates to a target-driven multimodal trajectory prediction method, vehicle, equipment and medium. Background Art

[0002] Autonomous driving technology is an innovative technology that uses computer and sensor technology to enable cars to drive independently. The goal of this technology is to enable vehicles to drive safely without human driver intervention, thereby improving traffic efficiency, reducing traffic accidents and providing a more convenient way to travel. Trajectory prediction plays a key role in autonomous driving. It is an important component of the autonomous driving system and helps achieve safe, efficient and intelligent autonomous driving.

[0003] Trajectory prediction technology is mainly used to predict the future driving trajectory of a vehicle so that the vehicle can make corresponding decisions and driving control. In an autonomous driving system, trajectory prediction is not only related to the safety and stability of vehicle driving, but is also closely related to the vehicle's path planning, obstacle avoidance and other functions. Only through accurate trajectory prediction can a vehicle better avoid traffic obstacles, predict the driving intentions of other vehicles, maintain a safe driving distance and speed, etc. Therefore, trajectory prediction technology is of great significance to the performance and reliability of autonomous driving systems. This technology has a wide range of applications in various fields, including autonomous driving, drone navigation, logistics, and aerospace.

[0004] Previous trajectory prediction technologies only focused on the historical trajectory and driving environment of the vehicle, and mainly focused on studying the interaction between vehicles and the interaction between vehicles and the environment within the observation time window. However, they rarely explored the driving behavior of the vehicle in the prediction time window, such as the vehicle making an evasive move at some point in the future because of compliance with driving ethics. In addition, in a real driving environment, human intentions are very complex. For example, at an intersection, the driver may have multiple options, such as stopping, accelerating, or turning, and different behaviors may lead to completely opposite results, which makes it very important to understand the behavioral intention for future trajectory prediction.

[0005] With the development of autonomous driving and related fields, the requirements for trajectory prediction technology are becoming increasingly higher, especially when solving complex, diverse and uncertain traffic scenarios. Goal-driven multimodal trajectory prediction technology has become the forefront of this field and plays a key role in meeting these high requirements.

[0006] Target-driven multimodal trajectory prediction technology is an advanced trajectory prediction method that aims to more accurately predict the future trajectory of mobile targets (such as vehicles, pedestrians, bicycles, etc.) to improve decision-making and safety in autonomous driving and related fields. The core idea of ​​this technology is to combine multimodal data (such as sensor data) with the intention and possible behavior of the target to generate multiple possible trajectory predictions. These trajectory predictions can cover a variety of possibilities in different situations, helping the system to better understand the surrounding environment and make smarter decisions. Multimodal trajectory prediction technology has multiple advantages such as data fusion, diversity, uncertainty processing, target intention understanding, real-time, interactive considerations and a wide range of application areas, making it one of the key technologies in autonomous driving and related fields. This technology helps to improve the intelligence and safety of transportation systems, thereby providing more possibilities for future mobility. Summary of the invention

[0007] The purpose of the present invention is to provide a target-driven multimodal trajectory prediction method, vehicle, device and medium, which can more accurately predict the behavior of other vehicles, pedestrians and bicycles, so that the autonomous driving system can better avoid potential dangerous situations, thereby improving road safety and reducing traffic accidents.

[0008] According to a first aspect of the present invention, a multi-modal trajectory prediction method based on target driving is provided, the method comprising:

[0009] Obtain a high-precision autonomous driving dataset, which includes trajectory data of road participants and vectorized high-precision maps;

[0010] The historical trajectory data S of road participants are analyzed by U-Net obs Perform feature extraction to obtain trajectory features H s ;

[0011] Construct a lane graph based on the vectorized high-precision map, and use LaneGCN to perform convolution operations on the lane graph to obtain the map feature H g ;

[0012] Using LaneGCN and spatial attention layer to fuse trajectory features H s and map feature H g , to capture all interactive information between road participants and roads, and obtain the fused feature x;

[0013] Based on the fused feature x, K is predicted for each road participant. e Candidate targets And apply MLP to each candidate target Calculate the confidence score and use the smoothe1 loss function to calculate the error between the best candidate target and the true target;

[0014] The candidate target with the highest confidence score is selected as the prediction target, and the area within a certain range of the prediction target is regarded as the GOI area. The GOI area is encoded to obtain the final feature X;

[0015] The final feature X is used to predict K for each road participant. a The predicted trajectories and the confidence scores corresponding to the trajectories are calculated; the classification error is calculated by applying the max-margin loss function, and the displacement error between all predicted trajectories and their corresponding ground actual trajectory coordinates at each time step is calculated by applying the smoothe1 loss function;

[0016] After the training is completed, the predicted trajectory with the highest confidence score is selected as the final trajectory prediction result.

[0017] In the above scheme, the method further comprises:

[0018] Preprocess the data, clean and process the raw data, deal with missing values, outliers and noise, and perform coordinate system conversion and unit standardization.

[0019] In the above scheme, the historical trajectory data S of road participants are analyzed by U-Net. obs Perform feature extraction to obtain trajectory features H s ,include:

[0020] The trajectory data of road participants are divided into observed historical trajectory data S obs And the actual trajectory data S that needs to be predicted pred , use U-Net encoder to analyze the historical trajectory data S obs Perform feature extraction, and then use the U-Net decoder to map the features extracted by the encoder back to the form of original data to obtain the trajectory feature H s .

[0021] In the above scheme, all interaction information between road participants and roads includes: road participant to road A2L, road to road L2L, road to road participant L2A and road participant to road participant A2A.

[0022] In the above scheme, based on the fused feature x, K is predicted for each road participant. e Candidate targets And apply MLP to each candidate target Calculate the confidence score and use the smooth l1 loss function to calculate the error between the best candidate target and the true target, including:

[0023] Applying Transformer Generate a set of candidate targets for each road participant:

[0024]

[0025] Among them, G is the candidate target set, N is the number of road participants, represents the jth candidate target of the i-th road participant;

[0026] In order to make the candidate targets closer to the real targets, a classification branch consisting of MLP M(·) is designed to output the confidence score corresponding to each candidate target:

[0027]

[0028] Among them, C is the confidence score set, represents the confidence score of the jth candidate target of the i-th road participant;

[0029] Calculate the smoothe1 loss between the candidate target and the true target to further improve the accuracy of the predicted target:

[0030]

[0031] Among them, ∈ represents margin, It represents the one with the best prediction result among the candidate targets, and r represents the number of candidate targets;

[0032] During the regression process, smooth l1 loss is applied to calculate the positive target:

[0033]

[0034] in, represents the last coordinate position of the real trajectory of the ith road participant, reg(z)=∑ i d(z i ):

[0035]

[0036] Among them, z i represents smoothe1 loss, ‖z i ‖ represents z i The l1 norm of .

[0037] In the above scheme, the candidate target with the highest confidence score is selected as the prediction target, and the area within a certain range of the prediction target is regarded as the GOI area. The GOI area is encoded to obtain the final feature X, including:

[0038]

[0039] Among them, the i'th feature of the i-th road participant is the traffic feature extracted with the road participant as the center, and x i Indicates; the j'th feature value of the jth lane node is the environmental feature of the current traffic scene, expressed as y j Represents; weight matrix is ​​represented by W i Indicates that φ i Represents the layer normalization of the ReLU function, △ i,j =MLP(v i -v j ), v i represents the i'th coordinate of the i-th road participant, v j Represents the j'th coordinate of the j'th lane node; concat() means splicing.

[0040] In the above scheme, the final feature X is used to predict K for each road participant. a The predicted trajectories and the confidence scores corresponding to the trajectories are calculated; the classification error is calculated by applying the max-margin loss function, and the displacement error between all predicted trajectories and their corresponding ground actual trajectory coordinates at each time step is calculated by applying the smooth l1 loss function, including:

[0041] A linear layer is used to train the K a The predicted trajectories are regressed:

[0042]

[0043] in, represents the predicted coordinates of the kth mode of the nth road participant at the i-th time step; the classification branch outputs a confidence score for each predicted trajectory:

[0044]

[0045] In the regression process, the displacement error between all predicted trajectories and their corresponding true coordinates at each time step is calculated by applying smooth l1 loss:

[0046]

[0047] In the formula, represents the real coordinates of the nth road participant at the tth time step, and T represents the total number of time steps;

[0048] The model is trained end-to-end using classification and regression loss functions:

[0049] l=l cls +l reg

[0050] In the formula, l represents the total loss, l cls represents the classification loss, l reg represents the regression loss.

[0051] According to a second aspect of the present invention, a vehicle is provided, comprising any one of the steps of the target-driven multi-modal trajectory prediction method described above.

[0052] According to a third aspect of the present invention, an electronic device is provided, comprising: a processor and a memory, the memory storing a program or instruction that can be executed on the processor, and the program or instruction, when executed by the processor, implements the steps of any one of the above-mentioned target-driven multimodal trajectory prediction methods.

[0053] According to a fourth aspect of the present invention, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the target-driven multimodal trajectory prediction method described in any one of the above are implemented.

[0054] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0055] (1) The present invention introduces target region coding to encode the scene context and motion trajectory in the vicinity of the predicted target, and fuses the features in the target of interest (GOI) region to help better understand the target's intention.

[0056] (2) The present invention predicts multiple candidate targets and a set of future trajectories to reflect different target behaviors or intentions, which helps the system adapt to diverse traffic conditions and improves decision-making flexibility.

[0057] (3) By estimating and modeling uncertainty, the present invention enables the system to better handle complex and uncertain traffic conditions and improves the robustness of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of the target-driven multi-modal trajectory prediction method;

[0059] Figure 2 A flowchart of the multimodal trajectory prediction process;

[0060] Figure 3 Diagram of the multimodal trajectory prediction network framework. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0062] The present invention solves three main technical problems:

[0063] (1) The present invention uses U-Net to process vehicle trajectories. The U-Net network retains the contextual information of the trajectory data through an encoder-decoder structure, and can capture the global features in the trajectory data, thereby better understanding the overall structure of the trajectory. In addition, the network structure can process features of different scales at the same time, which is very useful for processing vehicle trajectories because trajectory data may include information of different scales, such as slight changes and large-scale movements of the vehicle. Therefore, using the U-Net network to process vehicle trajectory data can make full use of its outstanding performance in the field of image segmentation, thereby improving the efficiency and performance of trajectory data processing and helping the autonomous driving system to better understand.

[0064] (2) The present invention generates multiple candidate targets for road participants. In real driving environments, human intentions are very complex. For example, at an intersection, a driver may have multiple options, such as stopping, accelerating, or turning, and different behaviors may lead to completely opposite results. This makes it very important to understand their behavioral intentions for future trajectory prediction. By understanding the intentions of others, more reasonable predictions can be made. Therefore, in the target prediction stage, multiple candidate targets are predicted for road participants through Transformer (the final destination expected by the given participant within the prediction time window), aiming to solve the participant's intention problem as accurately as possible.

[0065] (3) Previous trajectory prediction technologies only focus on the historical trajectory and driving environment of the vehicle, and mainly focus on studying the interaction between vehicles and the interaction between the vehicle and the environment within the observation time window. However, they rarely explore the driving behavior of the vehicle in the prediction time window, such as the vehicle making an evasive action at some point in the future due to compliance with driving ethics. In order to better understand the driving intention of the vehicle, the present invention selects the candidate target with the highest confidence score as the prediction target, that is, the destination that the vehicle is most likely to reach, and regards the area within 6 meters of the prediction target as the GOI area, encodes the GOI area, updates the road and vehicle features within the range, and aggregates the features.

[0066] like Figure 1 , Figure 2 and Figure 3 As shown in the figure, the present invention first performs preprocessing operations on the data, cleans and processes the raw data, handles missing values, outliers and noise; performs coordinate system conversion and unit standardization to ensure data consistency; and constructs a lane graph based on the vectorized high-precision map. The first stage involves extracting corresponding features from the data: First, the vehicle historical trajectory S is analyzed by U-Net. obs Perform feature extraction to obtain trajectory features H s ; Then convolution operation is performed through LaneGCN to obtain map features H g The second stage uses the spatial attention mechanism to fuse the trajectory features H s and map feature H g To obtain the interactive information between the road and the vehicle, the fused feature x is obtained. The third stage is to sample candidate targets. First, the Transformer predicts K for each vehicle. e Potential candidate targets Apply MLP to each Calculate its confidence score and use the smooth l1 loss function to calculate the error between the candidate target and the real target; finally, select the target with the highest confidence score as the predicted target, that is, the destination that the agent is most likely to reach, and regard the area within 6 meters of the predicted target as the GOI area. After encoding the GOI area, the final feature X is obtained. The fourth stage uses the final agent feature X to predict K for each vehicle. a The classification error is calculated by applying the max-margin loss function, and the displacement error between all predicted trajectories and their corresponding ground actual trajectory coordinates at each time step is calculated by applying the smooth l1 loss function.

[0067] like Figure 1 , Figure 2 and Figure 3 As shown, the specific implementation steps of the present invention are as follows:

[0068] S1: Obtain a high-precision autonomous driving dataset and divide it into training set, validation set, and test set; the dataset contains the vehicle's driving trajectory data and the lane information of the high-precision map.

[0069] S2: Preprocess the data, clean and process the raw data, handle missing values, outliers and noise; perform coordinate system conversion and unit standardization to ensure data consistency.

[0070] S3: Divide the vehicle's trajectory data into observed historical trajectory data S obs and the trajectory S to be predicted pred (i.e., the real trajectory), and then use the U-Net encoder to encode the historical trajectory data S obs Processing is performed to reduce the spatial dimension of the data, and then the decoder is used to map the features extracted by the encoder back to the form of the original data to obtain the final trajectory feature H s ; The collaborative work of the encoder and decoder allows architectures such as U-Net to process multi-scale information and better capture and generate spatiotemporal patterns related to trajectory data; build a lane graph based on the vectorized high-precision map, and use LaneGCN to perform convolution operations on the lane graph to obtain map features H g .

[0071] S4: Considering that the behavior of the target vehicle is affected by the surrounding vehicles and road environment, in order to improve the authenticity of the prediction, a wider range of contextual information needs to be considered. To do this, an interaction fusion module is introduced, which uses LaneGCN and spatial attention layers to capture interaction information at different levels, including four feature attention modules: A2L (vehicle to road), L2L (road to road), L2A (road to vehicle), and A2A (vehicle to vehicle), thereby providing a more comprehensive feature x for the target prediction module.

[0072] S5: In the target prediction network module, K is predicted for each road participant. e potential goals (i.e., the desired final destination for a given participant within the prediction time window).

[0073] First apply Transformer Generate a set of potential targets for each road participant:

[0074]

[0075] Where N is the number of participants in the process, represents the jth potential target of the i-th road participant.

[0076] In order to make the predicted target closer to the real target, the present invention designs a classification branch consisting of MLP M(·) to output the confidence score corresponding to the potential target:

[0077]

[0078] in, represents the confidence score of the jth potential target of the i-th road participant.

[0079] In addition, the present invention calculates the smooth l1 loss between the best predicted target and the ground truth target to further improve the accuracy of the predicted endpoint:

[0080]

[0081] Among them, ∈ represents margin, Indicates the candidate target with the best prediction result.

[0082] During the regression process, smooth l1 loss is applied to calculate the positive target:

[0083]

[0084] in, represents the last coordinate position of the real trajectory of the ith road participant, reg(z)=∑ i d(z i ):

[0085]

[0086] Among them, z i represents smooth l1 loss, ‖z i ‖ represents z i The e1 norm of .

[0087] Subsequently, the present invention selects the target with the highest confidence as the predicted target, that is, the destination that the vehicle is most likely to reach, and regards the area within 6 meters of the predicted target as the GOI area. In addition, the behavior of the vehicle is closely related to the historical trajectory and driving environment. Previous work mainly focused on studying the interaction between agents within the observation time window, while ignoring the interaction between the vehicle and the map in the GOI area. Therefore, the GOI area is encoded and the features are aggregated as follows:

[0088]

[0089] Among them, the i′th feature of the i-th vehicle is the traffic feature extracted with the vehicle as the center, and x i The j′th feature value of the jth lane node is the environmental feature of the current traffic scene, represented by y j Indicates that the weight matrix is ​​Wi i Indicates; Φ i Represents the layer normalization of the ReLU function, △ i,j =MLP(v i -v j ), v i represents the i'th coordinate belonging to the i-th participant, v jrepresents the j′th coordinate belonging to the jth lane node.

[0090] S6: After integrating traffic flow features and environmental features, the final feature X is obtained, and K is predicted for each road participant. a candidate prediction trajectories and their corresponding confidence scores.

[0091] A linear layer is used to train the K a The predicted trajectories are regressed:

[0092]

[0093] in, represents the predicted coordinates of the kth mode of the nth road participant at the i-th time step. The classification branch outputs a confidence score for each predicted trajectory:

[0094]

[0095] In the regression process, the displacement error between all predicted trajectories and their corresponding true coordinates at each time step is calculated by applying smooth l1 loss:

[0096]

[0097] in, represents the real coordinates of the nth road participant at the tth time step, and T represents the total number of time steps;

[0098] The model is trained end-to-end using classification and regression loss functions:

[0099] l=l cls +l reg

[0100] In the formula, l represents the total loss, l cls represents the classification loss, l reg represents the regression loss.

[0101] Finally, after the training is completed, the current vehicle's driving trajectory data and the lane information of the high-precision map are input, and the candidate prediction trajectory with the highest confidence score is selected as the prediction output.

[0102] The present invention also provides a vehicle, which includes the steps of any one of the target-driven multi-modal trajectory prediction methods described above.

[0103] The present invention also provides an electronic device, comprising: a processor and a memory, wherein the memory stores a program or instruction that can be executed on the processor, and when the program or instruction is executed by the processor, the steps of the target-driven multimodal trajectory prediction method described in any one of the above are implemented.

[0104] The present invention also provides a readable storage medium having a program or instruction stored thereon, and when the program or instruction is executed by a processor, the steps of any one of the above-mentioned target-driven multi-modal trajectory prediction methods are implemented.

[0105] In summary, the present invention provides a target-driven multimodal trajectory prediction method, vehicle, device and medium, which enables the autonomous driving system to better avoid potential dangerous situations by more accurately predicting the behavior of other vehicles, pedestrians and bicycles, thereby improving road safety and reducing traffic accidents.

[0106] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0107] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0108] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A target-driven multimodal trajectory prediction method, characterized in that: The method includes: Obtain a high-precision autonomous driving dataset, which includes trajectory data of road participants and vectorized high-precision maps; The historical trajectory data S of road participants are analyzed by U-Net obs Perform feature extraction to obtain trajectory features H s ; Construct a lane graph based on the vectorized high-precision map, and use LaneGCN to perform convolution operations on the lane graph to obtain the map feature H g ; Using LaneGCN and spatial attention layer to fuse trajectory features H s and map feature H g , to capture all interactive information between road participants and roads, and obtain the fused feature x; Based on the fused feature x, K is predicted for each road participant. e Candidate targets And apply MLP to each candidate target Calculate the confidence score and use the smooth l1 loss function to calculate the error between the best candidate target and the true target; The candidate target with the highest confidence score is selected as the prediction target, and the area within a certain range of the prediction target is regarded as the GOI area. The GOI area is encoded to obtain the final feature X; The final feature X is used to predict K for each road participant. a The predicted trajectories and the confidence scores corresponding to the trajectories are calculated; the classification error is calculated by applying the max-margin loss function, and the displacement error between all predicted trajectories and their corresponding ground actual trajectory coordinates at each time step is calculated by applying the smooth l1 loss function; After the training is completed, the predicted trajectory with the highest confidence score is selected as the final trajectory prediction result.

2. The target-driven multi-modal trajectory prediction method according to claim 1, characterized in that: The method further includes: Preprocess the data, clean and process the raw data, deal with missing values, outliers and noise, and perform coordinate system conversion and unit standardization.

3. The target-driven multi-modal trajectory prediction method according to claim 1, characterized in that: The historical trajectory data S of road participants are analyzed by U-Net obs Perform feature extraction to obtain trajectory features H s ,include: The trajectory data of road participants are divided into observed historical trajectory data S obs And the actual trajectory data S that needs to be predicted pred , use U-Net encoder to analyze the historical trajectory data S obs Perform feature extraction, and then use the U-Net decoder to map the features extracted by the encoder back to the form of original data to obtain the trajectory feature H s .

4. The target-driven multi-modal trajectory prediction method according to claim 1, characterized in that: All interaction information between road participants and roads includes: road participant to road A2L, road to road L2L, road to road participant L2A and road participant to road participant A2A.

5. The target-driven multi-modal trajectory prediction method according to claim 1, characterized in that: Based on the fused feature x, K is predicted for each road participant. e Candidate targets And apply MLP to each candidate target Calculate the confidence score and use the smooth l1 loss function to calculate the error between the best candidate target and the true target, including: Applying Transformer Generate a set of candidate targets for each road participant: Among them, G is the candidate target set, N is the number of road participants, represents the jth candidate target of the i-th road participant; In order to make the candidate targets closer to the real targets, a classification branch consisting of MLP M(·) is designed to output the confidence score corresponding to each candidate target: Among them, C is the confidence score set, represents the confidence score of the jth candidate target of the i-th road participant; Calculate the smooth l1 loss between the best candidate target and the true target to further improve the accuracy of the predicted target: Among them, ∈ represents margin, It represents the one with the best prediction result among the candidate targets, and r represents the number of candidate targets; During the regression process, smooth l1 loss is applied to calculate the positive target: in, represents the last coordinate position of the real trajectory of the ith road participant, reg(z)=∑ i d(z i ): Among them, z i represents smooth l1 loss, ‖z i ‖ represents z i The l1 norm of .

6. The target-driven multi-modal trajectory prediction method according to claim 1, characterized in that: The candidate target with the highest confidence score is selected as the prediction target, and the area within a certain range of the prediction target is regarded as the GOI area. The GOI area is encoded to obtain the final feature X, including: Among them, the i'th feature of the i-th road participant is the traffic feature extracted with the road participant as the center, and x i Indicates; the j'th feature value of the jth lane node is the environmental feature of the current traffic scene, expressed as y j Represents; weight matrix is ​​represented by W i Indicates that φ i Represents the layer normalization of the ReLU function, △ i,j =MLP(v i -v j ), v i represents the i'th coordinate of the i-th road participant, v j Represents the j'th coordinate of the j'th lane node; concat() means concatenation.

7. The target-driven multi-modal trajectory prediction method according to claim 6, characterized in that: The final feature X is used to predict K for each road participant. a The predicted trajectories and the confidence scores corresponding to the trajectories are calculated; the classification error is calculated by applying the max-margin loss function, and the displacement error between all predicted trajectories and their corresponding ground actual trajectory coordinates at each time step is calculated by applying the smoothhl1loss function, including: A linear layer is used to train the K a Regression on the predicted trajectories: in, represents the predicted coordinates of the kth mode of the nth road participant at the i-th time step; the classification branch outputs a confidence score for each predicted trajectory: In the regression process, the displacement error between all predicted trajectories and their corresponding true coordinates at each time step is calculated by applying smooth l1 loss: In the formula, represents the real coordinates of the nth road participant at the tth time step, and T represents the total number of time steps; The model is trained end-to-end using classification and regression loss functions: and=and cls +e reg In the formula, e represents the total loss, e cls represents the classification loss, l reg represents the regression loss.

8. A vehicle, characterized in that: The vehicle includes the steps of the target-driven multi-modal trajectory prediction method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the target-driven multimodal trajectory prediction method described in any one of claims 1 to 7 are implemented.

10. A readable storage medium, characterized in that: A program or instruction is stored thereon, and when the program or instruction is executed by a processor, the steps of the target-driven multi-modal trajectory prediction method described in any one of claims 1 to 7 are implemented.

Images