Parallel iterative joint prediction planning method considering bidirectional interaction characteristics

By employing a parallel iterative joint prediction and planning method, the problem of insufficient coupling between prediction and planning in complex traffic environments for autonomous vehicles is solved, generating safe, smooth, and efficient driving trajectories, thereby improving vehicle safety and traffic efficiency.

CN121982877APending Publication Date: 2026-05-05TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511908478.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-17
Publication Date
2026-05-05

AI Technical Summary

Technical Problem

In existing technologies, autonomous vehicles struggle to fully consider the two-way coupling between prediction and planning in complex and dynamic traffic environments, resulting in untimely trajectory adjustments that affect safety and traffic efficiency, and may even lead to traffic accidents.

Method used

A parallel iterative joint prediction and planning method is adopted. By iteratively executing interactive trajectory prediction and interactive trajectory planning based on a two-way interactive modeling mechanism in parallel within the planning time domain, environmental information and agent interaction are integrated to generate safe, smooth and efficient driving trajectories.

Benefits of technology

It enables accurate modeling of traffic participant behavior in complex and dynamic traffic scenarios, generating safe, smooth, and efficient driving trajectories, thereby improving the overall safety and traffic efficiency of vehicles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121982877A_ABST
    Figure CN121982877A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of trajectory prediction and planning, in particular to a parallel iterative joint prediction planning method considering bidirectional interaction characteristics, and the method comprises the steps: constructing a bidirectional interaction modeling mechanism considering future trajectory interaction; iteratively executing interactive trajectory prediction and interactive trajectory planning based on a bidirectional interactive modeling mechanism in parallel until a preset condition is met, so as to respectively obtain a final planning code and a final prediction code; and decoding the final planning code and the final prediction code to obtain a multi-modal prediction trajectory and a planning trajectory of a planning time domain, and obtaining a final prediction trajectory and a planning trajectory according to the confidence coefficient. Therefore, the problems that the coupling relation between prediction and planning is difficult to fully reflect and the vehicle track cannot be timely adjusted according to the traffic environment change due to the fact that only the one-way influence of prediction on planning is established in the related technology are solved; therefore, the problems of conservative planning, influence on road traffic efficiency, even traffic accidents and the like easily occur in a complex dynamic scene.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of trajectory prediction and planning technology, and in particular to a parallel iterative joint prediction and planning method that considers bidirectional interaction characteristics. Background Technology

[0002] In real-world dynamic traffic environments, especially in complex urban conditions such as intersections and ramps, traffic flow is dense, participants are diverse, and traffic order is relatively chaotic. Various dynamic interactive behaviors are intertwined and superimposed, which significantly increases the difficulty of trajectory planning for autonomous vehicles.

[0003] In related technologies, a sequential “prediction-planning” approach is typically used. This involves first predicting the future state of surrounding traffic participants, and then generating planned trajectories for vehicles based on the prediction results, thereby improving the rationality of the planning results.

[0004] However, the relevant technologies only establish the one-way influence of prediction on planning, and it is difficult to fully consider the two-way coupling relationship between prediction and planning. As a result, the safety and traffic efficiency of assisted driving vehicles are still limited in complex scenarios, resulting in insufficient response of assisted driving vehicles to environmental changes, untimely trajectory adjustment, and decreased safety and reliability of vehicle driving decisions, which urgently need to be solved. Summary of the Invention

[0005] This application provides a parallel iterative joint prediction and planning method that considers bidirectional interaction characteristics, in order to solve the problem in related technologies that only establish the one-way influence of prediction on planning, which makes it difficult to fully reflect the coupling relationship between prediction and planning, and the vehicle trajectory cannot be adjusted in a timely manner according to changes in the traffic environment. As a result, in complex dynamic scenarios, conservative planning may occur, affecting road traffic efficiency and even causing traffic accidents.

[0006] The first aspect of this application provides a parallel iterative joint prediction planning method considering bidirectional interaction characteristics, comprising the following steps: dividing the planning time domain into multiple stages; within the multiple stages, iteratively executing interactive trajectory prediction based on a bidirectional interaction modeling mechanism and interactive trajectory planning based on the bidirectional interaction modeling mechanism in parallel according to the future trajectory encoding of the previous stage, until a preset condition is met, so as to obtain the final planning code and the final prediction code respectively; decoding the final planning code and the final prediction code to obtain the predicted trajectory, planning distance and corresponding confidence level of the planning time domain, and mapping the planning distance to a lateral trajectory, so as to obtain the planning trajectory of the planning time domain according to the predicted trajectory, the confidence level and the lateral trajectory.

[0007] Through the above technical means, the embodiments of this application can obtain the predicted trajectory of surrounding traffic participants and the planned trajectory of the vehicle by executing interactive trajectory prediction and interactive trajectory planning in parallel. This fully integrates the interactive information between intelligent agents and environmental constraints, and realizes accurate modeling of interactive relationships in complex dynamic traffic scenarios. It can guide assisted driving or autonomous driving vehicles to generate a final driving trajectory that is both safe and smooth, while taking into account traffic efficiency and compliance with traffic rules.

[0008] Optionally, in one embodiment of this application, before obtaining the future trajectory code of the previous stage, the method further includes: performing the interactive trajectory prediction and interactive trajectory planning in parallel in the first stage of the planning time domain to obtain the initial planning code and the initial prediction code; and concatenating the initial planning code and the initial prediction code to obtain the future trajectory code of the first stage.

[0009] Through the above technical means, the embodiments of this application obtain future trajectory codes by splicing planning codes and predictive codes. This can integrate the future state information of surrounding traffic participants with the planned trajectory information of the vehicle, thereby forming a high-dimensional feature representation that takes into account both interaction relationships and environmental constraints. At the same time, it can be used for subsequent trajectory generation, decision optimization and safety assessment, enabling assisted driving or autonomous driving vehicles to generate safe, smooth and efficient driving trajectories in complex dynamic traffic scenarios.

[0010] Optionally, in one embodiment of this application, the interactive trajectory prediction based on the bidirectional interactive modeling mechanism includes the following steps: based on the bidirectional interactive modeling mechanism, performing cross-attention calculation on the scene encoding of the prediction target with the prediction query vector to obtain the current environmental relational semantic conditions of the prediction target; performing the cross-attention calculation on the future trajectory encoding of the previous stage with the query vector to obtain the future intention conditions of the agent; concatenating and temporally pooling the current environmental semantic conditions and the future intention conditions to obtain the prediction encoding of the current stage.

[0011] Through the above technical means, the embodiments of this application can perform interactive trajectory prediction based on current environmental information and semantic constraints of the prediction target, so as to obtain the prediction codes of surrounding traffic participants in the planning time domain. It can integrate environmental constraints and multi-agent interaction information to reflect the possible evolution trend of the future state of each traffic participant, thereby providing a reliable feature representation for subsequent trajectory planning, decision optimization and safety assessment.

[0012] Optionally, in one embodiment of this application, the formula for calculating the first cross-attention is: , in, This represents the predicted query vector after the update and integration of environmental information. This indicates cross-attention calculation. This represents the predicted query vector. This represents the scene encoding; The formula for calculating the second cross-attention is: , in, The predicted query vector represents the updated future intention of the fused agent. This represents the encoding of the future trajectory.

[0013] Through the above technical means, the embodiments of this application can calculate the predicted query vector of the agent's future intention after fusing environmental information based on cross-attention calculation, and then perform trajectory prediction. This can enable the prediction results to simultaneously reflect the coupling relationship between environmental constraints and the vehicle's intention, thereby improving the accuracy, coherence and scene adaptability of trajectory prediction.

[0014] Optionally, in one embodiment of this application, the splicing formula is: , in, Indicates intermediate variables. Indicates the first linear layer; This indicates the first concatenation operation; The formula for the time pooling truncation is: , in, Indicates predictive coding. This indicates a max pooling operation. Represents the planning time domain of stage j. This indicates that the data is being extracted along the time dimension.

[0015] Through the above technical means, the embodiments of this application can obtain predictive coding based on max pooling, which can effectively compress feature dimensions while retaining key feature responses, thereby enhancing the predictive coding's ability to represent significant environmental information and motion patterns, and improving the stability and accuracy of subsequent trajectory prediction.

[0016] Optionally, in one embodiment of this application, the interactive trajectory planning based on the bidirectional interactive modeling mechanism includes the following steps: determining the lateral candidate trajectory of the planning target according to a preset rule class strategy, and performing geometric feature encoding on the lateral candidate trajectory to obtain the lateral query vector of the planning target; concatenating and linearly transforming the lateral query vector with the vertical query vector of the planning target, and performing self-attention mechanism calculation to obtain the initial planning query vector of the planning target; and performing the interactive trajectory planning on the initial planning query vector according to the bidirectional interactive modeling mechanism to obtain the planning code for the current stage.

[0017] Through the above technical means, the embodiments of this application can perform interactive trajectory planning based on lateral rules, environmental information interaction and future trajectory interaction to obtain the planning code of the vehicle in the planning time domain. It can comprehensively consider road constraints, surrounding vehicle status and potential behavior, feasibility and safety of candidate lateral trajectories, thereby providing a high-fidelity feature representation for generating the final planned trajectory.

[0018] Optionally, in one embodiment of this application, the formula for the linear change is: , in, Indicates intermediate variables. Indicates the second linear layer; This indicates the second splicing operation. Indicates a horizontal query. Indicates a vertical query; The formula for calculating the self-attention mechanism is as follows: , in, This represents the initial planning query vector. This indicates that the self-attention mechanism is calculated.

[0019] Through the above technical means, the embodiments of this application can calculate the initial planning query vector according to the self-attention mechanism, and can autonomously extract planning-related key features without relying on external information, thereby achieving efficient encoding of the vehicle's current state, historical trajectory features and potential intentions, thus providing a more consistent, stable and scalable feature foundation for subsequent trajectory generation, constraint evaluation and optimization solution.

[0020] Optionally, in one embodiment of this application, the formula for predicting the trajectory is: , in, Indicates decoder; This represents the predicted trajectory; This represents the final predicted code; The formula for the planned distance is: , in, Indicates the planned distance; This represents the final planning code; The formula for the confidence level is: , in, This indicates the confidence level.

[0021] Through the above technical means, the embodiments of this application decode the predictive coding and planning coding according to the decoder to obtain the predicted trajectory of surrounding traffic participants, the planned distance of the vehicle and the corresponding confidence level, so as to form structured feature information, which can be used to generate the final planned trajectory, perform trajectory selection and optimization, and assist driving or autonomous driving vehicles to achieve safe, smooth and efficient longitudinal and lateral motion control in complex dynamic traffic scenarios.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a parallel iterative joint predictive planning method considering bidirectional interaction characteristics provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the principle of a parallel iterative joint predictive planning method considering bidirectional interaction characteristics according to an embodiment of this application. Figure 3 This is a schematic diagram illustrating the planning results for four complex urban conditions according to one embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0025] The following description, with reference to the accompanying drawings, describes a parallel iterative joint predictive planning method considering bidirectional interaction characteristics, according to embodiments of this application. Addressing the technical problems mentioned in the background art, where related technologies only establish a unidirectional influence of prediction on planning, making it difficult to fully reflect the coupling relationship between prediction and planning, and vehicle trajectories cannot be adjusted in a timely manner according to changes in the traffic environment, thus easily leading to conservative planning, affecting road traffic efficiency, and even causing traffic accidents in complex dynamic scenarios, this application provides a parallel iterative joint predictive planning method considering bidirectional interaction characteristics. In this method, interactive trajectory prediction and interactive trajectory planning based on a bidirectional interaction modeling mechanism are iteratively executed in parallel within the planning time domain to obtain the final planning code and the final prediction code, which are then decoded to obtain the predicted trajectory and the planned distance. The system integrates environmental information and agent interaction relationships, enabling it to understand traffic lights, road topology, surrounding vehicle states, and potential behavioral intentions in complex urban scenarios with multiple agents. This ensures the planned trajectory avoids collisions with surrounding vehicles, doesn't cross road boundaries, and prevents unnecessary stops, thus maintaining traffic safety and efficiency, rather than relying solely on unidirectional prediction or fixed rules to generate trajectories. Simultaneously, it guarantees computational efficiency, allowing assisted driving or autonomous driving vehicles to generate safe, smooth, and efficient driving trajectories in complex dynamic traffic scenarios, thereby improving overall vehicle safety and traffic efficiency. This solves the problem that related technologies only establish a unidirectional influence of prediction on planning, making it difficult to fully reflect the coupling relationship between prediction and planning. Vehicle trajectories cannot be adjusted in a timely manner according to changes in the traffic environment, leading to conservative planning, impacting road efficiency, and even causing traffic accidents in complex dynamic scenarios.

[0026] Specifically, Figure 1 This is a flowchart illustrating a parallel iterative joint predictive planning method that considers bidirectional interaction characteristics, provided in an embodiment of this application.

[0027] like Figure 1 As shown, the parallel iterative joint forecasting planning method considering bidirectional interaction characteristics includes the following steps: In step S101, the planning time domain is divided into multiple stages.

[0028] The planning time domain can refer to the future time range covered by trajectory planning, which is used to determine the motion state and decision-making strategy of the assisted driving vehicle within that time period.

[0029] It can be noted that the division method can be equal division or adaptive dynamic division based on road structure characteristics, traffic flow change trends, and uncertainty of interactive subject behavior, so as to better match the planning accuracy requirements under different scenarios; it can be set by those skilled in the art according to the actual situation, and no specific restrictions are made here.

[0030] For example, embodiments of this application can plan the time domain. Divided into multiple stages ,in, .

[0031] In step S102, within multiple stages, interactive trajectory prediction and interactive trajectory planning based on a two-way interactive modeling mechanism are executed in parallel based on the future trajectory encoding of the previous stage until preset conditions are met, so as to obtain the final planning code and the final prediction code respectively.

[0032] In the embodiments of this application, future trajectory encoding may include prediction encoding and planning encoding, which can simultaneously reflect the dynamic feedback relationship between prediction and planning, thereby improving the responsiveness of trajectory generation to complex traffic evolution; interactive trajectory prediction may refer to considering not only the historical motion state of the target traffic participant itself during the prediction process, but also introducing the behavioral correlation and dynamic interaction influence with surrounding vehicles, thereby obtaining future motion trajectory prediction results that are more in line with the real traffic evolution law; interactive trajectory planning may refer to explicitly considering the two-way behavioral influence between the planned vehicle and the surrounding traffic subjects during the trajectory planning process, and improving the trajectory through online evaluation of interactive feedback and iterative correction, so that the generated vehicle trajectory can better adapt to the changes in complex dynamic traffic environment.

[0033] One possible approach is a two-way interactive modeling mechanism that models trajectory prediction and trajectory planning as parallel processes executed within the same stage, jointly modeling their relationship through interactive iteration. Under this mechanism, trajectory prediction can update its predictions of future behavior of surrounding traffic entities in real time based on candidate trajectories generated by trajectory planning. Simultaneously, trajectory planning can also use the dynamic prediction information output by trajectory prediction to synchronously revise its planning strategy. Through this two-way information transmission and iterative update, a closed-loop interactive relationship can be formed between prediction and planning, thereby more accurately depicting the mutual influence between the vehicle's own intentions and the behavior of environmental entities, and improving the synergy and reliability of the overall decision-making / planning model.

[0034] It can be explained that the preset condition can be reaching the final stage N of the planning time domain, or it can be that a certain evaluation index (such as safety margin, spatiotemporal conflict degree, etc.) meets the set threshold within the planning time domain, thereby triggering trajectory output and decision termination. It can be set by those skilled in the art according to the actual situation, and no specific restrictions are made here.

[0035] Optionally, in one embodiment of this application, interactive trajectory prediction based on a bidirectional interactive modeling mechanism includes the following steps: based on the bidirectional interactive modeling mechanism, performing cross-attention calculation on the scene encoding of the prediction target using the prediction query vector to obtain the current environmental semantic conditions of the prediction target; performing cross-attention calculation on the future trajectory encoding of the previous stage using the query vector to obtain the future intention conditions of the agent; and concatenating and temporally pooling the current environmental semantic conditions and the future intention conditions to obtain the prediction encoding of the current stage.

[0036] Among them, the predicted query vector can refer to the query vector related to the future state of the target vehicle; the query vector can be mapped from the target state of the current task (such as the current position of the vehicle, the state of the planning stage, the state of the predicted node, etc.); cross-attention calculation can refer to the cross-matching calculation of a set of query vectors with key vectors and value vectors from another feature domain. By calculating the matching weight and weighted fusion of the value vectors, the association modeling and complementary enhancement between information from different sources can be realized, thereby improving the model's ability to express complex traffic interaction relationships.

[0037] In the embodiments of this application, the current environmental semantic conditions may include, but are not limited to, the relative distance, relative speed, and acceleration relationship with surrounding vehicles; the spatial positional relationship with lane lines, curbs, and obstacles; interaction modes such as merging, overtaking, queuing, and yielding; and may also include, but are not limited to, compliance with traffic regulations such as traffic lights, speed limits, and right-of-way rules. Future intention conditions may include, but are not limited to, the vehicle's task objectives (such as going straight, turning left, changing lanes, and overtaking), road traffic rules (such as traffic lights, yielding, and speed limits), and driving preferences (such as conservative or aggressive driving styles).

[0038] Optionally, in one embodiment of this application, the formula for calculating the first cross-attention can be: , in, This represents the predicted query vector after the update and integration of environmental information. This indicates cross-attention calculation. This represents the predicted query vector. This represents the scene encoding; The formula for calculating the second cross-attention can be expressed as: , in, The predicted query vector represents the updated future intention of the fused agent. This represents the encoding of the future trajectory.

[0039] For example, such as Figure 2As shown, obtaining the prediction code for the current stage in this embodiment may include the following steps: (1) Environmental information interaction: The embodiments of this application can predict query vectors. Scene coding Perform the first cross-attention calculation to obtain the current environmental relationships and semantic constraints.

[0040] (2) Future trajectory interaction: Specifically, embodiments of this application can use query vectors right Future trajectory coding of the stage Perform a second cross-attention computation to explicitly condition the agent’s future intentions.

[0041] (3) Furthermore, embodiments of this application can aggregate the outputs of the aforementioned two interactive functions, and perform time-based extraction and aggregation on them, using the result as... Stage prediction coding It can be represented as: , , in, Indicates a linear layer; Indicates a splicing operation; This indicates a max pooling operation.

[0042] Optionally, in one embodiment of this application, interactive trajectory planning based on a two-way interactive modeling mechanism includes the following steps: determining the lateral candidate trajectory of the planning target according to a preset rule class strategy, and encoding the lateral candidate trajectory with geometric features to obtain the lateral query vector of the planning target; concatenating and linearly transforming the lateral query vector with the longitudinal query vector of the planning target, and performing self-attention mechanism calculation to obtain the initial planning query vector of the planning target; and performing interactive trajectory planning on the initial planning query vector according to the two-way interactive modeling mechanism to obtain the planning code for the current stage.

[0043] Here, lateral candidate trajectories refer to a set of possible lateral movement paths generated by assisted or autonomous driving vehicles, which can be used to describe multiple feasible movement modes of the vehicle within or across lanes within a given planning time domain. Lateral query vectors may include, but are not limited to, lateral candidate trajectories, lane boundaries, and the lateral positions of surrounding vehicles; longitudinal query vectors may include, but are not limited to, the longitudinal velocities and accelerations of the preceding vehicle, following vehicle, and the vehicle itself. Planning query vectors refer to query vectors related to the planned trajectory, and may include, but are not limited to, interaction information and task objectives.

[0044] It is understandable that the preset rule-based strategy can be based on the position of the lane centerline to keep the vehicle as close to the center of the lane as possible; or it can be based on a curve (such as a cubic polynomial or a Bézier curve) to plan the lateral candidate trajectory, etc.; these can be set by those skilled in the art according to the actual situation, and no specific restrictions are made here.

[0045] The following is a specific example, such as Figure 2 As shown, the following steps can be included in the embodiments of this application to further illustrate the acquisition of the planning code at the current stage: (1) First, the embodiments of this application can obtain horizontal candidate trajectories based on rule-based strategies, and perform geometric feature encoding on each horizontal candidate trajectory to obtain a horizontal query vector. .

[0046] (2) Further, embodiments of this application can perform horizontal queries With learnable vertical queries The initial planning query vector is obtained by concatenating and linearly transforming the vector, followed by a self-attention mechanism. It can be represented as: , , in, This indicates computation using a self-attention mechanism.

[0047] (3) Finally, the embodiments of this application can... Interactive planning is performed using a two-way interactive modeling strategy to obtain the planning code. .

[0048] Optionally, in one embodiment of this application, before obtaining the future trajectory code of the previous stage, the method further includes: performing interactive trajectory prediction and interactive trajectory planning in parallel in the first stage of the planning time domain to obtain initial planning code and initial prediction code; and concatenating the initial planning code and initial prediction code to obtain the future trajectory code of the first stage.

[0049] As one possible implementation, embodiments of this application can concatenate the obtained planning code and prediction code to obtain... Future trajectory coding of the stage It can be represented as: , Furthermore, embodiments of this application can iteratively and repeatedly execute interactive trajectory prediction and interactive trajectory planning in parallel until the final stage. To obtain the final predicted code and planning codes It can simultaneously reflect the interaction information between the future state of surrounding traffic participants and the planned trajectory of the vehicle, thereby providing a structured and high-fidelity feature representation for subsequent trajectory generation, decision optimization and safety assessment, and improving the response capability and planning accuracy of assisted driving or autonomous driving vehicles in complex dynamic traffic scenarios.

[0050] In step S103, the final planning code and the final prediction code are decoded to obtain the predicted trajectory, planning distance and corresponding confidence level in the planning time domain, and the planning distance is mapped to the lateral trajectory to obtain the planning trajectory in the planning time domain based on the predicted trajectory, confidence level and lateral trajectory.

[0051] In the embodiments of this application, the predicted trajectory can refer to the possible future movement paths of surrounding traffic participants generated by interactive trajectory prediction within the planning time domain, which can be used to reflect the potential behavioral trends of each target vehicle; the planning distance can refer to the minimum lateral and longitudinal distances between each candidate trajectory of the vehicle and surrounding vehicles, road boundaries or obstacles within the planning time domain, which can be used to measure the feasibility and safety of the trajectory; the confidence level can refer to the quantitative index of the credibility of the predicted trajectory or the planned trajectory, which can be used to evaluate the credibility of the trajectory in complex dynamic traffic scenarios, thereby guiding the safe driving decisions of assisted driving or autonomous driving vehicles.

[0052] The planned trajectory refers to the final feasible trajectory generated by combining the predicted trajectory, the planned distance, and the corresponding confidence level. It can simultaneously take into account safety, smoothness, and traffic efficiency, and is used to guide the longitudinal and lateral control of assisted driving or autonomous driving vehicles to achieve effective adaptation to complex dynamic traffic scenarios.

[0053] Optionally, in one embodiment of this application, the formula for predicting the trajectory is: , in, Indicates decoder; Indicates the predicted trajectory; Indicates the final predicted code; The formula for planning distance is: , in, Indicates the planned distance; Indicates the final planning code; The formula for confidence level is: , in, Indicates the confidence level.

[0054] The embodiments of this application can generate a planned trajectory by decoding the predicted trajectory, the planned distance and the corresponding confidence level, so as to fully integrate the future movement trends of surrounding traffic participants, the spatial relationship between the vehicle and the environment and the trajectory feasibility information, and to guide the motion control of assisted driving or autonomous driving vehicles in complex dynamic traffic scenarios.

[0055] like Figure 3 As shown, a specific example is used to illustrate the effectiveness of the parallel iterative joint predictive planning method that considers bidirectional interaction characteristics in the embodiments of this application.

[0056] like Figure 3 As shown, this embodiment of the application conducted closed-loop simulation tests on the nuPlan simulation platform, selecting four scenarios for verification: encountering a vehicle crossing while driving straight, encountering a reversing vehicle after turning right, making a left turn at a busy intersection, and changing lanes to overtake a stationary vehicle. It can be seen that the parallel iterative joint predictive planning method considering bidirectional interaction characteristics provided in this embodiment of the application can make decision-making and planning decisions consistent with human driving based on reasoning about bidirectional interaction relationships. In complex interaction scenarios, it balances efficiency and safety, ensuring driving within the drivable area without collisions with surrounding vehicles, and its driving strategy is flexible and efficient without exhibiting overly conservative behavior that negatively impacts traffic efficiency.

[0057] The parallel iterative joint prediction and planning method based on bidirectional interaction characteristics proposed in this application executes interactive trajectory prediction and interactive trajectory planning based on a bidirectional interaction modeling mechanism in parallel within the planning time domain to obtain the final planning code and the final prediction code. This is then decoded to obtain the predicted trajectory, planning distance, corresponding confidence level, and planned trajectory. This method can simultaneously integrate environmental information and agent interaction relationships, enabling it to understand traffic lights, road topology, surrounding vehicle states, and potential behavioral intentions in complex multi-agent interactions in urban environments. This ensures that the planned trajectory does not collide with environmental vehicles, cross road boundaries, or generate unnecessary waiting times, thereby maintaining traffic safety and efficiency, rather than relying solely on unidirectional prediction or fixed rules to generate trajectories. Simultaneously, it guarantees computational efficiency, enabling assisted driving or autonomous driving vehicles to generate safe, smooth, and efficient driving trajectories in complex dynamic traffic scenarios, thereby improving overall vehicle safety and traffic efficiency.

[0058] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0059] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0060] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0061] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0062] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0063] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0064] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0065] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A parallel iterative joint predictive programming method considering bidirectional interaction characteristics, characterized in that, Includes the following steps: The planning time domain is divided into multiple stages; Within the multiple stages, interactive trajectory prediction based on a two-way interactive modeling mechanism and interactive trajectory planning based on the two-way interactive modeling mechanism are executed in parallel according to the future trajectory encoding of the previous stage until the preset conditions are met, so as to obtain the final planning code and the final prediction code respectively. The final planning code and the final prediction code are decoded to obtain the predicted trajectory, planning distance and corresponding confidence level in the planning time domain, and the planning distance is mapped to the lateral trajectory to obtain the planning trajectory in the planning time domain based on the predicted trajectory, the confidence level and the lateral trajectory.

2. The method according to claim 1, characterized in that, Before obtaining the future trajectory encoding from the previous stage, the process also includes: The interactive trajectory prediction and interactive trajectory planning are executed in parallel in the first stage of the planning time domain to obtain the initial planning code and the initial prediction code; The initial planning code and the initial prediction code are concatenated to obtain the future trajectory code for the first stage.

3. The method according to claim 1, characterized in that, The interactive trajectory prediction based on the bidirectional interactive modeling mechanism includes the following steps: Based on the bidirectional interactive modeling mechanism, the first cross-attention calculation is performed on the scene encoding of the prediction target using the prediction query vector to obtain the current environmental semantic conditions of the prediction target. The second cross-attention calculation is performed on the future trajectory encoding of the previous stage using the predicted query vector to obtain the agent's future intention conditions; The current environmental semantic conditions and the future intention conditions are concatenated and truncated using time pooling to obtain the predictive code for the current stage.

4. The method according to claim 3, characterized in that, The formula for calculating the first cross-attention is: , in, This represents the predicted query vector after the update and integration of environmental information. This indicates cross-attention calculation. This represents the predicted query vector. This represents the scene encoding; The formula for calculating the second cross-attention is: , in, The predicted query vector represents the updated future intention of the fused agent. This represents the encoding of the future trajectory.

5. The method according to claim 4, characterized in that, The formula for splicing is: , in, Indicates intermediate variables. Indicates the first linear layer; This indicates the first concatenation operation; The formula for the time pooling truncation is: , in, Indicates predictive coding. This indicates a max pooling operation. Represents the planning time domain of stage j. This indicates that the data is being extracted along the time dimension.

6. The method according to claim 1, characterized in that, The interactive trajectory planning based on the bidirectional interactive modeling mechanism includes the following steps: The horizontal candidate trajectory of the planning target is determined according to the preset rule-based strategy, and the horizontal candidate trajectory is geometrically encoded to obtain the horizontal query vector of the planning target. The horizontal query vector and the vertical query vector of the planning target are concatenated and linearly transformed, and a self-attention mechanism is used to calculate the initial planning query vector of the planning target. The interactive trajectory planning is performed on the initial planning query vector according to the bidirectional interactive modeling mechanism to obtain the planning code for the current stage.

7. The method according to claim 6, characterized in that, The formula for the linear change is: , in, Indicates intermediate variables. Indicates the second linear layer; This indicates the second splicing operation. Indicates a horizontal query. Indicates a vertical query; The formula for calculating the self-attention mechanism is as follows: , in, This represents the initial planning query vector. This indicates that the self-attention mechanism is calculated.

8. The method according to claim 1, characterized in that, The formula for predicting the trajectory is: , in, Indicates decoder; This represents the predicted trajectory; This represents the final predicted code; The formula for the planned distance is: , in, Indicates the planned distance; This represents the final planning code; The formula for the confidence level is: , in, This indicates the confidence level.