Task coordination decision planning network architecture for platooning vehicles, devices and media
By employing a decision-making model based on deep learning neural networks in platooning vehicles, the collaborative decision-making and planning capabilities of the lead vehicle and follower vehicles have been improved, solving the problems of high computational complexity and poor adaptability to environmental uncertainties, and achieving more efficient collaborative decision-making.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI DEEPWAY TECHNOLOGY CO LTD
- Filing Date
- 2026-03-13
- Publication Date
- 2026-07-10
AI Technical Summary
Existing formation autonomous driving decision-making and planning algorithms suffer from high computational complexity, insufficient real-time performance, poor adaptability to environmental uncertainties, and difficulty in capturing complex interaction features.
By employing deep learning neural networks, decision-making models for the lead vehicle and the follower vehicle are configured. Cross Attention structure and fully connected layers are used for information fusion and planning to improve complex interaction capabilities.
It improves the collaborative decision-making and planning capabilities of platooned vehicles and enhances their adaptability to dynamic traffic environments and uncertain behaviors.
Smart Images
Figure CN122369247A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of vehicle technology, and in particular to a mission collaborative decision-making and planning network architecture, device and medium for platooned vehicles. Background Technology
[0002] Formation-based autonomous driving decision-making and planning technologies mainly fall into two categories: centralized and distributed. Centralized methods involve a lead vehicle uniformly planning the entire platoon's trajectory, achieving formation maintenance and energy minimization through global optimization. Distributed methods, on the other hand, are based on vehicle-to-vehicle (V2V) communication, where each vehicle collaborates with its neighbors based on local information and their states. Due to computational resource and real-time requirements, distributed methods are more commonly used in mass production.
[0003] Currently, decision-making and planning algorithms for platooning autonomous driving primarily employ a "planning + optimization" approach. This involves first generating reference trajectories or formations based on geometric, rule-based, or heuristic methods, and then optimizing the trajectory using techniques such as model predictive control (MPC), distributed optimization, or nonlinear programming to meet objectives such as safe distance, vehicle dynamics constraints, and energy minimization. In recent years, some research has begun to introduce neural network models, using deep reinforcement learning, graph neural networks, or end-to-end policy learning to learn the collaborative decision-making patterns among platooning vehicles. This enables a direct mapping from perception to control, thereby improving generalization and adaptive capabilities in uncertain scenarios.
[0004] While the "planning + optimization" approach offers good interpretability and engineering feasibility, it also suffers from the following major drawbacks: High computational complexity and insufficient real-time performance: Optimization problems typically involve nonlinear constraints and high-dimensional state spaces, making real-time solutions particularly challenging in multi-vehicle platooning scenarios. Poor adaptability to environmental uncertainties: Traditional methods rely heavily on deterministic models and static constraints, making it difficult to handle dynamic traffic environments or the uncertain behaviors of other vehicles. Difficulty in capturing complex interactions: Rule-based planning struggles to reflect the complex interactive characteristics of real-world traffic, such as mixed pedestrian and vehicle traffic and game-theoretic relationships. Summary of the Invention
[0005] Based on this, it is necessary to provide a task collaborative decision-making and planning network architecture, device and medium for platooning vehicles to address the above-mentioned technical problems. This architecture employs deep learning neural networks to enhance complex interaction capabilities, and enables collaborative decision-making and planning between the lead vehicle and the following vehicles. Compared with traditional algorithms, this architecture improves collaborative decision-making and planning capabilities.
[0006] Firstly, a task coordination decision-making and planning method for platooned vehicles is provided, including: The first decision-making model is configured on the lead vehicle, and the second decision-making model is configured on the follower vehicle, wherein: The first decision model includes a first encoder, a first Cross Attention structure, and a first fully connected layer, wherein, The first encoder is used to receive the vehicle status information of the lead vehicle and the environmental information detected by the lead vehicle, and to encode the vehicle status information of the lead vehicle and the environmental information detected by the lead vehicle. The first Cross Attention structure is connected to the first encoder and is used to receive the pilot vehicle control command and fuse the pilot vehicle control command and the output data from the first encoder. The first fully connected layer is connected to the first Cross Attention structure and is used to determine the planned trajectory of the lead vehicle based on the output of the first Cross Attention structure. The lead vehicle then sends the planned trajectory of the lead vehicle to the following vehicle via V2V communication. The lead vehicle is also used to send the vehicle status information of the lead vehicle and the control commands of the following vehicle to the following vehicle via the V2V communication, so that the following vehicle can collaboratively decide on the planned trajectory of the following vehicle through the second decision model.
[0007] In some examples, the second decision model for following the vehicle includes a caching module, a fusion module, a stitching module, a second encoder, a second Cross Attention structure, and a second fully connected layer, wherein: The caching module is used to cache historical data of the lead vehicle's vehicle status information; The fusion module is connected to the cache module and the second encoder respectively, and is used to fuse short-term historical data of vehicle status information from the lead vehicle and environmental information detected by the following vehicle from the cache module to obtain fused information, and send the fused information to the second encoder. The stitching module is connected to the cache module and the second Cross Attention structure respectively, and is used to stitch together the long-term historical data of the vehicle status information of the navigator vehicle, the control command of the follower vehicle and the planned trajectory of the navigator vehicle from the cache module to obtain stitched data, and provide the stitched data to the second Cross Attention structure. The second encoder is used to encode the fused information and the vehicle status information of the following vehicle; The second Cross Attention structure is used to fuse the spliced data and the output of the second encoder; The second fully connected layer is connected to the second Cross Attention structure and is used to determine the planned trajectory of the following vehicle based on the output from the second Cross Attention structure.
[0008] In some examples, the first encoder and the second encoder are the same.
[0009] In some examples, the environmental information includes obstacle and lane line information.
[0010] In some examples, the navigator control commands include lane change information.
[0011] In some examples, the control commands for the following vehicle include a safe following distance and a safe following time interval.
[0012] Secondly, a task coordination decision-making and planning method for platooned vehicles is provided, which is implemented based on the task coordination decision-making and planning network architecture of platooned vehicles described in the first aspect.
[0013] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the steps of the task collaborative decision-making and planning method for platooning vehicles described in the second aspect above.
[0014] Fourthly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the task collaborative decision-making and planning method for platooning vehicles described in the second aspect above.
[0015] Fifthly, a computer program product is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the task collaborative decision-making and planning method for platooning vehicles described in the second aspect above.
[0016] The embodiments of this application employ deep learning neural networks to enhance complex interaction capabilities, and the navigator and follower vehicles make collaborative decision-making and planning, which improves collaborative decision-making and planning capabilities compared to traditional algorithms. Attached Figure Description
[0017] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A structural block diagram of the task collaborative decision-making and planning network architecture for platooned vehicles provided in this application embodiment; Figure 2 A schematic diagram of the task collaborative decision-making and planning network architecture for platooning vehicles provided in an embodiment of this application; Figure 3 A schematic diagram of the encoder in the task collaborative decision-making and planning network architecture of platooning vehicles provided in the embodiments of this application; Figure 4 This is a structural block diagram of a computer device provided in an embodiment of this application. Detailed Implementation
[0018] The present application will now be described in further detail with reference to the embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the application. Furthermore, it should be noted that, for ease of description, only the parts relevant to the application are shown in the accompanying drawings.
[0019] It should be noted that, unless otherwise specified, the embodiments and features of the embodiments in this application can be combined with each other. The present application will now be described in detail with reference to the accompanying drawings and embodiments.
[0020] The following describes in detail, with reference to the accompanying drawings, the mission collaborative decision-making and planning network architecture, device and medium of platooning vehicles according to embodiments of this application.
[0021] For convoy vehicles, it usually refers to multiple vehicles traveling in a formation, with the lead vehicle acting as the navigator and the remaining vehicles acting as followers to follow the lead vehicle.
[0022] Figure 1 This is a schematic diagram of a task collaborative decision-making and planning network architecture for platooned vehicles according to an embodiment of this application. Figure 1 As shown, a task collaborative decision-making and planning network architecture for platooning vehicles according to an embodiment of this application includes: a first decision model 110 and a second decision model 120, wherein the first decision model 110 is configured on the lead vehicle and the second decision model is configured on the following vehicles. Specifically, in combination with... Figure 2 As shown, where: The first decision model 110 of the lead vehicle (i.e., Lead Vehicle Mode) includes a first encoder, a first cross attention structure (i.e., Cross Attention1), and a first fully connected layer MLP1. The first encoder receives the vehicle state information and environmental information detected by the lead vehicle and encodes these information. The first cross attention structure is connected to the first encoder and receives the lead vehicle control commands and fuses them with the output data from the first encoder. The first fully connected layer is connected to the first cross attention structure and determines the planned trajectory of the lead vehicle based on the output of the first cross attention structure. It then sends the planned trajectory of the lead vehicle to the following vehicle via V2V communication. The lead vehicle also sends its vehicle state information and control commands to the following vehicle via V2V communication, so that the following vehicle can collaboratively decide its planned trajectory using the second decision model.
[0023] Combination Figure 2 As shown, the second decision model 120 of the Follow Vehicle (i.e., Follow VehicleMode) includes a cache module, a fusion module, an embedding module, a second encoder, a second Cross Attention structure (i.e., Cross Attention1), and a second fully connected layer MLP2. Specifically: the cache module caches historical data of the lead vehicle's vehicle state information; the fusion module is connected to both the cache module and the second encoder, and is used to fuse short-term historical data of the lead vehicle's vehicle state information from the cache module with environmental information detected by the follow vehicle to obtain fused information, which is then sent to the second encoder; the embedding module is connected to both the cache module and the second Cross Attention structure, and is used to embed long-term historical data of the lead vehicle's vehicle state information from the cache module, the follow vehicle's control commands, and the lead vehicle's planned trajectory to obtain embedded data, which is then used for the second Cross Attention layer. The Attention structure provides the stitched data; the second encoder encodes the fused information and the vehicle state information of the following vehicle; the second CrossAttention structure fuses the stitched data and the output of the second encoder; the second fully connected layer is connected to the second CrossAttention structure and is used to determine the planned trajectory of the following vehicle based on the output of the second CrossAttention structure.
[0024] In one embodiment of this application, the first encoder and the second encoder are identical. Environmental information includes obstacle and lane line information. The lead vehicle control commands include lane change information. The follow vehicle control commands include following safety distance and following safety time distance.
[0025] by Figure 2 Taking the task coordination decision-making and planning network architecture of the platooning vehicles as an example, the lead vehicle and the follower vehicles use the same encoder. The lead vehicle's vehicle status (State_lv), platoon command, and planned trajectory (Traj_lv) are transmitted to the follower vehicles through the V2V module.
[0026] The Cache and Fusion modules in the following vehicle are traditional rule-based information integration modules that transmit the integrated information to the network model of the following vehicle.
[0027] in: Env_lv: Represents the environmental information received by the lead vehicle, including obstacle (Agents) and lane line (map) information. The obstacle and lane line information are outputs of the upstream perception model. Obstacle information includes position (x, y), heading angle, speed, type, length, and width. Lane line information includes the lane centerline (a series of points, each with x, y coordinates), lane lines (a series of points, each with x, y coordinates), lane type, and lane speed limit.
[0028] State_lv: Represents the positioning and chassis information of the navigator, including position (x, y), heading angle, speed, acceleration, and steering wheel angle.
[0029] Command: This indicates the instructions from the driver in the lead vehicle to the lead vehicle, including lane change information (go straight, change lanes to the left, and change lanes to the right, etc.) and target speed information.
[0030] Platoon Command: Instructions given by the driver in the lead vehicle to the following vehicles, including safe following distance and safe following time interval.
[0031] Traj_lv: The planned trajectory of the navigator vehicle output by the model, such as the driving trajectory in 8 seconds. The information of each trajectory point is position (x, y), heading angle, and speed.
[0032] Short_States_lv: Represents the short-term historical information of the lead vehicle, which is obtained through the information caching module Cache of the following vehicles.
[0033] Long_States_lv: Represents the long-term historical information of the lead vehicle, which is obtained through the information caching module Cache of the following vehicles.
[0034] Env_fv: Environmental information received by the following vehicle, including obstacle (Agents) and lane line (map) information. The obstacle and lane line information are outputs of the upstream perception model. Obstacle information includes position (x, y), heading angle, speed, type, length, and width. Lane line information includes the lane centerline (a series of points, each with x, y coordinates), lane lines (a series of points, each with x, y coordinates), lane type, and lane speed limit.
[0035] State_fv: Represents the vehicle's positioning and chassis information, including position (x, y), heading angle, speed, acceleration, and steering wheel angle.
[0036] Traj_fv: Represents the planned trajectory of the following vehicle output by the model, such as the driving trajectory in the future 8 seconds. The information of each trajectory point is position (x, y), heading angle, and speed.
[0037] like Figure 3As shown, for the Encoder, the input information includes obstacle agents, lane map, and information about the lead or follower vehicle. The output is processed through a standard N-layer Transformer encoder structure.
[0038] Cross Attention 1 is the standard Cross Attention architecture. MLP 1 is a fully connected layer.
[0039] Cache: This refers to the module that caches the historical information of the lead vehicle while following the vehicle.
[0040] Fusion: This refers to the merging of navigator information and obstacle information. Essentially, it processes the navigator information into an information structure consistent with that of the obstacles.
[0041] Embeddding: This indicates that the Long_States_lv, Traj_lv, and Platoon Command are encoded separately and then concatenated.
[0042] Cross Attention 2 is the standard Cross Attention structure, MLP 2: Fully Connected Layer.
[0043] When training the task collaborative decision planning network architecture of the platooning vehicles, the loss of the navigator vehicle is set to L_lv, which is the L1 loss value of the navigator vehicle model's output trajectory Traj_lv and the ground truth value in the training data.
[0044] The loss for training the following vehicle is set to L_fv, which is the L1 loss value of the output trajectory Traj_fv of the following vehicle model and the ground truth value in the training data. Then the total loss is set to L_lv + L_fv.
[0045] According to the task collaborative decision-making and planning network architecture of platooning vehicles in this application embodiment, a deep learning neural network is used to improve complex interaction capabilities, and the lead vehicle and the following vehicle make collaborative decisions and plans, which improves collaborative decision-making and planning capabilities compared with traditional algorithms.
[0046] Furthermore, embodiments of this application disclose a task collaborative decision-making and planning method for platooned vehicles, which is implemented based on the task collaborative decision-making and planning network architecture for platooned vehicles described in any of the above embodiments.
[0047] by Figure 2Taking the task coordination decision-making and planning network architecture of the platooning vehicles as an example, the lead vehicle and the follower vehicles use the same encoder. The lead vehicle's vehicle status (State_lv), platoon command, and planned trajectory (Traj_lv) are transmitted to the follower vehicles through the V2V module.
[0048] The Cache and Fusion modules in the following vehicle are traditional rule-based information integration modules that transmit the integrated information to the network model of the following vehicle.
[0049] in: Env_lv: Represents the environmental information received by the lead vehicle, including obstacle (Agents) and lane line (map) information. The obstacle and lane line information are outputs of the upstream perception model. Obstacle information includes position (x, y), heading angle, speed, type, length, and width. Lane line information includes the lane centerline (a series of points, each with x, y coordinates), lane lines (a series of points, each with x, y coordinates), lane type, and lane speed limit.
[0050] State_lv: Represents the positioning and chassis information of the navigator, including position (x, y), heading angle, speed, acceleration, and steering wheel angle.
[0051] Command: This indicates the instructions from the driver in the lead vehicle to the lead vehicle, including lane change information (go straight, change lanes to the left, and change lanes to the right, etc.) and target speed information.
[0052] Platoon Command: Instructions given by the driver in the lead vehicle to the following vehicles, including safe following distance and safe following time interval.
[0053] Traj_lv: The planned trajectory of the navigator vehicle output by the model, such as the driving trajectory in 8 seconds. The information of each trajectory point is position (x, y), heading angle, and speed.
[0054] Short_States_lv: Represents the short-term historical information of the lead vehicle, which is obtained through the information caching module Cache of the following vehicles.
[0055] Long_States_lv: Represents the long-term historical information of the lead vehicle, which is obtained through the information caching module Cache of the following vehicles.
[0056] Env_fv: Environmental information received by the following vehicle, including obstacle (Agents) and lane line (map) information. The obstacle and lane line information are outputs of the upstream perception model. Obstacle information includes position (x, y), heading angle, speed, type, length, and width. Lane line information includes the lane centerline (a series of points, each with x, y coordinates), lane lines (a series of points, each with x, y coordinates), lane type, and lane speed limit.
[0057] State_fv: Represents the vehicle's positioning and chassis information, including position (x, y), heading angle, speed, acceleration, and steering wheel angle.
[0058] Traj_fv: Represents the planned trajectory of the following vehicle output by the model, such as the driving trajectory in the future 8 seconds. The information of each trajectory point is position (x, y), heading angle, and speed.
[0059] like Figure 3 As shown, for the Encoder, the input information includes obstacle agents, lane map, and information about the lead or follower vehicle. The output is processed through a standard N-layer Transformer encoder structure.
[0060] Cross Attention 1 is the standard Cross Attention architecture. MLP 1 is a fully connected layer.
[0061] Cache: This refers to the module that caches the historical information of the lead vehicle while following the vehicle.
[0062] Fusion: This refers to the merging of navigator information and obstacle information. Essentially, it processes the navigator information into an information structure consistent with that of the obstacles.
[0063] Embeddding: This indicates that the Long_States_lv, Traj_lv, and Platoon Command are encoded separately and then concatenated.
[0064] Cross Attention 2 is the standard Cross Attention structure, MLP 2: Fully Connected Layer.
[0065] The task collaborative decision-making and planning method for platooning vehicles according to the embodiments of this application adopts a deep learning neural network to improve complex interaction capabilities, and the lead vehicle and the following vehicles make collaborative decision-making and planning, which improves collaborative decision-making and planning capabilities compared with traditional algorithms.
[0066] Specific limitations regarding the task coordination decision-making and planning method for platooned vehicles can be found in the above description of the network architecture for task coordination decision-making and planning of platooned vehicles, and will not be repeated here. Each module of the aforementioned network architecture for task coordination decision-making and planning of platooned vehicles can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.
[0067] In one embodiment, a computer device is provided. Figure 4 This is a structural block diagram of the computer device provided in the embodiments of this application, with reference to... Figure 4 The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned task collaborative decision-making and planning method for platooned vehicles.
[0068] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned task collaborative decision-making and planning method for platooned vehicles.
[0069] This application provides a computer program product including instructions that, when executed, cause the method described in this application embodiment to be performed. For example, a task collaborative decision-making and planning method for platooning vehicles can be executed.
[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0071] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0072] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A task collaborative decision-making and planning network architecture for platooned vehicles, characterized in that, This includes a first decision model configured on the lead vehicle and a second decision model configured on the follower vehicle, wherein: The first decision model includes a first encoder, a first Cross Attention structure, and a first fully connected layer, wherein, The first encoder is used to receive the vehicle status information of the lead vehicle and the environmental information detected by the lead vehicle, and to encode the vehicle status information of the lead vehicle and the environmental information detected by the lead vehicle. The first Cross Attention structure is connected to the first encoder and is used to receive the pilot vehicle control command and fuse the pilot vehicle control command and the output data from the first encoder. The first fully connected layer is connected to the first Cross Attention structure and is used to determine the planned trajectory of the lead vehicle based on the output of the first Cross Attention structure. The lead vehicle then sends the planned trajectory of the lead vehicle to the following vehicle via V2V communication. The lead vehicle is also used to send the vehicle status information of the lead vehicle and the control commands of the following vehicle to the following vehicle via the V2V communication, so that the following vehicle can collaboratively decide on the planned trajectory of the following vehicle through the second decision model.
2. The task collaborative decision-making and planning network architecture for platooned vehicles according to claim 1, characterized in that, The second decision model for following the vehicle includes a caching module, a fusion module, a stitching module, a second encoder, a second CrossAttention structure, and a second fully connected layer, wherein: The caching module is used to cache historical data of the lead vehicle's vehicle status information; The fusion module is connected to the cache module and the second encoder respectively, and is used to fuse short-term historical data of vehicle status information from the lead vehicle and environmental information detected by the following vehicle from the cache module to obtain fused information, and send the fused information to the second encoder. The stitching module is connected to the cache module and the second Cross Attention structure respectively, and is used to stitch together the long-term historical data of the vehicle status information of the navigator vehicle, the control command of the follower vehicle and the planned trajectory of the navigator vehicle from the cache module to obtain stitched data, and provide the stitched data to the second Cross Attention structure. The second encoder is used to encode the fused information and the vehicle status information of the following vehicle; The second Cross Attention structure is used to fuse the spliced data and the output of the second encoder; The second fully connected layer is connected to the second Cross Attention structure and is used to determine the planned trajectory of the following vehicle based on the output from the second Cross Attention structure.
3. The task collaborative decision-making and planning network architecture for platooned vehicles according to claim 2, characterized in that, The first encoder and the second encoder are the same.
4. The task collaborative decision-making and planning network architecture for platooned vehicles according to claim 1, characterized in that, The environmental information includes obstacle and lane line information.
5. The task collaborative decision-making and planning network architecture for platooned vehicles according to claim 1, characterized in that, The navigator control commands include lane change information.
6. The task collaborative decision-making and planning network architecture for platooned vehicles according to claim 1, characterized in that, The control commands for the following vehicle include the following safe distance and the following safe time interval.
7. A task collaborative decision-making and planning method for platooned vehicles, characterized in that, The task collaborative decision-making and planning method for platooned vehicles is implemented based on the task collaborative decision-making and planning network architecture for platooned vehicles as described in any one of claims 1-6.
8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the task collaborative decision-making and planning method for platooning vehicles as described in claim 7.
9. A computationally readable storage medium, comprising a memory and a computer program stored on the memory and executable on a processor, characterized in that, When the program is executed by the processor, it implements the task collaborative decision-making and planning method for platooned vehicles as described in claim 7.
10. A computer program product, comprising a memory and a computer program stored in the memory and executable on a processor, characterized in that, When the program is executed by the processor, it implements the task collaborative decision-making and planning method for platooned vehicles as described in claim 7.