Automatic driving trajectory planning device, method and electronic equipment

Abstract

This application provides an autonomous driving trajectory planning device, method, and electronic device, including at least one path planning module and at least one speed planning module. The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule generates multiple candidate paths, and the path optimization submodule optimizes each candidate path. The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule generates corresponding speed curves based on the multiple candidate paths, and the speed optimization submodule optimizes the speed curves. At least one submodule in both the path planning module and the speed planning module is a neural network model. In the autonomous driving trajectory planning problem, using a neural network model to replace the existing QP solver enables parallel optimization of multiple candidate paths, accelerating computation and improving the efficiency of solving multiple paths.

Description

Technical Field

[0001] This application relates to the field of autonomous driving, and more particularly to an autonomous driving trajectory planning device, method, and electronic device. Background Technology

[0002] Autonomous driving technology is a key technology for improving vehicle active safety, and trajectory planning is one of the core technologies of autonomous driving.

[0003] Trajectory planning in dynamic scenarios is a typical spatiotemporally coupled problem, as the spatial positions of obstacles change dynamically over time, requiring consideration of both spatial and temporal dimensions. To reduce the difficulty, some studies have proposed decoupling strategies, among which the path-velocity decoupling (PV) method transforms a three-dimensional spatiotemporal planning problem into a two-dimensional path planning and a one-dimensional velocity planning problem.

[0004] However, existing technologies mainly use QP solvers for path optimization and speed optimization, which has certain drawbacks. It is difficult to perform parallel optimization on multiple trajectories, and it consumes a lot of CPU resources, causing severe CPU load and affecting the lifespan of the CPU and equipment. Summary of the Invention

[0005] This application provides an autonomous driving trajectory planning device to solve the problem of parallel optimization of multiple trajectories, including at least one path planning module and at least one speed planning module, wherein...

[0006] The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule is used to generate multiple candidate paths, and the path optimization submodule is used to optimize each candidate path.

[0007] The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule is used to generate corresponding speed curves based on multiple candidate paths, and the speed optimization submodule is used to optimize the speed curves.

[0008] At least one sub-module of at least one path planning module and at least one speed planning module is a neural network model.

[0009] In some embodiments, neural network models prioritize the use of GPU resources.

[0010] In some embodiments, the device further includes:

[0011] The acquisition module is used to acquire information related to vehicles and their surrounding environment in real time and generate datasets.

[0012] The processing module is used to extract key point information from the dataset and input the key point information into the path planning module or the speed planning module.

[0013] Key information includes road scene information, the starting point of path planning, and the ending point of path planning.

[0014] In some embodiments, the path planning module is used to generate a global path based on the initial point and the endpoint of the path planning.

[0015] The path decision submodule is used to generate multiple candidate paths based on road scene information.

[0016] In some embodiments, the optimized candidate paths and multiple velocity curves form multiple trajectories. The device also includes an evaluation module for evaluating the multiple trajectories and outputting the optimal trajectory.

[0017] This application also proposes an autonomous driving trajectory planning method, which includes the following steps:

[0018] Real-time acquisition of vehicle and surrounding environment information and generation of datasets;

[0019] Feature extraction is performed on the dataset to obtain key point information, and this key point information is input into the path planning module or the speed planning module; at least one sub-module of at least one path planning module and at least one speed planning module is a neural network model.

[0020] Multiple trajectories are output after the neural network model is used;

[0021] The system evaluates multiple trajectories and outputs the optimal trajectory, which the autonomous vehicle then uses to drive.

[0022] In some embodiments, key point information includes road scene information, the initial point of path planning, and the endpoint of path planning. Inputting this key point information into the speed planning module specifically includes:

[0023] Multiple speed curves are generated based on dynamic information from road scene data;

[0024] Optimize each speed curve and output it.

[0025] In some embodiments, key point information includes road scene information, the initial point of path planning, and the endpoint of path planning. Inputting key point information into the path planning module specifically includes:

[0026] A global path is generated based on the initial point and the endpoint of the path planning.

[0027] Multiple candidate paths are generated based on road scene information;

[0028] Optimize each path and output the results.

[0029] In some embodiments, real-time acquisition of vehicle and surrounding environment-related information and generation of a dataset are specifically included:

[0030] Use the QP solver to run the simulator test set to generate the dataset; or

[0031] A dataset is generated by acquiring information about vehicles and their surrounding environment through cameras.

[0032] One embodiment of this application provides an electronic device, which includes: a processor and a memory communicatively connected to the processor;

[0033] The memory stores instructions that the computer executes;

[0034] The processor executes computer execution instructions stored in the memory to implement the autonomous driving trajectory planning method involved in the above embodiments.

[0035] This application provides an autonomous driving trajectory planning device, method, and electronic device, including at least one path planning module and at least one speed planning module. The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule generates multiple candidate paths, and the path optimization submodule optimizes each candidate path. The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule generates corresponding speed curves based on the multiple candidate paths, and the speed optimization submodule optimizes the speed curves. At least one submodule in both the path planning module and the speed planning module is a neural network model. In the autonomous driving trajectory planning problem, using a neural network model to replace the existing QP solver enables parallel optimization of multiple candidate paths, accelerating computation and improving the efficiency of solving multiple paths. Attached Figure Description

[0036] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0037] Figure 1 A schematic diagram illustrating lane changing in dynamic scenarios for autonomous vehicles;

[0038] Figure 2 This is a schematic diagram of an autonomous driving trajectory planning device in one embodiment;

[0039] Figure 3 This is a flowchart of an autonomous driving trajectory planning method using a neural network model in one embodiment;

[0040] Figure 4 This is a flowchart of an autonomous driving trajectory planning method using a neural network model in another embodiment;

[0041] Figure 5This is a flowchart of an autonomous driving trajectory planning method using a neural network model in another embodiment;

[0042] Figure 6 This is a schematic diagram of an autonomous driving trajectory planning device in one embodiment;

[0043] Figure 7 This is a schematic diagram of an electronic device in one embodiment.

[0044] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0045] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0046] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0047] Trajectory planning, as the core of autonomous driving technology, requires comprehensive consideration of information perceived by sensors, including driving environment information and vehicle status, to generate a safe, smooth, and feasible trajectory, which is then input into the control module. However, achieving fully autonomous driving in dynamic environments remains a challenge.

[0048] Among them, lane changing is the most typical dynamic scenario, such as Figure 1 As shown, when an autonomous vehicle wants to leave the highway or turn left at the next intersection, it must change lanes. In this situation, the autonomous vehicle's trajectory is more susceptible to the influence of surrounding vehicles and road conditions. For lane-changing scenarios, some research has proposed decoupling strategies to simplify trajectory planning in dynamic scenarios, reducing a three-dimensional constraint optimization problem to a two-dimensional problem involving space and time: path and speed.

[0049] In existing technologies, path-velocity decoupling mainly uses the QP solver to solve for the optimal trajectory. However, the operation time is long and it cannot perform parallel optimization for multiple trajectories. At the same time, the QP solver mainly consumes CPU computing resources, causing a serious CPU load and affecting the lifespan of the CPU and the entire device.

[0050] Based on the above considerations, one embodiment of this application provides an autonomous driving trajectory planning device, such as... Figure 2 As shown,

[0051] To address the problem of parallel optimization of multiple trajectories, the device includes at least one path planning module and at least one velocity planning module, wherein...

[0052] The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule is used to generate multiple candidate paths, and the path optimization submodule is used to optimize each candidate path. The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule is used to generate corresponding speed curves based on multiple candidate paths, and the speed optimization submodule is used to optimize the speed curves. At least one of the path planning module and at least one of the submodules of the speed planning module is a neural network model.

[0053] The path decision module generates multiple candidate paths, and each candidate path forms a "path optimization convex space" within the module. The path optimization module operates within each "path optimization convex space" to find the optimal path within that space.

[0054] Similarly, the speed planning module generates multiple speed curves, and each speed curve within the module forms a "speed optimization convex space". The speed optimization module operates within each "speed optimization convex space" to find the optimal speed within that space.

[0055] In the problem of autonomous driving trajectory planning, a neural network model is used to replace the QP solver in the existing technology. The solution speed of the neural network model can reach 10 times that of the QP solver. On this basis, it can also solve multiple reference paths in parallel, thereby improving the solution efficiency.

[0056] In some embodiments, such as Figure 6 As shown, the device also includes:

[0057] The acquisition module is used to acquire information related to vehicles and their surrounding environment in real time and generate datasets.

[0058] The relevant information about the vehicle and its surrounding environment specifically includes environmental perception information and vehicle location and navigation information. Environmental perception information mainly includes information about obstacles, road conditions, and lane markings in the vehicle's surroundings, provided in real time by sensors such as LiDAR, radar, and cameras. Vehicle location and navigation information includes the position and speed of the vehicle and other vehicles, which can be obtained using GPS combined with inertial navigation.

[0059] The acquired information about vehicles and their surrounding environment is organized into a dataset for storage and processing.

[0060] The processing module is used to extract key point information from the dataset and input the key point information into the path planning module or the speed planning module.

[0061] Key information includes road scene information, the starting point of path planning, and the ending point of path planning.

[0062] In some embodiments, the path planning module is used to generate a global path based on the initial point and the endpoint of the path planning.

[0063] The path decision submodule is used to generate multiple candidate paths based on road scene information.

[0064] First, the autonomous driving road segment is divided into cells, and then points within these cells are randomly sampled. By taking one point from each cell and connecting multiple points, a candidate path can be constructed. Multiple candidate paths can be constructed by repeating this process.

[0065] In some embodiments, the optimized candidate paths and multiple velocity curves form multiple trajectories. The device also includes an evaluation module for evaluating the multiple trajectories and outputting the optimal trajectory.

[0066] The optimal trajectory is generated by evaluating multiple trajectories using a cost function and selecting the one with the lowest cost. This cost function may take into account factors such as changes in speed and curvature, distance from obstacles, deviation from the lane centerline, or any other factors that may be desired.

[0067] In some embodiments, the neural network model is a transformer-based neural network model.

[0068] In some embodiments, the neural network model prioritizes GPU resources. This saves CPU resources, as the QP solver primarily consumes CPU computing resources, causing a heavy CPU load and affecting the lifespan of the CPU and the entire device.

[0069] This neural network model prioritizes GPU resources and can solve for optimal trajectory information at a speed up to 10 times faster than the QP solver. Furthermore, it can solve multiple reference paths in parallel, further improving efficiency.

[0070] This application conducted corresponding simulation and real-vehicle tests on the optimization data of the neural network model and the optimization data of the QP solver, as shown in the table below. The data used for simulation and real-vehicle testing consisted of 200,000 data points. When optimizing the neural network model based on the transformer, the simulation time was nearly 6 times that of the QP solver, and the real-vehicle testing time was nearly 9 times that of the QP solver.

[0071] Table 1 Comparison of simulation and real-vehicle testing times for neural network models and QP solvers

[0072] Simulation time 4ms 0.7ms Actual vehicle time 12ms 1.5ms

[0073] Based on the aforementioned technical problems, this application also proposes an autonomous driving trajectory planning method, such as... Figure 3 As shown, the method includes the following steps:

[0074] S100: Real-time acquisition of vehicle and surrounding environment information and generation of dataset;

[0075] Information about the vehicle and its surrounding environment is obtained in real time by the vehicle's environmental perception module.

[0076] The relevant information about the vehicle and its surrounding environment specifically includes environmental perception information and vehicle location and navigation information. Environmental perception information mainly includes information about obstacles, road conditions, and lane markings in the vehicle's surroundings, provided in real time by sensors such as LiDAR, radar, and cameras. Vehicle location and navigation information includes the position and speed of the vehicle and other vehicles, which can be obtained using GPS combined with inertial navigation.

[0077] The acquired information about the vehicle and its surrounding environment is organized into a dataset for storage and as input for the next step.

[0078] S200. Perform feature extraction on the dataset to obtain key point information, and input the key point information into the path planning module or the speed planning module; at least one sub-module of at least one path planning module and at least one speed planning module is a neural network model.

[0079] The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule is used to generate multiple candidate paths, and the path optimization submodule is used to optimize each candidate path. The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule is used to generate corresponding speed curves based on multiple candidate paths, and the speed optimization submodule is used to optimize the speed curves.

[0080] Path planning is mainly for static environments, such as roads, stationary or low-speed obstacles; speed planning is mainly for dynamic environments, such as medium-speed or high-speed obstacles.

[0081] Path planning includes path decision-making and path optimization. Path decision-making mainly involves determining whether to use a detour based on road scene information, such as lane conditions, obstacles blocking the road, and whether the detour is a solid line. Path optimization selects the optimal path by judging rules such as path length and the possibility of collision with obstacles.

[0082] For example, if it is a detour route, a decision to detour is made; if there is an obstacle blocking the way, a decision to stop is made.

[0083] Speed ​​planning includes speed decision-making and speed optimization. For dynamic obstacles, the system first makes speed decisions, such as acceleration or deceleration, lateral acceleration, jerk, etc. In order to ensure safe, courteous driving and a comfortable driving experience, the speed curve needs to be optimized to make it smooth.

[0084] S300, after passing through a neural network model, outputs multiple trajectories;

[0085] Multiple trajectories are generated and output based on the optimized candidate paths and velocity curves. All trajectories are feasible. The optimization of the candidate paths and / or velocity curves can be performed using a neural network model.

[0086] S400 evaluates multiple trajectories and outputs the optimal trajectory, which the autonomous vehicle then uses to drive.

[0087] The optimal trajectory is generated by evaluating multiple trajectories using a cost function and selecting the one with the lowest cost. This cost function may take into account factors such as changes in speed and curvature, distance from obstacles, deviation from the lane centerline, or any other factors that may be desired.

[0088] In some embodiments, key point information includes road scene information, the initial point of path planning, and the endpoint of path planning. Inputting key point information into the speed planning module specifically includes the following steps:

[0089] S201. Generate multiple speed curves based on dynamic information in road scene information;

[0090] The speed planning module mainly deals with dynamic obstacles, such as vehicles traveling at medium / high speeds. It inputs the necessary key information into the speed decision submodule. In order to avoid collisions with obstacles, the speed decision submodule needs to continuously adjust the speed, which will form a speed curve.

[0091] In one embodiment, the process can be performed by a neural network model.

[0092] S202, Optimize each speed curve and output it.

[0093] Feasibly, the speed curve can be optimized using a neural network model. After optimization by the algorithm, each speed curve becomes smooth. Using a neural network model, optimization can be performed quickly to avoid collisions with medium- and high-speed objects at the fastest speed. In this way, autonomous vehicles can also respond to emergencies autonomously and in a timely manner.

[0094] Optimization algorithms use complex mathematical operations to search for low-cost solutions subject to various constraints, such as legal restrictions (e.g., speed limits), distance restrictions (e.g., distance to obstacles), and physical limitations of the vehicle (e.g., acceleration limits).

[0095] In some embodiments, key point information includes road scene information, the initial point of path planning, and the endpoint of path planning. Inputting key point information into the path planning module specifically includes:

[0096] S301. Generate a global path based on the initial point and the endpoint of the path planning;

[0097] In this step, the autonomous vehicle generates a global path from the initial point to the destination based on the initial point and the destination point of the path planning, combined with the map scene.

[0098] The path planning module includes a path decision submodule and a path optimization submodule.

[0099] S302: Generate multiple candidate paths based on road scene information;

[0100] In the path decision submodule, candidate curves are generated, which represent the drivable paths for the vehicle. To generate candidate paths, we first divide the global path into numerous cells, then randomly sample the points in these cells; select one point from each cell, and connect these points to create a candidate path. By repeating this process, multiple candidate paths can be constructed.

[0101] It is important to understand that there can be multiple candidate curves or a single curve, which is not limited here, and the candidate curves here can be part or all of the global path.

[0102] S303, Optimize each candidate path and output it.

[0103] The path decision submodule generates multiple candidate paths, and the path optimization submodule optimizes each candidate path to make each candidate path the best path, and outputs the optimized multiple candidate paths.

[0104] In one embodiment, such as Figure 4 As shown, this application also proposes an autonomous driving trajectory planning method, which includes the following steps:

[0105] Input key point information into the speed planning module and the path planning module;

[0106] S1. Generate and optimize multiple speed curves based on dynamic information in road scene information;

[0107] S2. Generate a global path based on the initial point and the endpoint of the path planning, generate multiple candidate paths based on road scene information, and optimize each candidate path.

[0108] The optimized candidate paths and multiple velocity curves form multiple trajectories, and the evaluation module determines the optimal trajectory.

[0109] Based on the information surrounding the vehicle, steps S1 and S2 can be iterated continuously to obtain the optimal trajectory, and the vehicle trajectory information can also be updated in real time.

[0110] In this embodiment, a neural network model is used to make decisions on speed and obtain multiple speed curves. This can quickly solve obstacle problems in dynamic scenes. For example, in lane changing situations, a decision can be made quickly to avoid collisions with dynamic obstacles.

[0111] Secondly, by iterating through steps S1 and S2, the path and velocity curves can be smoothed out. Finally, the optimal trajectory is determined through evaluation among multiple trajectories.

[0112] In one embodiment, such as Figure 5 As shown, this application also proposes an autonomous driving trajectory planning method, which includes the following steps:

[0113] Input key point information into the speed planning module and the path planning module;

[0114] S1. Generate a global path based on the initial point and the endpoint of the path planning, generate multiple candidate paths based on road scene information, optimize each candidate path and output it;

[0115] S2. Generate multiple speed curves based on dynamic information in the road scene, optimize each speed curve and output it;

[0116] The optimized candidate paths and multiple velocity curves form multiple trajectories, and the evaluation module determines the optimal trajectory.

[0117] Based on the information surrounding the vehicle, the optimal trajectory for autonomous driving is obtained through optimization using a neural network model, enabling real-time updates of vehicle trajectory information.

[0118] In this embodiment, multiple trajectories can be quickly obtained through a neural network model, and the optimal trajectory can be determined through evaluation, which speeds up the calculation and improves the calculation efficiency of multiple trajectories.

[0119] In some embodiments, real-time acquisition of vehicle and surrounding environment-related information and generation of a dataset are specifically included:

[0120] Use the QP solver to run the simulator test set to generate the dataset; or

[0121] Data sets are generated by acquiring information about the surrounding environment of a driving vehicle using cameras and other means.

[0122] In some embodiments, the processing module 402 is specifically used for:

[0123] S100: Real-time acquisition of vehicle and surrounding environment information and generation of dataset;

[0124] S200: Extract key point information from the dataset and input the key point information into the path planning module or speed planning module.

[0125] S300, after passing through a neural network model, outputs multiple trajectories;

[0126] S400 evaluates multiple trajectories and outputs the optimal trajectory, which the autonomous vehicle then uses to drive.

[0127] In some embodiments, the processing module 402 is further configured to:

[0128] S201. Generate multiple speed curves based on dynamic information in road scene information;

[0129] S202, Optimize each speed curve and output it.

[0130] In some embodiments, the processing module 402 is further configured to:

[0131] S301. Generate a global path based on the initial point and the endpoint of the path planning;

[0132] S302: Generate multiple candidate paths based on road scene information;

[0133] S303, Optimize each candidate path and output it.

[0134] In some embodiments, the processing module 402 is further configured to:

[0135] Input key point information into the speed planning module and the path planning module;

[0136] S1. Generate and optimize multiple speed curves based on dynamic information in road scene information;

[0137] S2. Generate a global path based on the initial point and the endpoint of the path planning, generate multiple candidate paths based on road scene information, and optimize each candidate path.

[0138] The optimized candidate paths and multiple velocity curves form multiple trajectories, and the evaluation module determines the optimal trajectory.

[0139] Based on the information surrounding the vehicle, steps S1 and S2 can be iterated continuously to obtain the optimal trajectory, and the vehicle trajectory information can also be updated in real time.

[0140] In some embodiments, the processing module 402 is further configured to:

[0141] Input key point information into the speed planning module and the path planning module;

[0142] S1. Generate a global path based on the initial point and the endpoint of the path planning, generate multiple candidate paths based on road scene information, optimize each candidate path and output it;

[0143] S2. Generate multiple speed curves based on dynamic information in the road scene, optimize each speed curve and output it;

[0144] The optimized candidate paths and multiple velocity curves form multiple trajectories, and the evaluation module determines the optimal trajectory.

[0145] like Figure 7 As shown, one embodiment of this application provides an electronic device 500, which includes a memory 501 and a processor 502.

[0146] Among them, memory 501 is used to store computer instructions that can be executed by the processor;

[0147] The processor 502 implements the various steps of the method in the above embodiments when executing computer instructions. For details, please refer to the relevant descriptions in the foregoing method embodiments.

[0148] Optionally, the memory 501 can be either independent or integrated with the processor 502. When the memory 501 is configured independently, the electronic device also includes a bus for connecting the memory 501 and the processor 502.

[0149] This application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the various steps of the methods described above.

[0150] This application also provides a computer program product, including computer instructions that, when executed by a processor, implement the various steps in the methods described above.

[0151] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0152] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. An automatic driving trajectory planning device characterized by comprising: It includes at least one path planning module and at least one speed planning module, wherein, The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule is used to generate multiple candidate paths, and the path optimization submodule is used to optimize each candidate path. The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule is used to generate multiple corresponding speed curves based on the multiple candidate paths, and the speed optimization submodule is used to optimize each speed curve generated by the speed decision submodule. The path optimization submodule and / or the speed optimization submodule are configured as neural network models to perform parallel optimization of the candidate paths and / or the speed curves. The path decision submodule is used to divide the autonomous driving road segment into cells, randomly sample the points in the cells, and construct a candidate path by taking one point from each cell and connecting multiple points. Multiple candidate paths are constructed by repeating this process. In the path optimization submodule, each candidate path forms a path optimization convex space. The path optimization submodule is used to find the optimal path within each path optimization convex space. In the speed optimization submodule, each speed curve forms a speed optimization convex space. The speed optimization submodule is used to find the optimal speed within each speed optimization convex space.

2. The automatic driving trajectory planning apparatus according to claim 1, wherein The neural network model prioritizes the use of GPU resources.

3. The autonomous driving trajectory planning device as described in claim 1, characterized in that, The device further includes: The acquisition module is used to acquire information related to vehicles and their surrounding environment in real time and generate datasets. The processing module is used to extract key point information from the dataset and input the key point information into the path planning module or the speed planning module, wherein... The key information includes road scene information, the initial point of path planning, and the endpoint of path planning.

4. The autonomous driving trajectory planning device as described in claim 3, characterized in that, The path planning module is used to generate a global path based on the initial point and the endpoint of the path planning. The path decision submodule is used to generate multiple candidate paths based on the road scene information. 5.The automatic driving trajectory planning apparatus of claim 1, wherein the optimized plurality of candidate paths and the plurality of velocity profiles form a plurality of trajectories. The device also includes an evaluation module, which evaluates the multiple trajectories and outputs the optimal trajectory.

6. An automatic driving trajectory planning method characterized by comprising: The method includes the following steps: Real-time acquisition of vehicle and surrounding environment information and generation of datasets; Feature extraction is performed on the dataset to obtain key point information, which is then input into a path planning module or a speed planning module. The path planning module includes a path decision submodule and a path optimization submodule. The path decision submodule generates multiple candidate paths based on the key point information, and the path optimization submodule optimizes each candidate path generated by the path decision submodule. The speed planning module includes a speed decision submodule and a speed optimization submodule. The speed decision submodule generates multiple corresponding speed curves based on the multiple candidate paths output by the path planning module, and the speed optimization submodule optimizes each speed curve generated by the speed decision submodule. The path optimization submodule and / or the speed optimization submodule are configured as neural network models to perform parallel optimization of the candidate paths and / or the speed curves. The path decision submodule is used to divide the autonomous driving road segment into cells, randomly sample the points in the cells, and construct a candidate path by taking one point from each cell and connecting multiple points. Multiple candidate paths are constructed by repeating this process. In the path optimization submodule, each candidate path forms a path optimization convex space, and the path optimization submodule is used to find the optimal path within each path optimization convex space. In the speed optimization submodule, each speed curve forms a speed optimization convex space, and the speed optimization submodule is used to find the optimal speed within each speed optimization convex space. The neural network model outputs multiple trajectories; The multiple trajectories are evaluated and the optimal trajectory is output. The autonomous vehicle then uses the optimal trajectory to drive.

7. The automatic driving trajectory planning method of claim 6, wherein, The key point information includes road scene information, the initial point of path planning, and the endpoint of path planning. Inputting the key point information into the speed planning module specifically includes: Multiple speed curves are generated based on the dynamic information in the road scene information; Optimize each speed curve and output it.

8. The automatic driving trajectory planning method according to claim 6 or 7, characterized in that, The key point information includes road scene information, the initial point of path planning, and the endpoint of path planning. Inputting the key point information into the path planning module specifically includes: A global path is generated based on the initial point and the endpoint of the path planning. Multiple candidate paths are generated based on the road scene information; Optimize each candidate path and output the result.

9. The autonomous driving trajectory planning method as described in claim 6, characterized in that, Real-time acquisition of vehicle and surrounding environment information and generation of datasets, specifically including: Use the QP solver to run the simulator test set to generate the dataset; or A dataset is generated by acquiring information about vehicles and their surrounding environment through cameras.

10. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 6 to 9.

Images