Parameter debugging method, device and equipment for trajectory planning module
By using regression testing and simulation scenarios to generate trajectories in the autonomous driving planning module, scores are obtained and parameters are automatically debugged using the Optuna module, the problem of difficult and time-consuming parameter debugging in the existing technology is solved, and an efficient and automatic parameter debugging process is achieved.
Patent Information
- Application Number
- CN202411811959.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2025-04-18
AI Technical Summary
The parameters debugging of existing autonomous driving planning modules is difficult and complex, requiring highly professional and experienced debugging personnel, and it takes a long time.
By generating trajectories in preset regression test scenarios and simulation scenarios, obtaining regression test scores and trajectory difference scores, determining parameter scores based on these scores, and automatically debugging parameters using the Optuna module until the preset conditions are met.
Automatic debugging of parameters is realized, reducing debugging difficulty and requirements for debugging personnel, improving the debugging efficiency of trajectory planning module parameters, and effectively avoiding the fallback of trajectory planning quality.
Smart Images
Figure CN119847062A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of autonomous driving technology, and in particular to a trajectory planning module parameter debugging method, device and equipment. Background Art
[0002] Autonomous driving technology is one of the important directions for future transportation development. It brings many potential benefits, including improving traffic safety, improving travel efficiency, increasing travel convenience, expanding public transportation coverage, and protecting the environment. These advantages make autonomous driving technology a focus of attention and development in various countries. The core goal of autonomous driving technology is to replace human drivers with artificial intelligence, saving labor costs and regulating driving behavior while ensuring the safety and efficiency of unmanned vehicles.
[0003] Autonomous driving algorithms generally include perception modules, planning modules, and control modules. Among them, the planning module, as the core decision-making module, is responsible for planning a safe and efficient driving trajectory that conforms to the vehicle dynamics model based on the environmental information provided by the upstream perception module, and handing it over to the downstream control module for execution. There are a large number of custom parameters or mapping tables in the planning algorithm of the existing planning module, which makes parameter debugging difficult and complex. This not only requires extremely high professionalism and experience of the debugger, but also takes a lot of time to debug the parameters.
[0004] Therefore the prior art still needs to be improved and enhanced. Summary of the invention
[0005] The technical problem to be solved by the present application is to provide a trajectory planning module parameter debugging method, device and equipment in view of the deficiencies in the prior art.
[0006] In order to solve the above technical problems, the first aspect of the present application provides a trajectory planning module parameter debugging method, wherein the trajectory planning module parameter debugging method comprises:
[0007] Generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory;
[0008] Generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario;
[0009] Determining a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score;
[0010] The parameters to be tested are debugged based on the parameter scores to obtain trajectory planning module parameters.
[0011] The trajectory planning module parameter debugging method, wherein the debugging of the parameters to be tested based on the parameter score to obtain the trajectory planning module parameters specifically includes:
[0012] generating candidate test parameters according to a preset parameter space based on the parameter scores;
[0013] The candidate test parameters are used as parameters to be tested, and a regression trajectory is generated based on the parameters to be tested in a preset regression test scenario until a preset end condition is met, and the parameters to be tested when the preset end condition is met are used as trajectory planning module parameters.
[0014] The trajectory planning module parameter debugging method, wherein the preset parameter space is configured with the parameter range of each test parameter item in the parameters to be tested and the data distribution of each test parameter item; the candidate test parameters are generated according to the preset parameter space based on the parameter score, specifically:
[0015] The Optuna module generates candidate test parameters based on the parameter scores according to a preset parameter space.
[0016] The trajectory planning module parameter debugging method, wherein, when there are multiple candidate test parameters, taking the candidate test parameters as the parameters to be tested, and executing the generation of a regression trajectory based on the parameters to be tested in a preset regression test scenario specifically includes:
[0017] Treat each candidate test parameter as a parameter to be tested;
[0018] Each parameter to be tested is assigned to a distributed node in the distributed task cluster, and the distributed node generates a regression trajectory based on the parameter to be tested in a preset regression test scenario to obtain a parameter score corresponding to each parameter to be tested.
[0019] The trajectory planning module parameter debugging method, wherein the step of obtaining the regression test score of the regression trajectory specifically includes:
[0020] Evaluating the evaluation score of the regression trajectory on each preset evaluation indicator in a preset evaluation indicator set;
[0021] A regression test score for the regression trajectory is calculated based on all evaluation scores.
[0022] The trajectory planning module parameter debugging method, wherein the determining of the trajectory difference score based on each planned trajectory and the human driving trajectory corresponding to each preset simulation scenario specifically includes:
[0023] For each planned trajectory, calculating a scene trajectory difference score between the planned trajectory and a human driving trajectory corresponding to a preset simulation scene in which the planned trajectory is located;
[0024] The average of all scene trajectory difference scores is calculated to obtain the trajectory difference score.
[0025] The trajectory planning module parameter debugging method, wherein the calculating of the scene trajectory difference score between the planned trajectory and the human driving trajectory corresponding to the preset simulation scene in which the planned trajectory is located specifically includes:
[0026] For each trajectory moment in the planned trajectory, obtaining a simulated host vehicle state at the trajectory moment in the planned trajectory, and obtaining a real host vehicle state at the trajectory moment in the human driving trajectory;
[0027] The main vehicle state difference between the simulated main vehicle state and the real main vehicle state at each trajectory moment is calculated, and the scene trajectory difference score between the planned trajectory and the human driving trajectory is determined based on the main vehicle state difference at each trajectory moment.
[0028] A second aspect of the present application provides a trajectory planning module parameter debugging device, wherein the trajectory planning module parameter debugging device comprises:
[0029] A trajectory planning regression test module is used to generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory;
[0030] A scenario simulation module, configured to generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario;
[0031] A parameter score determination module, used to determine the parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score;
[0032] A parameter debugging module is used to debug the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters.
[0033] A third aspect of the present application provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in any of the above-described trajectory planning module parameter debugging methods.
[0034] A fourth aspect of the present application provides a server, which includes a plurality of distributed nodes and a central node;
[0035] The distributed node is used to generate a regression trajectory in a preset regression test scenario based on the parameter to be tested, and obtain a regression test score of the regression trajectory; generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameter to be tested, and determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario; determine a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score;
[0036] The central node is used to debug the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters.
[0037] Beneficial effects:
[0038] 1. This application realizes automatic debugging of parameters, reduces the difficulty of program debugging and the requirements for debuggers, and can improve the debugging efficiency of trajectory planning module parameters;
[0039] 2. This application introduces trajectory planning regression testing. The regression test score prevents the regression of the overall trajectory planning performance during the trajectory planning module parameter debugging process, thereby effectively avoiding the situation where "the simulation and comparison scenarios are improved but the overall trajectory planning quality is reduced" during the parameter adjustment iteration process;
[0040] 3. This application determines the trajectory difference score by comparing the planned trajectory with the human driving trajectory, and then evaluates the performance of the trajectory planning module parameters through the trajectory difference score and regression test score, providing real-time feedback for the debugging process, so as to better debug the trajectory planning module parameters. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0042] Figure 1 A flowchart of a trajectory planning module parameter debugging method provided in an embodiment of the present application.
[0043] Figure 2 A flowchart of an example of a trajectory planning module parameter debugging method provided in an embodiment of the present application.
[0044] Figure 3 Schematic diagram of the scene for the side vehicle insertion.
[0045] Figure 4 A schematic diagram of the big bend scene.
[0046] Figure 5 A schematic diagram of a lane merging scenario.
[0047] Figure 6 Schematic diagram of an unprotected left turn scenario.
[0048] Figure 7 Schematic diagram of the trajectory planning module.
[0049] Figure 8 Schematic diagram of the trajectory planning module parameter debugging device.
[0050] Fig. 9 A functional block diagram of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0051] The embodiments of the present application provide a method, device and equipment for debugging parameters of a trajectory planning module. In order to make the purpose, technical solution and effect of the present application clearer and more specific, the present application is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0052] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.
[0053] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.
[0054] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not mean the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.
[0055] After research, it is found that autonomous driving technology is one of the important directions for future transportation development. It brings many potential benefits, including improving traffic safety, improving travel efficiency, increasing travel convenience, expanding public transportation coverage, and protecting the environment. These advantages make autonomous driving technology a focus of attention and development in various countries. The core goal of autonomous driving technology is to replace human drivers with artificial intelligence, while saving labor costs and regulating driving behavior, ensuring the safety and efficiency of unmanned vehicles.
[0056] Autonomous driving algorithms generally include perception modules, planning modules, and control modules. Among them, the planning module, as the core decision-making module, is responsible for planning a safe and efficient driving trajectory that conforms to the vehicle dynamics model based on the environmental information provided by the upstream perception module, and handing it over to the downstream control module for execution. Existing planning modules generally use machine learning-based methods and rule-based methods for trajectory planning. Among them, trajectory planning based on machine learning directly generates trajectories through neural networks and relies on human driving trajectories as core training data. Generally speaking, if the training data is of high quality and diverse, the trained neural network will have better ability to cope with different road conditions. However, the disadvantages of this type of method are that it requires high computing power and poor interpretability, and cannot guarantee the safety of the generated trajectory. Rule-based trajectory planning refers to the human driving trajectory, builds corresponding mathematical models for different scene road conditions, and obtains the final trajectory by solving the debugging problem. Due to the existence of mathematical models, this type of method has high interpretability, and various requirements for trajectories (such as upper and lower speed limits, comfort, safety, etc.) can be intuitively embedded into the debugging problem to constrain the solution of the trajectory. However, as the road conditions become more complicated, the rules and requirements that need to be added will become more complex, and the coupling relationship between rules and requirements will also be strengthened accordingly. This will result in a large number of custom parameters or mapping tables in the trajectory planning algorithm code, making debugging more difficult and requiring extremely high professionalism and experience from the debugger.
[0057] In order to solve the above problems, in an embodiment of the present application, a regression trajectory is generated in a preset regression test scenario based on the parameters to be tested, and a regression test score of the regression trajectory is obtained; a planned trajectory is generated in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and a trajectory difference score is determined based on each planned trajectory and the human driving trajectory corresponding to each preset simulation scenario; a parameter score corresponding to the parameter to be tested is determined based on the trajectory difference score and the regression test score; and the parameter to be tested is debugged based on the parameter score to obtain trajectory planning module parameters. The embodiment of the present application compares the planned trajectory with the human driving trajectory to determine the trajectory difference score, and then debugs the trajectory planning module parameters through the trajectory difference score and the regression test score to improve the performance of the trajectory planning algorithm using the trajectory planning module parameters, thereby generating a more accurate planned trajectory.
[0058] The application content is further explained below through the description of embodiments in conjunction with the accompanying drawings.
[0059] This embodiment provides a trajectory planning module parameter debugging method, such as Figure 1 and Figure 2 As shown, the method includes:
[0060] S10: Generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory.
[0061] Specifically, the parameters to be tested are the parameters to be debugged of the trajectory planning module used to generate the driving trajectory, and the trajectory planning module uses the debugged test parameters (i.e., trajectory planning module parameters) to generate the driving trajectory. Among them, the trajectory planning module plans the driving trajectory in a rule-based manner. Among them, the parameters to be tested can be generated using the Optuna tool, or they can be random module parameters randomly generated for the trajectory planning module, or they can be initial module parameters of the trajectory planning module, etc.
[0062] The preset regression test scenario is a pre-collected driving scenario, and the driving scenario can be a challenging driving scenario, and the trajectory quality of the regression trajectory can be more effectively tested through the challenging driving scenario. For example, the preset regression test scenario is as follows Figure 3 The side vehicle insertion scenario shown, Figure 4 The big bend scene shown, such as Figure 5 The lane merging scenario shown is as follows: Figure 6The unprotected left turn scenario shown in the figure is used as a monitoring scenario for the overall performance of the trajectory planning module. The regression trajectory is generated under the preset regression test scenario, and then the regression test score based on the regression trajectory is used as a penalty score to ensure that the trajectory planning quality (such as safety, physical perception, efficiency, traffic, etc.) does not regress.
[0063] It should be noted that the predicted regression scenario can be one or more predicted regression scenarios, and when there are multiple predicted regression scenarios, a regression trajectory can be generated in each predicted regression scenario, and the regression test score of the regression trajectory can be obtained, and then the mean of all regression test scores can be used as the final regression test score, or all regression test scores can be weighted as the final regression test score, etc.
[0064] Furthermore, the regression trajectory can be generated by a trajectory planning module configured with the parameters to be tested in the predicted regression scenario, specifically, Figure 7 As shown in the figure, the trajectory planning module accepts the known information (such as lane lines, traffic lights, speed limits, sidewalks and obstacle locations and corresponding predicted trajectories) provided by upstream modules (such as perception, navigation, prediction, etc.), and plans the driving trajectory based on the predicted goals (such as comfort, safety, efficiency, compliance with traffic rules, etc.) through deep learning and other methods.
[0065] In the embodiment of the present application, the trajectory planning module is embedded in the dynamic model, and then the dynamic model is directly used to plan the driving trajectory, that is, the regression trajectory can be generated by the dynamic model in the preset regression test scenario, and the planning trajectory can be generated in each simulation scenario in the preset simulation scenario set. The dynamic model can be expressed as:
[0066]
[0067] Among them, z(t) represents the value of the main vehicle state at time t, represents the derivative of the main vehicle state at time t, x and y represent the position of the main vehicle in the world coordinate system, h represents the direction of the main vehicle, and v x represents the longitudinal velocity of the host vehicle, v y represents the lateral speed of the main vehicle, δ represents the steering angle of the front wheels of the main vehicle, Represents the trajectory planning module parameters (i.e. the parameters to be tested).
[0068] To this end, after obtaining the parameters to be tested, the above-mentioned dynamic model can be used to plan the regression trajectory under the preset regression test scenario, and then the regression test score of the regression trajectory can be obtained, where the regression test score is used to reflect the trajectory quality of the regression trajectory (such as safety, physical sensation, efficiency, passage, etc.). The regression test score can be determined by a trained quality assessment model, or it can be determined by pre-setting several preset evaluation indicators, and then determining the evaluation score of the regression trajectory in each preset evaluation indicator, and finally determining the regression test score of the regression trajectory based on all evaluation scores.
[0069] In a specific implementation, obtaining the regression test score of the regression trajectory specifically includes:
[0070] Evaluating the evaluation score of the regression trajectory on each preset evaluation indicator in a preset evaluation indicator set;
[0071] A regression test score for the regression trajectory is calculated based on all evaluation scores.
[0072] Specifically, the preset evaluation index set is pre-set, and includes several preset evaluation indicators, each of which is an evaluation indicator for evaluating the quality of the regression trajectory, and each preset evaluation indicator corresponds to an evaluation score, which is used to represent the trajectory quality of the regression trajectory on the preset evaluation indicator. Among them, the preset evaluation indicators may include the acceleration and deceleration of the main vehicle, body sensation, steering angular velocity, distance to surrounding obstacles, distance to lane lines or road boundaries, etc.
[0073] After obtaining the evaluation scores of the regression trajectory on each preset evaluation index, the weighted sum of all evaluation scores can be used as the regression test score of the regression trajectory, or the mean of all evaluation scores can be used as the regression test score of the regression trajectory, etc. In the embodiment of the present application, the weighted sum of all evaluation scores is used as the regression test score of the regression trajectory, and the regression test score is used as a penalty score to guide the debugging process of the parameters to be tested, wherein the relationship between the regression test score and the trajectory quality is that the lower the regression test score, the higher the trajectory quality, and vice versa, the higher the regression test score, the lower the trajectory quality.
[0074] S20, generating a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determining a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario.
[0075] Specifically, the preset simulation scene set is formed by collecting real driving data, and the preset simulation scene set includes several simulation scenes, each of which is formed by collecting human driving trajectories for the same collection scene, and the human driving trajectory is a driving trajectory formed by the driver's artificial driving. For example, several simulation scenes are formed by collecting human driving trajectories in lane change scenes. In this way, by collecting multiple human driving trajectories for the same scene to form multiple simulation scenes, the quality of the driving trajectory planned for the scene can be further improved. In addition, each of the several simulation scenes has a corresponding human driving trajectory, and the human driving trajectory is used as a reference object for the planned trajectory generated in the simulation scene.
[0076] The planned trajectory is the driving trajectory planned by the trajectory planning module in the simulation scenario. That is, the planned trajectory is the driving trajectory generated for the simulation scenario by the closed-loop dynamics model of the main vehicle using the parameters to be tested, and the environmental information of different simulation scenarios may be different. Therefore, when the planned trajectory is generated for the simulation scenario based on the closed-loop dynamics model, the closed-loop dynamics model of the main vehicle can be expanded to: Wherein, i represents the i-th simulation scenario. The expanded closed-loop dynamic model carries the influence of environmental information on the main vehicle, and the integral result of the expanded closed-loop dynamic model is the planned trajectory of the main vehicle. At the same time, the trajectory of the main vehicle essentially depends on the simulation scenario and the parameters to be tested. Therefore, in order to emphasize this dependence, the present application expands the state of the main vehicle, and the expanded state of the main vehicle is recorded as z(t; θ, i). z(t; θ, i) represents the state of the main vehicle at time t based on the parameter θ to be tested in the i-th simulation scenario, wherein each dimensional element of the main vehicle state is dependent on θ and i, that is, z(t; θ, i) = [x(t; θ, i), y(t; θ, i), h(t; θ, i), v x (t;θ,i),v y (t;θ,i),δ(t;θ,i)] T .
[0077] Exemplarily, determining the trajectory difference score based on each planned trajectory and the human driving trajectory corresponding to each preset simulation scenario specifically includes:
[0078] For each planned trajectory, calculating a scene trajectory difference score between the planned trajectory and a human driving trajectory corresponding to a preset simulation scene in which the planned trajectory is located;
[0079] The average of all scene trajectory difference scores is calculated to obtain the trajectory difference score.
[0080] Specifically, the scene trajectory difference score is used to reflect the difference between the planned trajectory and its corresponding human driving trajectory, and the scene trajectory difference score is used to evaluate the performance of the trajectory planning of the parameter θ to be tested in the i-th simulation scene. The trajectory difference score can be the average of all scene trajectory difference scores, the maximum value of all scene trajectory difference scores, or the weighted sum of all scene trajectory difference scores.
[0081] Exemplarily, the calculating of the scene trajectory difference score between the planned trajectory and the human driving trajectory corresponding to the preset simulation scene in which the planned trajectory is located specifically includes:
[0082] For each trajectory moment in the planned trajectory, obtaining a simulated host vehicle state at the trajectory moment in the planned trajectory, and obtaining a real host vehicle state at the trajectory moment in the human driving trajectory;
[0083] The main vehicle state difference between the simulated main vehicle state and the real main vehicle state at each trajectory moment is calculated, and the scene trajectory difference score between the planned trajectory and the human driving trajectory is determined based on the main vehicle state difference at each trajectory moment.
[0084] Specifically, the simulated main vehicle state is the main vehicle state at a certain time in the planned trajectory, denoted as z(t; θ, i), and the real main vehicle state is the main vehicle state at this time in the human driving trajectory, denoted as z h (t; i). The main vehicle state difference is the difference between the simulated main vehicle state and the real main vehicle state. Since the main vehicle state includes the position state of the main vehicle in the world coordinate system, the main vehicle orientation, the longitudinal speed, the lateral speed and the main vehicle front wheel steering angle, the main vehicle state difference can be determined based on the difference in one or more of the position state of the main vehicle in the world coordinate system, the main vehicle orientation, the longitudinal speed, the lateral speed and the main vehicle front wheel steering angle. For example, it can be determined based on the main vehicle orientation, or based on the main vehicle orientation, the longitudinal speed and the lateral speed, or based on the position state, the main vehicle orientation, the longitudinal speed, the lateral speed and the main vehicle front wheel steering angle.
[0085] Furthermore, the state items of the main vehicle state used to determine the main vehicle state difference can be determined based on the real scene corresponding to the simulation scene. Different real scenes can select different state dimensions. For example, the lane change scene corresponds to the position state, the main vehicle orientation, the longitudinal speed, the lateral speed and the main vehicle front wheel steering. The starting scene corresponds to the position state, the main vehicle orientation, the longitudinal speed and the main vehicle front wheel steering. In the embodiment of the present application, the process of determining the main vehicle state difference is explained by taking the lane change scene as an example. Specifically, the process of determining the main vehicle state difference can be:
[0086] First, the dimensional differences of each state dimension are calculated respectively. The calculation formulas for the dimensional differences of the position state, the main vehicle orientation, the longitudinal speed, the lateral speed and the main vehicle front wheel steering at the t trajectory time in the i-th simulation scenario are:
[0087] The lateral displacement difference c1(θ)=||g(x h (t; i),y h (t; i))-g(x(t; θ, i), y(t; θ, i))||2, where function g is used to calculate the lateral displacement of the main vehicle at the corresponding time in the reference frame defined by the initial posture of the main vehicle in the current scene; x h (t; i),y h (t; i) represents the position state of the human driving trajectory at the time t in the i-th simulation scenario, and x(t; θ, i), y(t; θ, i) represents the position state of the planned trajectory at the time t in the i-th simulation scenario;
[0088] The angle difference of the main vehicle direction c2(θ)=|h h (t;i)-h(t;θ,i)|, where, h h (t; i) represents the direction of the main vehicle at the time t of the human driving trajectory in the i-th simulation scenario, and h(t; θ, i) represents the direction of the main vehicle at the time t of the planned trajectory in the i-th simulation scenario;
[0089] Longitudinal speed difference of the host vehicle in, represents the longitudinal velocity of the human driving trajectory at the time t in the ith simulation scenario, v x (t;θ,i) represents the longitudinal velocity of the planned trajectory at time t in the i-th simulation scenario;
[0090] Lateral speed difference of the host vehicle in, represents the lateral velocity of the human driving trajectory at the time t in the ith simulation scenario, v y (t;θ,i) represents the lateral velocity of the planned trajectory at time t in the i-th simulation scenario;
[0091] The front wheel rotation angular velocity difference of the main vehicle c5(θ): = |δ h (t;i)-δ(t;θ,i)|, where, δ h (t; i) represents the front wheel angular velocity of the human driving trajectory at the time t in the i-th simulation scenario, and δ(t; θ, i) represents the front wheel angular velocity of the planned trajectory at the time t in the i-th simulation scenario;
[0092] Among them, c1(θ), c2(θ), c3(θ), c4(θ) and c5(θ) are all functions of t, θ and i, but for the sake of simplicity and highlighting the purpose of the scheme, only θ is indicated here and t and i are omitted.
[0093] Secondly, by weighting the dimension differences of each state dimension, the state difference of the main vehicle can be obtained, where the expression of the state difference of the main vehicle is:
[0094] Main vehicle state difference = w1c1(θ)+w2c2(θ)+w3c3(θ)+w4c4(θ)+w5c5(θ)
[0095] Among them, w1, w2, w3, w4 and w5 all represent weight coefficients.
[0096] Further, after obtaining the host vehicle state difference at each trajectory moment, the scene trajectory difference score between the planned trajectory and the human driving trajectory is calculated according to all host vehicle state differences, wherein the scene trajectory difference score and the trajectory difference score may be calculated as follows:
[0097]
[0098] Among them, c i (θ) represents the scene trajectory difference score of the i-th simulation scene, w1, w2, w3, w4 and w5 all represent weight coefficients, T i represents the time interval of the i-th simulation scenario, I represents the simulation scenario set, |I| represents the number of simulation scenarios in the simulation scenario set, and c(θ) represents the trajectory difference score.
[0099] S30: Determine a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score.
[0100] Specifically, the parameter score is used to reflect the parameter quality of the parameter to be tested and to evaluate the performance of the parameter to be tested, wherein the parameter score is determined based on the trajectory difference score and the regression test score. For example, the sum of the trajectory difference score and the regression test score can be directly used as the parameter score, or the trajectory difference score and the regression test score can be weighted to determine the parameter score. In the embodiment of the present application, the calculation formula for the parameter score can be:
[0101] Parameter score = c(θ) + q(θ)
[0102] Where c(θ) represents the trajectory difference score and q(θ) represents the regression test score.
[0103] S40: Debugging the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters.
[0104] Specifically, debugging the parameters to be tested based on the parameter score refers to determining the performance of the parameters to be tested based on the parameter score, providing real-time feedback for the debugging process of the parameters to be tested, so as to promote the parameters to be tested to be adjusted in the direction of better performance. That is to say, the parameter score is used as the real-time feedback of the debugging process of the parameters to be tested, and the parameters to be tested are adjusted so that the parameter score corresponding to the parameters to be tested after debugging is better than the parameters to be tested before debugging, and then the trajectory planning module parameters can be obtained, and the trajectory quality of the planned trajectory and the fit of the planned trajectory with the human driving trajectory can be improved. To this end, in the process of debugging the parameters to be tested, the calculation formula of the parameter score can be used as the objective function, and then the parameters to be tested can be iteratively debugged with the minimum objective function as the goal to obtain the parameters of the trajectory planning module. That is to say, when determining the parameters of the trajectory planning module, multiple debugging processes of the parameters to be tested will be performed, and for two adjacent debugging processes, the parameters to be tested adopted by the latter debugging process are obtained based on the debugging of the previous debugging process.
[0105] Based on this, the parameters to be tested are debugged based on the parameter score to obtain trajectory planning module parameters, which specifically include:
[0106] generating candidate test parameters according to a preset parameter space based on the parameter scores;
[0107] The candidate test parameters are used as parameters to be tested, and a regression trajectory is generated based on the parameters to be tested in a preset regression test scenario until a preset end condition is met, and the parameters to be tested when the preset end condition is met are used as trajectory planning module parameters.
[0108] Specifically, the preset parameter space is pre-set, and is configured with the parameter range of each test parameter item in the parameters to be tested and the data distribution of each test parameter item, that is, the change rule of each test parameter item and the range that can be changed can be determined by the preset parameter space. The Optuna module can be used to generate candidate test parameters according to the preset parameter space. It is an automated tool for debugging parameters, which uses the TPE (Tree-structured Parzen Estimator) algorithm of Bayesian debugging to balance the exploration and utilization of parameters, and gradually approach the optimal solution. To this end, after obtaining the parameter score, the Optuna module is used to generate candidate test parameters according to the preset parameter space based on the parameter score, and the candidate test parameters are used as the next parameter to be tested to carry out the next round of parameter debugging process, which realizes the automation of the complete debugging process, reduces the difficulty of program debugging and the requirements for debuggers.
[0109] Furthermore, the candidate test parameter can be one or more. When there is one candidate test parameter, the candidate test parameter is directly used as the parameter to be tested, and then the step of generating a regression trajectory based on the parameter to be tested in the preset regression test scenario is re-executed to perform the parameter debugging process of the next iteration round. When there are multiple candidate test parameters, each candidate test parameter can be used as a parameter to be tested, and a parameter debugging task can be created for each parameter to be tested, and then the parameter debugging tasks can be executed in parallel through a distributed task cluster (such as a multi-task cluster built using the distributed computing framework ray-kubernates, etc.) to improve the test adjustment speed.
[0110] For example, when there are multiple candidate test parameters, such as Figure 2 As shown, the method of using the candidate test parameters as the parameters to be tested and generating regression trajectories based on the parameters to be tested in a preset regression test scenario specifically includes:
[0111] Treat each candidate test parameter as a parameter to be tested;
[0112] Each parameter to be tested is assigned to a distributed node in the distributed task cluster, and the distributed node generates a regression trajectory based on the parameter to be tested in a preset regression test scenario to obtain a parameter score corresponding to each parameter to be tested.
[0113] Specifically, the distributed node is a work node in the distributed task cluster, which is used to perform the parameter scoring determination process, that is, each distributed node will execute the steps S10-S30, and then obtain the parameter score of the parameter to be tested assigned to it, and then feed the parameter score back to the Optuna module, which summarizes the parameter scores of all parameters to be tested through the Optuna module, and then generates multiple candidate test parameters based on the preset parameter space to enter the next parameter debugging round until the end condition is reached to obtain the trajectory planning module parameters. Among them, the end condition can be that the parameter score meets the parameter requirements (for example, less than the preset score threshold, etc.) and / or the debugging round reaches the round limit, etc.
[0114] In summary, this embodiment provides a method for debugging parameters of a trajectory planning module, the method comprising: generating a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtaining a regression test score of the regression trajectory; generating a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determining a trajectory difference score based on each planned trajectory and the human driving trajectory corresponding to each preset simulation scenario; determining a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score; debugging the parameter to be tested based on the parameter score to obtain trajectory planning module parameters. This application uses human driving trajectories as trajectory references under driving conditions corresponding to simulation scenarios, uses parameter scores based on human driving trajectories and planned trajectories as feedback information, and uses the Optuna module to debug parameters to be tested, thereby realizing automatic debugging of trajectory planning module parameters, which can not only reduce the difficulty of program debugging and the requirements for debuggers, but also improve the trajectory quality of the planned trajectory and its fit with the human driving trajectory.
[0115] Based on the above trajectory planning module parameter debugging method, this embodiment provides a trajectory planning module parameter debugging device, such as Figure 8 As shown, the trajectory planning module parameter debugging device includes:
[0116] The trajectory planning regression test module 100 is used to generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory;
[0117] A scenario simulation module 200, configured to generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and to determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario;
[0118] A parameter score determination module 300, configured to determine a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score;
[0119] The parameter debugging module 400 is used to debug the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters.
[0120] Based on the above-mentioned trajectory planning module parameter debugging method, this embodiment provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the trajectory planning module parameter debugging method as described in the above-mentioned embodiment.
[0121] Based on the above-mentioned trajectory planning module parameter debugging method, the present application also provides a server end, which includes a number of distributed nodes and a central node, each distributed node communicates with the central node, the central node is used to send the parameters to be tested to each distributed node, and each distributed node is used to generate a regression trajectory under a preset regression test scenario based on the parameters to be tested, and obtain the regression test score of the regression trajectory; based on the parameters to be tested, a planned trajectory is generated in each simulation scenario in a preset simulation scenario set, and a trajectory difference score is determined based on each planned trajectory and the human driving trajectory corresponding to each preset simulation scenario; based on the trajectory difference score and the regression test score, a parameter score corresponding to the parameter to be tested is determined, and the parameter score is fed back to the central node, and the central node is also used to debug the parameters to be tested based on the parameter score to obtain the trajectory planning module parameters.
[0122] Among them, Fig. 9 As shown, the distributed node and the central node may include at least one processor 20, a display screen 21, and a memory 22, and may also include a communications interface 23 and a bus 24. The processor 20, the display screen 21, the memory 22, and the communications interface 23 may communicate with each other through the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setting mode. The communications interface 23 may transmit information. The processor 20 may call the logic instructions in the memory 22 to execute the method in the above embodiment.
[0123] In addition, the logic instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.
[0124] The memory 22 is a computer-readable storage medium that can be configured to store software programs, computer executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions or modules stored in the memory 22, that is, implementing the methods in the above embodiments.
[0125] The memory 22 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 22 may include a high-speed random access memory and may also include a non-volatile memory. For example, a variety of media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, may also be a transient storage medium.
[0126] In addition, the specific process of loading and executing the multiple instruction processors in the above-mentioned storage medium and the terminal device has been described in detail in the above-mentioned method, and will not be described one by one here.
[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A trajectory planning module parameter debugging method, characterized in that: The trajectory planning module parameter debugging method includes: Generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory; Generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario; Determining a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score; The parameters to be tested are debugged based on the parameter scores to obtain trajectory planning module parameters.
2. The trajectory planning module parameter debugging method according to claim 1, characterized in that: The debugging of the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters specifically includes: generating candidate test parameters according to a preset parameter space based on the parameter scores; The candidate test parameters are used as parameters to be tested, and a regression trajectory is generated based on the parameters to be tested in a preset regression test scenario until a preset end condition is met, and the parameters to be tested when the preset end condition is met are used as trajectory planning module parameters.
3. The trajectory planning module parameter debugging method according to claim 2, characterized in that: The preset parameter space is configured with the parameter range of each test parameter item in the parameters to be tested and the data distribution of each test parameter item; the generation of candidate test parameters according to the preset parameter space based on the parameter score is specifically: The Optuna module generates candidate test parameters according to a preset parameter space based on the parameter scores.
4. The trajectory planning module parameter debugging method according to claim 2, characterized in that: When there are multiple candidate test parameters, the step of using the candidate test parameters as the parameters to be tested and generating a regression trajectory based on the parameters to be tested in a preset regression test scenario specifically includes: Treat each candidate test parameter as a parameter to be tested; Each parameter to be tested is assigned to a distributed node in the distributed task cluster, and the distributed node generates a regression trajectory based on the parameter to be tested in a preset regression test scenario to obtain a parameter score corresponding to each parameter to be tested.
5. The trajectory planning module parameter debugging method according to claim 1, characterized in that: The step of obtaining the regression test score of the regression trajectory specifically includes: Evaluating the evaluation score of the regression trajectory on each preset evaluation indicator in a preset evaluation indicator set; A regression test score for the regression trajectory is calculated based on all evaluation scores.
6. The trajectory planning module parameter debugging method according to claim 1, characterized in that: Determining the trajectory difference score based on each planned trajectory and the human driving trajectory corresponding to each preset simulation scenario specifically includes: For each planned trajectory, calculating a scene trajectory difference score between the planned trajectory and a human driving trajectory corresponding to a preset simulation scene in which the planned trajectory is located; The average of all scene trajectory difference scores is calculated to obtain the trajectory difference score.
7. The trajectory planning module parameter debugging method according to claim 6, characterized in that: The calculating of the scene trajectory difference score between the planned trajectory and the human driving trajectory corresponding to the preset simulation scene where the planned trajectory is located specifically includes: For each trajectory moment in the planned trajectory, obtaining a simulated host vehicle state at the trajectory moment in the planned trajectory, and obtaining a real host vehicle state at the trajectory moment in the human driving trajectory; The main vehicle state difference between the simulated main vehicle state and the real main vehicle state at each trajectory moment is calculated, and the scene trajectory difference score between the planned trajectory and the human driving trajectory is determined based on the main vehicle state difference at each trajectory moment.
8. A trajectory planning module parameter debugging device, characterized in that: The trajectory planning module parameter debugging device comprises: A trajectory planning regression test module is used to generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory; A scenario simulation module, configured to generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario; A parameter score determination module, used to determine the parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score; A parameter debugging module is used to debug the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the trajectory planning module parameter debugging method as described in any one of claims 1-7.
10. A server, characterized in that: include: Several distributed nodes and central nodes; The distributed node is used to generate a regression trajectory in a preset regression test scenario based on the parameters to be tested, and obtain a regression test score of the regression trajectory; Generate a planned trajectory in each simulation scenario in a preset simulation scenario set based on the parameters to be tested, and determine a trajectory difference score based on each planned trajectory and a human driving trajectory corresponding to each preset simulation scenario; Determining a parameter score corresponding to the parameter to be tested based on the trajectory difference score and the regression test score; The central node is used to debug the parameters to be tested based on the parameter scores to obtain trajectory planning module parameters.