Planning AI model online evaluation method and device, electronic equipment and storage medium

Abstract

The invention discloses a planning AI model online evaluation method and device, electronic equipment and a storage medium, and the method comprises the steps: enabling a planning AI model to be deployed on an automatic driving vehicle, and enabling the automatic driving vehicle to comprise a rule planning model; acquiring scene data; respectively inputting the scene data into the rule planning model and the planning AI model; and evaluating the track feasibility of the second output result of the planning AI model, and selecting the first output result of the rule planning model or the second output result of the planning AI model according to the evaluation result as the planning track of the autonomous vehicle. According to the method and the device, online evaluation on the planning AI model is realized, so that the AI model can be applied to various automatic driving scenes.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular, to a method, apparatus, electronic device, and storage medium for evaluating the online deployment of a planning AI model. Background Art

[0002] In an autonomous driving system, the planning module is located between the perception module and the control module. The perception module provides static and dynamic environment information, and the control module executes the planned actions. The role of the planning module is to convert this information into specific driving strategies and path planning. The planning model usually uses a rule-based planning model, and there are also those that use an AI-based planning model.

[0003] Regardless of the model used, when deployed on an autonomous driving vehicle, it is difficult to apply because the planning model is difficult to cover all scenarios. Summary of the Invention

[0004] Embodiments of this application provide a method, apparatus, electronic device, and storage medium for evaluating the online deployment of a planning AI model to evaluate the online deployment of the planning AI model based on determined scenarios.

[0005] Embodiments of this application adopt the following technical solutions:

[0006] In a first aspect, embodiments of this application provide a method for evaluating the online deployment of a planning AI model, where the method includes:

[0007] Deploy the planning AI model online on an autonomous driving vehicle, where the autonomous driving vehicle includes a rule-based planning model;

[0008] Obtain scenario data;

[0009] Input the scenario data into the rule-based planning model and the planning AI model respectively;

[0010] Evaluate the trajectory feasibility of the second output result of the planning AI model, and select the first output result of the rule-based planning model or the second output result of the planning AI model as the planning trajectory of the autonomous driving vehicle according to the evaluation result.

[0011] In some embodiments, the evaluating the trajectory feasibility of the second output result of the planning AI model, and selecting the first output result of the rule-based planning model or the second output result of the planning AI model as the planning trajectory of the autonomous driving vehicle includes:

[0012] Based on a pre-set feedback mechanism, obtain the first intention in the first output result of the rule-based planning model and the second intention in the second output result of the planning AI model;

[0013] If it is determined that the first intention is inconsistent with the second intention, then the first output result is used to determine the feasibility of the AI trajectory in the second output result;

[0014] If it is determined that the first intention is consistent with the second intention, then the trajectory similarity of the autonomous vehicle is evaluated.

[0015] In some embodiments, the evaluating the trajectory similarity of the autonomous vehicle includes:

[0016] In the case of having the same intention, obtain the trajectory information in the second output result of the planning AI model and the trajectory information in the first output result of the rule planning model;

[0017] Evaluate the similarity between the trajectory information in the second output result of the planning AI model and the trajectory information in the first output result of the rule planning model to determine the feasibility of the AI trajectory;

[0018] In the case of determining that the AI trajectory is feasible, use the second output result of the AI model as the planned trajectory of the autonomous vehicle.

[0019] In some embodiments, the using the first output result to determine the feasibility of the AI trajectory in the second output result includes:

[0020] In the case of having different intentions, feedback the second intention to the rule planning model to determine the feasibility of the AI trajectory under the current decision output by the rule planning model;

[0021] In the case of determining that the AI trajectory is infeasible, use the first output result of the rule planning model as the planned trajectory of the autonomous vehicle.

[0022] In some embodiments, the using the first output result of the rule planning model or the second output result of the planning AI model as the planned trajectory of the autonomous vehicle further includes:

[0023] Obtain the output result of the rule planning model for the second intention and record the scene data at the current moment;

[0024] Iterate the output result and the current moment scene data to the planning AI model;

[0025] And / or,

[0026] Save the entry conditions and exit conditions of the scene where the planning AI model is deployed on the autonomous vehicle after going online;

[0027] Use the entry conditions and exit conditions of the scenario where the planned AI model is deployed on the autonomous vehicle after going online as the trigger conditions for the rule planning model that replaces the planned AI model.

[0028] In some embodiments, the obtaining of scenario data includes:

[0029] During the training phase of the planned AI model, collect all the data of manual driving. The all data at least includes scenario data, and the scenario data includes the entry conditions and exit conditions of the scenario.

[0030] During the verification phase of the planned AI model, collect the feature data of the target at each moment. The feature data includes the data available for the rule planning model.

[0031] In some embodiments, the planned AI model includes:

[0032] Collect scenario data to improve the user's riding experience, and obtain a data set of preset scenarios.

[0033] Train an AI model according to the data set of the preset scenario.

[0034] Set a preset feedback mechanism.

[0035] According to the planned AI model and the preset feedback mechanism, obtain the planned AI model.

[0036] In a second aspect, an embodiment of the present application further provides an evaluation device for the online of a planned AI model. Among them, the device includes:

[0037] A deployment module for deploying the planned AI model online on an autonomous vehicle, and the autonomous vehicle includes a rule planning model.

[0038] An acquisition module for acquiring scenario data.

[0039] An input module for inputting the scenario data into the rule planning model and the planned AI model respectively.

[0040] An evaluation module for evaluating the trajectory feasibility of the second output result of the planned AI model, and selecting the first output result of the rule planning model or the second output result of the planned AI model according to the evaluation result as the planned trajectory of the autonomous vehicle.

[0041] In a third aspect, an embodiment of the present application further provides an electronic device, including: a processor; and a memory arranged to store computer-executable instructions, and the executable instructions, when executed, cause the processor to execute the above method.

[0042] Fourthly, an embodiment of the present application further provides a computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, cause the electronic device to execute the above method.

[0043] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: First, the planned AI model is deployed online on an autonomous vehicle, and the autonomous vehicle includes a rule planning model. Then, in an actual use scenario, scenario data is obtained; and the scenario data is input into the rule planning model and the planned AI model respectively. Further, the trajectory feasibility of the second output result of the planned AI model is evaluated, and the first output result of the rule planning model or the second output result of the planned AI model is used as the planned trajectory of the autonomous vehicle. Through the above method, the trajectory feasibility evaluation of the AI model deployed online on the autonomous vehicle under a determined scenario is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments and descriptions thereof of the present application are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0045] FIG. 1(a) is a schematic diagram of the implementation principle of the method for evaluating the online deployment of the planned AI model in the embodiment of the present application;

[0046] FIG. 1(b) is a schematic diagram of the training and deployment of the AI model in the method for evaluating the online deployment of the planned AI model in the embodiment of the present application;

[0047] Figure 2 is a schematic flowchart of the method for evaluating the online deployment of the planned AI model in the embodiment of the present application;

[0048] Figure 3 is a schematic structural diagram of the device for evaluating the online deployment of the planned AI model in the embodiment of the present application;

[0049] Figure 4 is a schematic structural diagram of an electronic device in the embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0051] The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0052] As shown in Figure 1(a), an autonomous driving vehicle mainly includes a perception module, a planning module, and a control module. The perception module includes the perception results of multiple sensors, and inputs the perception results of the target at each moment into the planning module. Data is collected at each moment and input into the planning model and the traditional planning model respectively according to the scenario conditions. Then, the feasibility of the planned output trajectory is judged, the traditional planning model is adopted, and the data is recorded for the iteration of the rule planning model. Finally, the planned trajectory is output according to the feasibility evaluation result. Among them, the planning model mainly refers to the planning AI model. The planning AI model is a system that can simulate human intelligent behavior obtained through computer algorithms and data training. It uses technologies such as machine learning to input known data into the computer for training, and finally generates a system that can automatically complete specific tasks. The core of the planning AI model is algorithms and data. Through continuous iterative learning and optimization, the model can continuously improve accuracy and efficiency. Traditional planning mainly refers to the rule-based RB (Rule Based) planning model. As shown in Figure 1(b), the training and deployment of the planning AI model include: collecting human driving data, constructing the planning AI model, and deploying it on the vehicle. That is to say, the planning AI model is pre-trained based on the collected human driving data and then deployed on the autonomous driving vehicle, and a feasibility evaluation is carried out after the deployment on the vehicle.

[0053] The embodiments of the present application provide a method for evaluating the online deployment of a planning AI model. As Figure 2 shown, a schematic flow chart of the method for evaluating the online deployment of the planning AI model in the embodiments of the present application is provided. The method at least includes the following steps S210 to step S240:

[0054] It should be noted that steps S210 to S240 refer to the actual verification stage, and the planning AI model has been obtained through training in the training stage.

[0055] Step S210, deploy the planning AI model online on an autonomous driving vehicle, and the autonomous driving vehicle includes a rule-based planning model.

[0056] After the planning AI model is pre-trained until convergence and meets the conditions for online deployment of the model, the planning AI model can be deployed online on the autonomous driving vehicle. At the same time, a rule-based planning model has been deployed and included in the autonomous driving system of the autonomous driving vehicle.

[0057] The rule-based planning model refers to the model of traditional planning algorithms. Traditional planning mainly refers to the rule-based RB (RuleBased) rule model. The planning AI model refers to the AI model.

[0058] Step S220: Obtain scenario data.

[0059] The scenario data needs to be obtained in advance for the online evaluation of the model, and the selection of scenario data mainly focuses on the scenario data that needs to improve the driving experience. The scenario data is mainly the full amount of data collected from manual driving data. The full amount of manual driving data refers to all data records in the autonomous driving system at the current time point. These data records include all information up to the current time point, and the latest status is retained regardless of whether the data has changed. The collected full amount of data can be used in scenarios that require a comprehensive understanding of the system status or historical data analysis.

[0060] The collected manual driving data also includes the entry conditions and exit conditions of the scenario. It should be noted that the "entry conditions of the scenario" include, but are not limited to, the trigger entry conditions for changes in the vehicle's own conditions such as the vehicle changing lanes or overtaking; the "exit conditions of the scenario" include, but are not limited to, the end exit conditions in response to changes in the vehicle's own conditions such as the vehicle changing lanes or overtaking. The above scenarios of the vehicle changing lanes are only examples, and the "entry conditions of the scenario" and the "exit conditions of the scenario" can also cover a wider range, such as true value scenarios based on maps (intersections / road segments), target (preceding vehicle) scenarios, etc. These true value scenarios of manual driving data can be used as conditions for judging autonomous driving data later.

[0061] It can be understood that obtaining scenario data in the current step S220 refers to the data verification stage and does not include the data training stage, but the scenario data obtained in the training stage and the verification stage is of the same dimension.

[0062] Step S230: Input the scenario data into the rule-based planning model and the planning AI model respectively.

[0063] The scenario data will be input into the rule-based planning model and the planning AI model. The planning AI model can include, but is not limited to, a combination of one AI model or multiple AI models, and the rule-based planning model uses one rule-based model or multiple rule-based models.

[0064] In the embodiments of the present application, the rule-based planning model is not specifically limited. Similarly, for the planning AI model, as long as it meets the requirements of planning, it does not need to be specifically limited.

[0065] Step S240: Evaluate the trajectory feasibility of the second output result of the planning AI model, and select the first output result of the rule-based planning model or the second output result of the planning AI model according to the evaluation result as the planned trajectory of the autonomous driving vehicle.

[0066] The first output result of the rule planning model includes a planned trajectory selected by a rule engine, and the second output result of the planning AI model includes a planned trajectory generated by AI. Further, the trajectory feasibility of the autonomous vehicle can be evaluated by comparing the first output result and the second output result.

[0067] If the trajectory feasibility of the second output result of the planning AI model is feasible, then use the planned trajectory of the planning AI model as the planned trajectory for autonomous driving. If the trajectory feasibility of the second output result of the planning AI model is infeasible, then use the planned trajectory of the rule planning model as the planned trajectory for autonomous driving.

[0068] According to the planned trajectory of the autonomous vehicle that meets the requirements of feasibility in the evaluation result, deploy the planning AI model on the autonomous vehicle. Thus, after the planning AI model is deployed online, the trajectory feasibility evaluation can be carried out on the vehicle. The planning AI model deployed on the autonomous vehicle has the ability of self-learning, so as to cover a variety of scenarios.

[0069] Through the above method, first deploy the planning AI model on the autonomous vehicle online. The autonomous vehicle includes a rule planning model. Then, in the actual use scenario, obtain the scenario data; and input the scenario data into the rule planning model and the planning AI model respectively. Further, evaluate the trajectory feasibility of the second output result of the planning AI model, and select the first output result of the rule planning model or the second output result of the planning AI model according to the evaluation result as the planned trajectory of the autonomous vehicle. Therefore, the experience can be improved for a specific scenario based on the manual driving data in the scenario, so that the planning AI model can not only cover most scenarios but also better improve the experience after being deployed online.

[0070] Through the above method, when evaluating the trajectory feasibility of the autonomous vehicle, first judge based on the trajectory intention, and then judge the trajectory similarity, so as to accurately judge the feasibility of the AI trajectory of the planning AI model.

[0071] Different from the related technology, when deployed on the vehicle, it is difficult to apply because the rule planning model is difficult to cover all scenarios. Through the above method, not only the planning AI model is introduced, but also the feasibility of the planned trajectory output by the planning AI model is evaluated, so as to ensure that the planning AI model can cover most scenarios, thereby improving the applicability of the model.

[0072] In an embodiment of the present application, evaluating the trajectory feasibility of the second output result of the planning AI model and selecting the first output result of the rule planning model or the second output result of the planning AI model as the planned trajectory of the autonomous vehicle according to the evaluation result includes: based on a preset feedback mechanism, obtaining the first intention in the first output result of the rule planning model and the second intention in the second output result of the planning AI model; if it is determined that the first intention is inconsistent with the second intention, using the first output result to judge the feasibility of the AI trajectory in the second output result; if it is determined that the first intention is consistent with the second intention, evaluating the trajectory similarity of the autonomous vehicle.

[0073] The feedback mechanism is used to evaluate the trajectory feasibility of the autonomous vehicle. Specifically, the feedback mechanism refers to the intention similarity and trajectory similarity between the trajectory output by the AI model and the trajectory output based on the rule algorithm. It can be understood that the intention similarity includes, but is not limited to, decision-making intention, lane-changing intention, overtaking intention, etc. The trajectory similarity includes, but is not limited to, the process of following the vehicle at different accelerations by the vehicle in front and behind, first accelerating and then decelerating.

[0074] Based on the feedback mechanism, the first intention in the first output result of the rule planning model and the second intention in the second output result obtained are used as the primary judgment basis. If the intentions are different, there is no need to continue the comparison. That is, if it is determined that the first intention is inconsistent with the second intention, the first output result is used to judge the feasibility of the AI trajectory in the second output result. The first output result, as the output result of the rule planning model, can be used to evaluate whether the feasibility of the AI trajectory in the second output result meets the requirements.

[0075] On the contrary, if the intentions are the same, further comparison is needed. That is, if it is determined that the first intention is consistent with the second intention, the trajectory similarity of the autonomous vehicle is evaluated. Only on the premise of similar intentions, will the trajectory similarity be further judged and the trajectory similarity of the autonomous vehicle be evaluated.

[0076] In an embodiment of the present application, evaluating the trajectory similarity of the autonomous vehicle includes: in the case of the same intention, obtaining the trajectory information in the second output result of the planning AI model and the trajectory information in the first output result of the rule planning model; evaluating the similarity between the trajectory information in the second output result of the planning AI model and the trajectory information in the first output result of the rule planning model to judge the feasibility of the AI trajectory; in the case of determining that the AI trajectory is feasible, using the second output result of the AI model as the planned trajectory of the autonomous vehicle.

[0077] To better obtain the evaluation results, when the intents are the same (the intent of the AI model and the intent of the rule-based model), the similarity between the trajectory information in the second output result of the evaluation planning AI model and the trajectory information in the first output result of the planning model is evaluated to determine the feasibility of the AI trajectory. Preferably, the similarity of the trajectory information can be obtained by calculating the Euclidean distance, thereby determining the feasibility of the AI trajectory. That is to say, the trajectory information in the first output result of the planning model is used as prior data to judge the feasibility of the AI trajectory. When it is determined that the AI trajectory is feasible, the second output result of the AI model is used as the planned trajectory of the autonomous vehicle.

[0078] In an embodiment of the present application, using the first output result to judge the feasibility of the AI trajectory in the second output result includes: in the case of different intents, feeding back the second intent to the rule-based planning model to judge the feasibility of the AI trajectory under the current decision output by the rule-based planning model; in the case where it is determined that the AI trajectory is not feasible, using the first output result of the rule-based planning model as the planned trajectory of the autonomous vehicle.

[0079] To better obtain the evaluation results, when there are different intents (the intent of the AI model and the intent of the rule-based model), the intent of the AI trajectory is fed back to the traditional rule algorithm model, so that the traditional rule model can judge the feasibility of the AI trajectory under the current decision.

[0080] The different intents mainly include, but are not limited to, the AI trajectory output having an overtaking intent while the traditional algorithm output has a following intent.

[0081] Determining the feasibility of the AI trajectory means that in the rule-based planning model, through optimization or search space, and different decisions will have different search spaces. For example, in the rule-based planning model, it is searched by following the vehicle to determine at what speed, acceleration, and steering angle to follow the vehicle.

[0082] When judging the feasibility of the AI trajectory under the current decision output by the rule-based planning model, for example, in the case of overtaking, a result is obtained under different decisions. If the direction of the rule-based planning model is inconsistent with the direction of the AI trajectory, it is considered that the feasibility of the AI trajectory does not meet the conditions. When it is determined that the AI trajectory is not feasible, the first output result of the rule-based planning model is used as the planned trajectory of the autonomous vehicle.

[0083] In one embodiment of the present application, using the first output result of the rule planning model or the second output result of the planning AI model as the planned trajectory of the autonomous vehicle includes: obtaining the output result of the rule planning model for the second intention and recording the scene data at the current moment; iterating the output result and the current moment scene data to the planning AI model; and / or saving the entry conditions and exit conditions of the scene after the planning AI model is deployed on the autonomous vehicle; using the entry conditions and exit conditions of the scene after the planning AI model is deployed on the autonomous vehicle as the trigger conditions for the rule planning model replacing the planning AI model.

[0084] If the trajectory evaluation feasibility in the evaluation result does not meet the requirements, then for the output result of the rule planning model for the second intention obtained and the scene data at the current moment recorded, at this time, "the output result of the planning model for the second intention and the scene data at the current moment" is the situation where the planning AI model cannot output the planned trajectory well. Therefore, it is necessary to iterate the output result and the current moment scene data to the planning AI model to improve and optimize the planning AI model. At this time, the planning data of the rule planning model is output.

[0085] If the trajectory evaluation feasibility in the evaluation result meets the requirements, then save the entry conditions and exit conditions of the scene after the planning AI model is deployed on the autonomous vehicle; the entry conditions and exit conditions of the scene after the planning AI model is deployed on the autonomous vehicle can be used as the trigger conditions for the rule planning model replacing the planning AI model. "The trigger conditions for the rule planning model replacing the planning AI model" is the situation where the planning AI model can output the planned trajectory well. Therefore, the entry conditions and exit conditions in the model can be iterated and used as the trigger conditions for the rule planning model replacing the planning AI model. At this time, the planning data of the planning AI model is output.

[0086] Specifically, when setting the entry and exit conditions of the scene, the collected data is mined according to the setting rules for entering and exiting the scene, so as to obtain the data set of the specified scene. The data set of the scene includes the following aspects:

[0087] (1) For scene classification, reference can be made to the division based on road scenes in the ISO-34504 standard, including but not limited to: going straight at an intersection, turning left at an intersection, turning right at an intersection, etc.

[0088] (2) For collecting scene data, the data includes but not limited to the information of the vehicle itself, the historical information of surrounding obstacles obtained from perception, the traffic light state, and the local map information, which are mainly used as the input of the planning AI model.

[0089] In addition, for the scene entry and exit conditions, road position information can be used. For example, the entry condition is: at a position 50 m from the solid line of the intersection ahead, and the exit condition is: at a position 30 m after entering the straight lane after passing through the intersection.

[0090] In an embodiment of the present application, the obtaining of the scene data includes: during the training stage of the planning AI model, collecting all the data of manual driving, where the all data at least includes scene data, and the scene data includes the entry condition and the exit condition of the scene; during the verification stage of the planning AI model, collecting the feature data of the target at each moment, where the feature data includes the data that can be used for the rule planning model.

[0091] During the training stage of the planning AI model, mainly collect the historical all data of manual driving. The all data at least includes scene data, and the scene data includes the entry condition and the exit condition of the scene. During the verification stage of the planning AI model, specifically collect the feature data of the target at each moment (the speed of the target, the heading angle, the position of the target), and the feature data includes the data that can be used for the rule planning model.

[0092] Preferably, constructing the AI model generally refers to the planning AI model, such as planTF, etc. First, the process of training on the all data set and then Finetuning is adopted because when collecting data, the data of other scenes is often much larger than the data of the specified scene. Using the all data set for pre-training is to give full play to the data advantages to improve the generalization of the model. And Finetuning on the data of the specified scene is to make its reasoning more reliable for this scene on the basis of having the above generalization.

[0093] It can be understood that the AI model of planTF is only an example and is not used to limit the protection scope in the embodiments of the present application.

[0094] In an embodiment of the present application, the planning AI model includes: collecting the scene data to be improved in the user riding experience to obtain a data set of the preset scene; training an AI model according to the data set of the preset scene; setting a preset feedback mechanism; and obtaining the planning AI model according to the planning AI model and the preset feedback mechanism.

[0095] When training the AI model, collect the scene data to be improved in the user riding experience to obtain a data set of the preset scene, and then train an AI model according to the data set of the preset scene. In addition, it also includes receiving the data of the rule planning model and recording the data for model iteration when the feasibility judgment of the output result of the planning AI model does not meet the requirements.

[0096] It is also necessary to set up a preset feedback mechanism, which takes into account not only the similarity of trajectory intentions but also the similarity of trajectories.

[0097] Finally, based on the planning AI model and the preset feedback mechanism, the planning AI model is obtained.

[0098] The embodiment of the present application also provides a device 300 for evaluating the online deployment of a planning AI model, as Figure 3 shown in the structural schematic diagram of the device for evaluating the online deployment of the planning AI model in the embodiment of the present application. The device 300 for evaluating the online deployment of the AI model at least includes: a deployment module 310, an acquisition module 320, an input module 330, and an evaluation module 340, where:

[0099] In an embodiment of the present application, the deployment module 310 is specifically configured to: deploy the planning AI model online on an autonomous vehicle, and the autonomous vehicle includes a rule planning model.

[0100] After the planning AI model is pre-trained until convergence and meets the conditions for online deployment of the model, the planning AI model can be deployed online on the autonomous vehicle. At the same time, a rule planning model has been deployed and included in the autonomous driving system of the autonomous vehicle.

[0101] The rule planning model refers to the model of traditional planning algorithms, and traditional planning mainly refers to the rule-based RB (Rule Based) rule model. The planning AI model refers to the AI model.

[0102] In an embodiment of the present application, the acquisition module 320 is specifically configured to: acquire scenario data.

[0103] The scenario data needs to be pre-acquired for the online evaluation of the model, and the selection of the scenario data mainly focuses on the scenario data that needs to improve the driving experience. The scenario data is mainly the full amount of data collected from manual driving data. The full amount of manual driving data refers to all the data records in the autonomous driving system at the current time point. These data records include all the information up to the current time point, and the latest status is retained regardless of whether the data has changed. The collected full amount of data can generally be used in scenarios that require a comprehensive understanding of the system status or historical data analysis.

[0104] The collected manual driving data also includes the entry conditions and exit conditions of the scenarios. It should be noted that the "entry conditions of the scenarios" include, but are not limited to, the trigger entry conditions for changes in the vehicle's own conditions such as the vehicle changing lanes or overtaking; the "exit conditions of the scenarios" include, but are not limited to, the end exit conditions in response to changes in the vehicle's own conditions such as the vehicle changing lanes or overtaking. The above scenarios of the vehicle changing lanes are only examples, and the "entry conditions of the scenarios" and the "exit conditions of the scenarios" can also cover a wider range, such as true value scenarios based on maps (intersections / road sections), target (preceding vehicle) scenarios, etc. These true value scenarios of the manual driving data can be used as conditions for judging autonomous driving data later.

[0105] It can be understood that obtaining scenario data currently refers to the data verification stage and does not include the data training stage, but the scenario data obtained in the training stage and the verification stage are data of the same dimension.

[0106] In an embodiment of the present application, the input module 330 is specifically configured to: input the scenario data into a rule planning model and a planning AI model respectively.

[0107] The scenario data will be input into the rule planning model and the planning AI model. The planning AI model can include, but is not limited to, a combination of one AI model or multiple AI models, and the rule planning model uses one rule-based model or multiple rule-based models.

[0108] In the embodiments of the present application, the rule planning model is not specifically limited. Similarly, for the planning AI model, as long as it meets the requirements of planning, it does not need to be specifically limited.

[0109] In an embodiment of the present application, the evaluation module 340 is specifically configured to: evaluate the trajectory feasibility of the first output result of the rule planning model and the second output result of the planning AI model, and use the output trajectory corresponding to the evaluation result of the trajectory feasibility as the planned trajectory of the autonomous driving vehicle.

[0110] The first output result of the rule planning model includes a planned trajectory selected by a rule engine, and the second output result of the planning AI model includes a planned trajectory generated by AI. Further, the trajectory feasibility of the autonomous driving vehicle can be evaluated by comparing the first output result and the second output result.

[0111] If the trajectory feasibility of the second output result of the planning AI model is feasible, the planned trajectory of the planning AI model is used as the planned trajectory for autonomous driving. If the trajectory feasibility of the second output result of the planning AI model is not feasible, the planned trajectory of the planning model is used as the planned trajectory for autonomous driving.

[0112] According to the planned trajectory of the autonomous vehicle that meets the requirements in the evaluation result, the planned AI model is deployed on the autonomous vehicle. Thus, after the planned AI model is deployed, the trajectory feasibility evaluation is carried out on the vehicle. The planned AI model deployed on the autonomous vehicle has the ability of self-learning, so as to cover a variety of scenarios.

[0113] In an embodiment of the present application, the evaluation module 340 is further configured to:

[0114] Based on a pre-set feedback mechanism, obtain the first intention in the first output result of the rule planning model and the second intention in the second output result of the planned AI model; if it is determined that the first intention is inconsistent with the second intention, then use the first output result to judge the feasibility of the AI trajectory in the second output result; if it is determined that the first intention is consistent with the second intention, then evaluate the trajectory similarity of the autonomous vehicle.

[0115] In an embodiment of the present application, the evaluation module 340 is further configured to:

[0116] In the case of the same intention, obtain the trajectory information in the second output result of the planned AI model and the trajectory information in the first output result of the rule planning model; evaluate the similarity between the trajectory information in the second output result of the planned AI model and the trajectory information in the first output result of the rule planning model to judge the feasibility of the AI trajectory; in the case of determining that the AI trajectory is feasible, use the second output result of the AI model as the planned trajectory of the autonomous vehicle.

[0117] In an embodiment of the present application, the evaluation module 340 is further configured to:

[0118] In the case of different intentions, feedback the second intention to the rule planning model to judge the feasibility of the AI trajectory under the current decision output by the rule planning model; in the case of determining that the AI trajectory is infeasible, use the first output result of the rule planning model as the planned trajectory of the autonomous vehicle.

[0119] In an embodiment of the present application, the evaluation module 340 is further configured to:

[0120] Obtain the output result of the rule planning model for the second intention and record the scene data at the current moment; Iterate the output result and the current moment scene data to the planning AI model; and / or, Save the entry conditions and exit conditions of the scene after the planning AI model is deployed on the autonomous vehicle; Use the entry conditions and exit conditions of the scene after the planning AI model is deployed on the autonomous vehicle as the trigger conditions for the rule planning model that replaces the planning AI model.

[0121] In an embodiment of the present application, the obtaining module 320 is further configured to:

[0122] In the training stage of the planning AI model, collect all the data of manual driving, and the all data at least includes scene data, and the scene data includes the entry conditions and exit conditions of the scene; In the verification stage of the planning AI model, collect the feature data of the target at each moment, and the feature data includes the data available for the rule planning model.

[0123] In an embodiment of the present application, the planning AI model includes: Collect the scene data to be improved for the user riding experience to obtain a data set of preset scenes; Train an AI model according to the data set of the preset scenes; Set a preset feedback mechanism; Obtain the planning AI model according to the planning AI model and the preset feedback mechanism.

[0124] It can be understood that the above AI model online evaluation device can implement each step of the AI model online evaluation method provided in the foregoing embodiments. The relevant explanations of the AI model online evaluation method are applicable to the AI model online evaluation device, and will not be elaborated here.

[0125] Figure 4 It is a schematic structural diagram of an electronic device according to an embodiment of the present application. Please refer to Figure 4 , at the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. Among them, the memory may include a memory, such as a high-speed random access memory (Random-Access Memory, RAM), and may also include a non-volatile memory, such as at least one disk memory, etc. Of course, the electronic device may also include other hardware required for other services.

[0126] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 4 only a bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0127] Memory, used to store programs. Specifically, the program can include program code, and the program code includes computer operation instructions. The memory can include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0128] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming an AI model online evaluation device at the logical level. The processor executes the program stored in the memory and is specifically used to perform the following operations:

[0129] Obtain scenario data;

[0130] Input the scenario data into a rule planning model and a planning AI model respectively;

[0131] Evaluate the trajectory feasibility of the autonomous vehicle according to the first output result of the rule planning model and the second output result of the planning AI model; and

[0132] Deploy the planning AI model online on the autonomous vehicle according to the evaluation result.

[0133] The above as in this application Figure 2The method executed by the AI model online evaluation device disclosed in the illustrated embodiment can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, the steps of the above method can be completed by the integrated logic circuit in the hardware of the processor or instructions in software form. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by a combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method.

[0134] The electronic device can also execute Figure 2 the method executed by the AI model online evaluation device in Figure 2 the illustrated embodiment and implement the functions of the AI model online evaluation device in

[0135] Embodiments of the present application also propose a computer-readable storage medium that stores one or more programs. The one or more programs include instructions that, when executed by an electronic device including multiple application programs, can enable the electronic device to execute Figure 2 the method executed by the AI model online evaluation device in the illustrated embodiment and specifically used to execute:

[0136] Obtain scenario data;

[0137] Input the scenario data into the rule planning model and the planning AI model respectively;

[0138] Evaluate the trajectory feasibility of the autonomous vehicle according to the first output result of the rule planning model and the second output result of the planning AI model; and

[0139] Deploy the planning AI model on the autonomous vehicle according to the evaluation result.

[0140] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0141] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in one or more flows in the flowchart and / or one or more blocks in the block diagram.

[0142] These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions specified in one or more flows in the flowchart and / or one or more blocks in the block diagram.

[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for realizing the functions specified in one or more flows in the flowchart and / or one or more blocks in the block diagram.

[0144] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0145] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0146] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0147] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0148] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0149] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A planning AI model online evaluation method, wherein: The method comprises: Deploy the planning AI model online on an autonomous driving vehicle, wherein the autonomous driving vehicle includes a rule-based planning model; Get scene data; Inputting the scenario data into the rule planning model and the planning AI model respectively; Evaluate the trajectory feasibility of the second output result of the planning AI model, and select the first output result of the rule planning model or the second output result of the planning AI model as the planned trajectory of the autonomous driving vehicle based on the evaluation result.

2. The method of claim 1, wherein: The evaluating the trajectory feasibility of the second output result of the planning AI model, and selecting the first output result of the rule planning model or the second output result of the planning AI model as the planned trajectory of the autonomous driving vehicle according to the evaluation result, includes: Based on a preset feedback mechanism, obtaining a first intention in a first output result of the rule planning model and a second intention in a second output result of the planning AI model; If it is determined that the first intention is inconsistent with the second intention, using the first output result to determine the feasibility of the AI ​​trajectory in the second output result; If it is determined that the first intention is consistent with the second intention, the trajectory similarity of the autonomous driving vehicle is evaluated.

3. The method of claim 2, wherein: The evaluating the trajectory similarity of the autonomous driving vehicle includes: In the case of having the same intention, obtaining the trajectory information in the second output result of the planning AI model and the trajectory information in the first output result of the rule planning model; Evaluate the similarity between the trajectory information in the second output result of the planning AI model and the trajectory information in the first output result of the rule planning model to determine the feasibility of the AI ​​trajectory; When it is determined that the AI ​​trajectory is feasible, the second output result of the planning AI model is used as the planned trajectory of the autonomous driving vehicle.

4. The method of claim 2, wherein: The using the first output result to judge the feasibility of the AI ​​trajectory in the second output result includes: In the case of different intentions, feeding back the second intention to the rule planning model to determine the feasibility of the AI ​​trajectory under the current decision output by the rule planning model; When it is determined that the AI ​​trajectory is not feasible, the first output result of the rule planning model is used as the planned trajectory of the autonomous driving vehicle.

5. The method of claim 2, wherein: The first output result of the rule planning model or the second output result of the planning AI model, as the planning trajectory of the autonomous driving vehicle, includes: Obtaining an output result of the rule planning model for the second intention and recording scene data at the current moment; Iterate the output result and the current scene data to the planning AI model; and / or, Saving the entry conditions and exit conditions of the scenario deployed on the autonomous driving vehicle after the planning AI model goes online; The entry conditions and exit conditions of the scenario deployed on the autonomous driving vehicle after the planning AI model goes online are used as trigger conditions for the rule planning model that replaces the planning AI model.

6. The method of claim 1, wherein: The acquiring of scene data comprises: In the training phase of the planning AI model, full data of manual driving is collected, wherein the full data at least includes scene data, and the scene data includes entry conditions and exit conditions of the scene; During the verification phase of the planning AI model, feature data of the target at each moment is collected, and the feature data includes data that can be used for the rule planning model.

7. The method of claim 1, wherein: The planning AI model includes: Collect scenario data for improving the user's riding experience and obtain a data set of preset scenarios; An AI model is obtained by training the data set according to the preset scenario; Set up a preset feedback mechanism; The planning AI model is obtained according to the planning AI model and the preset feedback mechanism.

8. A planning AI model online evaluation device, wherein: The device comprises: A deployment module, used to deploy the planning AI model online on an autonomous driving vehicle, wherein the autonomous driving vehicle includes a rule-based planning model; Acquisition module, used to acquire scene data; An input module, used to input the scenario data into the rule planning model and the planning AI model respectively; An evaluation module is used to evaluate the trajectory feasibility of the second output result of the planning AI model, and select the first output result of the rule planning model or the second output result of the planning AI model as the planned trajectory of the autonomous driving vehicle according to the evaluation result.

9. An electronic device, comprising: processor; as well as A memory arranged to store computer executable instructions, which when executed cause the processor to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, causes the electronic device to execute any one of the methods of claims 1 to 7.

Images