Decision-making method and device for autonomous vehicle and electronic equipment

Abstract

The invention provides a decision-making method and device for an automatic driving vehicle and electronic equipment. The method comprises the following steps: in a vehicle driving process, applying acquired perception information and vehicle state information to a target deep learning model, and acquiring a control parameter value of an initial control strategy; and applying the sensing information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy. The initial control strategy output by the target deep learning model is comprehensively checked and optimized through the hierarchical decision tree, instability caused by a direct linear weighted fusion mode is avoided, and the safety and stability of the obtained target control strategy are guaranteed.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and more specifically, to a decision-making method, device, medium and electronic device for an autonomous driving vehicle. Background Art

[0002] With the continuous development of autonomous driving technology, end-to-end large models are gradually emerging in the automotive industry. Compared with the traditional "perception-decision-planning-control" modular technical solution, the end-to-end large model has the characteristic of generating control output directly from sensor input. This feature gives it a significant advantage in information processing and decision-making speed. However, due to the inherent problems of machine learning technology itself, such as poor interpretability of output results, strong dependence on data, and large data demand, the application of machine learning models that directly output control instructions to vehicle chassis actuators in actual vehicles is accompanied by unexpected safety risks.

[0003] However, autonomous driving technology faces the dilemma of "no iteration and optimization without driving". On the one hand, machine learning models need to continuously accumulate data and experience in real driving environments in order to continuously improve and enhance their reliability and adaptability. On the other hand, companies and consumers are looking forward to the new generation of autonomous driving technology to bring a smoother and smarter vehicle travel experience. Therefore, how to reduce the safety risks brought by machine learning models in the process of driving has become an urgent problem to be solved.

[0004] Therefore, the present application provides a decision-making method for an autonomous driving vehicle to solve the above-mentioned technical problems. Summary of the invention

[0005] The purpose of this application is to provide a decision-making method, device, medium and electronic device for an autonomous driving vehicle, which can solve at least one of the technical problems mentioned above. The specific solution is as follows:

[0006] According to a specific implementation of the present application, in a first aspect, the present application provides a decision-making method for an autonomous driving vehicle, comprising:

[0007] Obtaining the perception information around the moving vehicle and the vehicle status information;

[0008] Applying the perception information and the vehicle state information to a target deep learning model to obtain control parameter values ​​of an initial control strategy;

[0009] The perception information, the vehicle state information and the control parameter values ​​of the initial control strategy are applied to a hierarchical decision tree to obtain a target control strategy.

[0010] Optionally, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy includes:

[0011] Performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule to obtain a hard rule check result;

[0012] When the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy to obtain a reliability check result;

[0013] When the reliability check result meets a preset low reliability condition, the perception information, the vehicle state information and the control parameter value of the initial control strategy are applied to a game model to obtain the target control strategy.

[0014] Optionally, when the reliability check result satisfies a preset low reliability condition, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a game model to obtain the target control strategy includes:

[0015] When the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy;

[0016] The benefit parameter value and the current rule weight value of the initial control strategy are applied to a dynamic adjustment model to obtain the target control strategy.

[0017] Optionally, when the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy includes:

[0018] When the reliability check result satisfies a preset low reliability condition, applying the vehicle state information and the control parameter value of the initial control strategy to a benefit measurement model of the target deep learning model to obtain a model benefit value;

[0019] The perception information, the vehicle state information, the preset traffic safety rules and the control parameter value of the initial control strategy are applied to a rule-constrained benefit model to obtain a rule-constrained benefit value.

[0020] Optionally, after applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy, the method further includes:

[0021] Performing a risk assessment on a control parameter value of the target control strategy based on the perception information to obtain a risk assessment result;

[0022] When the risk assessment result is danger information, the target control strategy is replaced in sequence according to the order of the initial control strategy, the preset rule control strategy and the preset braking strategy.

[0023] Optionally, the method further includes:

[0024] When the risk assessment result is safety information, in response to executing the target control strategy, obtaining feedback perception information and vehicle status information;

[0025] Based on the feedback perception information and vehicle state information, the execution effect of the target control strategy is evaluated to obtain an effect evaluation value;

[0026] When the effect evaluation value is greater than or equal to a preset good effect threshold, the training parameter values ​​of the target deep learning model and the performance parameter values ​​in the hierarchical decision tree are weighted based on the effect evaluation value.

[0027] Optionally, after performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule and obtaining a hard rule check result, the method further includes:

[0028] When the hard rule check result is abnormal, the target control strategy is determined to be a preset hard rule control strategy.

[0029] Optionally, when the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy, and after obtaining the reliability check result, further comprising:

[0030] When the reliability check result meets the preset high reliability condition, the target control strategy is determined to be the initial control strategy.

[0031] According to a specific implementation of the present application, in a second aspect, the present application provides a decision-making device for an autonomous driving vehicle, comprising:

[0032] An acquisition unit, used to acquire the perception information around the moving vehicle and the vehicle state information; an application unit, used to apply the perception information and the vehicle state information to the target deep learning model to obtain the control parameter value of the initial control strategy;

[0033] A decision unit is used to apply the perception information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy.

[0034] Optionally, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy includes:

[0035] Performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule to obtain a hard rule check result;

[0036] When the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy to obtain a reliability check result;

[0037] When the reliability check result meets a preset low reliability condition, the perception information, the vehicle state information and the control parameter value of the initial control strategy are applied to a game model to obtain the target control strategy.

[0038] Optionally, when the reliability check result satisfies a preset low reliability condition, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a game model to obtain the target control strategy includes:

[0039] When the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy;

[0040] The benefit parameter value and the current rule weight value of the initial control strategy are applied to a dynamic adjustment model to obtain the target control strategy.

[0041] Optionally, when the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy includes:

[0042] When the reliability check result satisfies a preset low reliability condition, applying the vehicle state information and the control parameter value of the initial control strategy to a benefit measurement model of the target deep learning model to obtain a model benefit value;

[0043] The perception information, the vehicle state information, the preset traffic safety rules and the control parameter value of the initial control strategy are applied to a rule-constrained benefit model to obtain a rule-constrained benefit value.

[0044] Optionally, after applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy, the method further includes:

[0045] Performing a risk assessment on a control parameter value of the target control strategy based on the perception information to obtain a risk assessment result;

[0046] When the risk assessment result is danger information, the target control strategy is replaced in sequence according to the order of the initial control strategy, the preset rule control strategy and the preset braking strategy.

[0047] Optionally, the device further comprises:

[0048] an execution unit, configured to obtain feedback perception information and vehicle status information in response to executing the target control strategy when the risk assessment result is safety information;

[0049] An evaluation unit, configured to evaluate the execution effect of the target control strategy based on the feedback perception information and vehicle state information to obtain an effect evaluation value;

[0050] An adjustment unit is used to adjust the weights of the training parameter values ​​of the target deep learning model and the performance parameter values ​​in the hierarchical decision tree based on the effect evaluation value when the effect evaluation value is greater than or equal to a preset good effect threshold.

[0051] Optionally, after performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule and obtaining a hard rule check result, the method further includes:

[0052] When the hard rule check result is abnormal, the target control strategy is determined to be a preset hard rule control strategy.

[0053] Optionally, when the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy, and after obtaining the reliability check result, further comprising:

[0054] When the reliability check result meets the preset high reliability condition, the target control strategy is determined to be the initial control strategy.

[0055] According to the specific implementation of the present application, in a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the decision-making method of an autonomous driving vehicle as described in any of the above items.

[0056] According to the specific implementation of the present application, in a fourth aspect, the present application provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the decision-making method of the autonomous driving vehicle as described in any of the above items.

[0057] Compared with the prior art, the above solution of the embodiment of the present application has at least the following beneficial effects:

[0058] The present application provides a decision-making method, device, medium and electronic device for an autonomous driving vehicle. During the driving process of the vehicle, the present application applies the acquired perception information and vehicle state information to the target deep learning model to obtain the control parameter values ​​of the initial control strategy; and applies the perception information, the vehicle state information and the control parameter values ​​of the initial control strategy to a hierarchical decision tree to obtain the target control strategy. The initial control strategy output by the target deep learning model is comprehensively checked and optimized through the hierarchical decision tree, avoiding the instability caused by the direct linear weighted fusion method and ensuring the safety and stability of the obtained target control strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A flowchart of a decision-making method for an autonomous driving vehicle according to an embodiment of the present application is shown;

[0060] Figure 2 A unit block diagram of a decision-making device for an autonomous driving vehicle according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0062] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two.

[0063] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0064] It should be understood that although the terms first, second, third, etc. may be used to describe in the embodiments of the present application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of the present application, the first may also be referred to as the second, and similarly, the second may also be referred to as the first.

[0065] As used herein, the words "if" and "if" may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0066] It should also be noted that the term "includes", "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, so that a 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 commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the commodity or device including the elements.

[0067] It should be particularly noted that any symbols and / or numbers in the specification that are not marked in the accompanying drawings are not drawing marks.

[0068] The optional embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0069] The embodiment provided in this application is an embodiment of a decision-making method for an autonomous driving vehicle.

[0070] Combine the following Figure 1 The embodiments of the present application are described in detail.

[0071] Step S101, acquiring the sensing information around the moving vehicle and the vehicle status information.

[0072] Perception information refers to the environmental information around the vehicle collected by the on-board camera, laser radar, sensor and positioning device, such as the positioning parameter values ​​of objects around the vehicle, the speed parameter values ​​of objects, and the road status parameter values.

[0073] The vehicle status information refers to the working status parameter value of the moving vehicle. For example, the vehicle status information includes: vehicle speed parameter value, vehicle acceleration parameter value and vehicle positioning parameter value.

[0074] Step S102: Apply the perception information and the vehicle state information to a target deep learning model to obtain control parameter values ​​of an initial control strategy.

[0075] For an embodiment of using the perception information and the vehicle status information to apply to a target deep learning model to obtain control parameter values ​​of an initial control strategy, please refer to the prior art and this application will not go into details.

[0076] Step S103, applying the perception information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy.

[0077] Different from the direct linear weighted fusion method in some embodiments, the embodiment of the present application provides a hierarchical decision tree fusion method. The target control strategy is determined by the hierarchical decision tree.

[0078] In some specific embodiments, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy includes:

[0079] Step S103 - 1 , performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule to obtain a hard rule check result.

[0080] Preset hard rules are common sense constraints on vehicle driving. For example, preset hard rules include: lane keeping rules, traffic light enforcement rules and collision avoidance rules.

[0081] In this specific embodiment, the control parameter values ​​of the initial control strategy output by the target deep learning model are checked to determine whether they conflict with preset hard rules.

[0082] Step S103 - 2a: when the hard rule check result is normal, a reliability check is performed on the control parameter value of the initial control strategy to obtain a reliability check result.

[0083] In this specific embodiment, if there is no conflict with the preset hard rules, a reliability check is performed on the control parameter values ​​of the initial control strategy.

[0084] In some specific embodiments, after performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule and obtaining a hard rule check result, the method further includes:

[0085] Step S103 - 2b: when the hard rule check result is abnormal, determining that the target control strategy is a preset hard rule control strategy.

[0086] If this specific embodiment conflicts with the preset hard rules, the preset hard rule control strategy will be used as the target control strategy to ensure that the safety strategy is directly adopted in situations such as red lights, lane keeping, collision risks, etc., to avoid model misjudgment and improve overall safety.

[0087] Step S103-3, when the reliability check result meets the preset low reliability condition, the perception information, the vehicle state information and the control parameter value of the initial control strategy are applied to the game model to obtain the target control strategy.

[0088] In this specific embodiment, if the reliability of the control parameter value of the initial control strategy is low, it means that the initial control strategy output by the target deep learning model is uncertain under the current circumstances, and the initial control strategy cannot be used as the target control strategy. In this case, a decision is made through a game model to ensure the safety and stability of the target control strategy.

[0089] In some specific embodiments, when the reliability check result satisfies a preset low reliability condition, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a game model to obtain the target control strategy includes:

[0090] Step S103-31, when the reliability check result meets the preset low reliability condition, the benefit parameter value is obtained based on the perception information, the vehicle state information and the control parameter value of the initial control strategy.

[0091] For example, the benefit parameter value includes the model benefit value and the rule constraint benefit value.

[0092] In some specific embodiments, when the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information, and the control parameter value of the initial control strategy includes:

[0093] Step S103-311, when the reliability check result meets the preset low reliability condition, the vehicle state information and the control parameter value of the initial control strategy are applied to the benefit measurement model of the target deep learning model to obtain the model benefit value.

[0094] Specifically, the following formulas are included:

[0095] U DL =β1×S+β2×R S ;

[0096] Among them, U DL represents the model return value, S represents the stability value, R S represents the reaction speed value, and β1 and β2 represent weight coefficients.

[0097] The stability value is a measure of the stability of the vehicle in the current operation, including the smoothness of the vehicle's acceleration changes and steering wheel angle changes. Higher stability means that the vehicle will reduce unnecessary sudden acceleration, sudden braking or sudden steering during driving, ensuring ride comfort and safety.

[0098] Specifically, the following formulas are included:

[0099] S=1-k1×σ a -k2×σ θ ;

[0100] Among them, S represents the stability value, k1 and k2 represent the adjustment coefficients, σ a Represents the standard deviation of the vehicle acceleration value within the preset window period, σ θ Indicates the standard deviation of the steering wheel angle change rate value within the preset window period. The vehicle state information includes the vehicle acceleration value and the steering wheel angle change rate value. The value range of S is between 0 and 1. The closer the value is to 1, the smoother the driving.

[0101] The reaction speed value is a measure of how quickly the target deep learning model responds to changes in the environment (such as sudden obstacles, lane line changes, etc.). A high reaction speed means that the model can quickly detect environmental changes and make corresponding decisions to avoid potential dangerous situations.

[0102] Specifically, the following formulas are included:

[0103]

[0104] Among them, RS represents the reaction speed value, Δt represents the reaction time, and ε is a constant used to prevent the denominator from being zero. After the vehicle senses the change in the environment (such as an obstacle in front), the delay time of the deceleration command or avoidance command in the initial control strategy. For example, after the obstacle in front is detected at time point t0, the model issues a response command at time point t1, then the reaction time Δt = t1-t0. s The higher the value of , the more timely the target deep learning model responds to environmental changes.

[0105] β1 and β2 are used to adjust the stability value S and the reaction speed value R s The relative importance of a factor in the model's return value.

[0106] In different driving scenarios (such as high-speed driving and urban congestion), the values ​​of β1 and β2 can be dynamically adjusted to achieve the optimal decision. For example, when driving at high speed, stability is more important, so the value of β1 is increased; on complex urban roads, quick response is more important, so the value of β2 is increased.

[0107] Step S103-312, applying the perception information, the vehicle state information, the preset traffic safety rules and the control parameter value of the initial control strategy to a rule-constrained benefit model to obtain a rule-constrained benefit value.

[0108] Specifically, the following formulas are included:

[0109] U R =δ×C-η×R c ;

[0110] Among them, U R represents the rule constraint benefit value, C represents the compliance value, R C represents the behavior conflict risk value, δ and η represent the weight coefficients.

[0111] The compliance value reflects the degree to which the vehicle's current driving behavior complies with traffic regulations, including whether it follows lane lines, maintains a safe distance, follows traffic signals, etc.

[0112] For example, vehicle status information includes vehicle positioning parameter values ​​and vehicle speed parameter values; perception information includes distance values ​​from the vehicle in front and traffic signal status values; preset traffic safety rules include: keeping lanes, maintaining a safe distance of at least 2 seconds from the vehicle in front, and stopping at red lights; the compliance conditions of each preset traffic safety rule are checked, with the number 1 indicating that the compliance conditions are met and the number 0 indicating that the compliance conditions are not met.

[0113] Specifically, the following formulas are included:

[0114]

[0115] Where C represents the compliance value, c i Indicates the state of the i-th compliance condition, and N indicates the total number of compliance conditions. The compliance value ranges from 0 to 1. The higher the compliance value, the more the vehicle's current behavior complies with traffic regulations.

[0116] The behavior conflict risk value measures the degree of conflict between the initial control strategy and the rule constraints, and checks whether there is a contradiction between the initial control strategy and the rule constraints. For example, the initial control strategy recommends accelerating to change lanes, while the rule requires decelerating to keep the lane. This conflict will increase driving uncertainty and risk.

[0117] Specifically, the following formulas are included:

[0118]

[0119] Among them, R C represents the behavior conflict risk value, r jrepresents the conflict state of the jth control parameter value of the initial control strategy, and M represents the number of control parameter values ​​of the initial control strategy. If there is a conflict, set r j =1; if there is no conflict, set r j = 0. The higher the conflict risk value, the more serious the conflict.

[0120] δ is used to adjust the importance of the compliance value C in the rule constraint benefit. Usually, δ is set to a higher value to ensure that the vehicle strictly abides by the traffic rules.

[0121] η is used to adjust the behavior conflict risk value R C Usually, η also takes a higher value to reduce the potential risk in driving behavior.

[0122] In this specific embodiment, step S103 - 311 and step S103 - 312 are executed in either order.

[0123] Step S103-32, applying the benefit parameter value and the current rule weight value of the initial control strategy to the dynamic adjustment model to obtain the target control strategy.

[0124] In this specific embodiment, the weighted fusion result between the benefit parameter value and the current rule weight value of the initial control strategy is used as the target control strategy. The benefit parameter value is used to give priority to operations that comply with traffic rules, ensuring the compliance and safety of the vehicle in various scenarios.

[0125] In some specific embodiments, when the hard rule check result is normal, the control parameter value of the initial control strategy is subjected to a reliability check. After obtaining the reliability check result, it also includes: when the reliability check result meets a preset higher reliability condition, determining that the target control strategy is the initial control strategy.

[0126] In this specific embodiment, if the reliability of the control parameter value of the initial control strategy is high, it means that the target deep learning model has a stronger decision-making ability under the current situation, and the initial control strategy output by the target deep learning model is directly adopted as the target control strategy.

[0127] In some specific embodiments, after applying the perception information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy, the method further includes:

[0128] Step S104 - 1 , performing risk assessment on the control parameter value of the target control strategy based on the perception information to obtain a risk assessment result.

[0129] In this specific embodiment, after obtaining the target control strategy, a risk assessment is performed on the target control strategy to determine whether there is a potential collision risk or other dangerous situation. For example, the distance between the vehicle and the preceding vehicle, the distance between the vehicle and the obstacle, and the vehicle speed are assessed.

[0130] Step S104-2a, when the risk assessment result is danger information, the target control strategy is replaced in sequence according to the order of the initial control strategy, the preset rule control strategy and the preset braking strategy.

[0131] In this specific embodiment, if it is detected that the target control strategy has a risk, the decision is downgraded and the following priorities are adopted in descending order: initial control strategy -> preset rule control strategy -> preset braking strategy.

[0132] If all strategies fail to eliminate the safety risk, safety braking is triggered to ensure that the vehicle slows down or stops immediately.

[0133] In some specific embodiments, the method further comprises:

[0134] Step S104-2b-1, when the risk assessment result is safety information, in response to executing the target control strategy, obtaining feedback perception information and vehicle status information.

[0135] In this specific embodiment, if it is detected that the target control strategy does not have risks, the target control strategy is executed, and in the process of executing the target control strategy, the feedback perception information and vehicle status information are obtained.

[0136] Step S104-2b-2: Evaluate the execution effect of the target control strategy based on the feedback perception information and vehicle status information to obtain an effect evaluation value.

[0137] Step S104-2b-3, when the effect evaluation value is greater than or equal to a preset good effect threshold, the training parameter values ​​of the target deep learning model and the performance parameter values ​​in the hierarchical decision tree are weighted based on the effect evaluation value.

[0138] In this specific embodiment, when executing the target control strategy, feedback information of the current scene is collected, the effect of the current decision is calculated based on the feedback information, and the training parameter values ​​of the target deep learning model and the weight values ​​of the performance parameter values ​​in the hierarchical decision tree are adjusted according to the effect, so that they are more adaptable in subsequent similar scenes, thereby achieving information feedback, automatically optimizing the target deep learning model and the hierarchical decision tree, and improving the reliability of the target deep learning model and the hierarchical decision tree.

[0139] In the embodiment of the present application, during the driving process of the vehicle, the acquired perception information and vehicle state information are applied to the target deep learning model to obtain the control parameter values ​​of the initial control strategy; the perception information, the vehicle state information and the control parameter values ​​of the initial control strategy are applied to the hierarchical decision tree to obtain the target control strategy. The initial control strategy output by the target deep learning model is comprehensively checked and optimized through the hierarchical decision tree, avoiding the instability caused by the direct linear weighted fusion method and ensuring the safety and stability of the obtained target control strategy.

[0140] The present application also provides a device embodiment that is based on the above embodiment, which is used to implement the method steps described in the above embodiment. The explanation based on the same name meaning is the same as the above embodiment, and has the same technical effect as the above embodiment, which will not be repeated here.

[0141] like Figure 2 As shown, the present application provides a decision-making device 200 for an autonomous driving vehicle, comprising:

[0142] An acquisition unit 201 is used to acquire the sensing information around the moving vehicle and the vehicle status information;

[0143] An application unit 202 is used to apply the perception information and the vehicle state information to a target deep learning model to obtain a control parameter value of an initial control strategy;

[0144] The decision unit 203 is used to apply the perception information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy.

[0145] Optionally, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy includes:

[0146] Performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule to obtain a hard rule check result;

[0147] When the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy to obtain a reliability check result;

[0148] When the reliability check result meets a preset low reliability condition, the perception information, the vehicle state information and the control parameter value of the initial control strategy are applied to a game model to obtain the target control strategy.

[0149] Optionally, when the reliability check result satisfies a preset low reliability condition, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a game model to obtain the target control strategy includes:

[0150] When the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy;

[0151] The benefit parameter value and the current rule weight value of the initial control strategy are applied to a dynamic adjustment model to obtain the target control strategy.

[0152] Optionally, when the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy includes:

[0153] When the reliability check result satisfies a preset low reliability condition, applying the vehicle state information and the control parameter value of the initial control strategy to a benefit measurement model of the target deep learning model to obtain a model benefit value;

[0154] The perception information, the vehicle state information, the preset traffic safety rules and the control parameter value of the initial control strategy are applied to a rule-constrained benefit model to obtain a rule-constrained benefit value.

[0155] Optionally, after applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy, the method further includes:

[0156] Performing a risk assessment on a control parameter value of the target control strategy based on the perception information to obtain a risk assessment result;

[0157] When the risk assessment result is danger information, the target control strategy is replaced in sequence according to the order of the initial control strategy, the preset rule control strategy and the preset braking strategy.

[0158] Optionally, the device further comprises:

[0159] an execution unit, configured to obtain feedback perception information and vehicle status information in response to executing the target control strategy when the risk assessment result is safety information;

[0160] An evaluation unit, configured to evaluate the execution effect of the target control strategy based on the feedback perception information and vehicle state information to obtain an effect evaluation value;

[0161] An adjustment unit is used to adjust the weights of the training parameter values ​​of the target deep learning model and the performance parameter values ​​in the hierarchical decision tree based on the effect evaluation value when the effect evaluation value is greater than or equal to a preset good effect threshold.

[0162] Optionally, after performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule and obtaining a hard rule check result, the method further includes:

[0163] When the hard rule check result is abnormal, the target control strategy is determined to be a preset hard rule control strategy.

[0164] Optionally, when the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy, and after obtaining the reliability check result, further comprising:

[0165] When the reliability check result meets the preset high reliability condition, the target control strategy is determined to be the initial control strategy.

[0166] In the embodiment of the present application, during the driving process of the vehicle, the acquired perception information and vehicle state information are applied to the target deep learning model to obtain the control parameter values ​​of the initial control strategy; the perception information, the vehicle state information and the control parameter values ​​of the initial control strategy are applied to the hierarchical decision tree to obtain the target control strategy. The initial control strategy output by the target deep learning model is comprehensively checked and optimized through the hierarchical decision tree, avoiding the instability caused by the direct linear weighted fusion method and ensuring the safety and stability of the obtained target control strategy.

[0167] Example 3

[0168] This embodiment provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method steps described in the above embodiment.

[0169] Example 4

[0170] An embodiment of the present application provides a non-volatile computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the method steps described in the above embodiment.

[0171] Finally, it should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the system or device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0172] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. 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 decision-making method for an autonomous driving vehicle, characterized in that: include: Obtaining the perception information around the moving vehicle and the vehicle status information; Applying the perception information and the vehicle state information to a target deep learning model to obtain control parameter values ​​of an initial control strategy; The perception information, the vehicle state information and the control parameter values ​​of the initial control strategy are applied to a hierarchical decision tree to obtain a target control strategy.

2. The method according to claim 1, characterized in that: The step of applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy includes: Performing a hard rule check on the control parameter value of the initial control strategy based on a preset hard rule to obtain a hard rule check result; When the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy to obtain a reliability check result; When the reliability check result meets a preset low reliability condition, the perception information, the vehicle state information and the control parameter value of the initial control strategy are applied to a game model to obtain the target control strategy.

3. The method according to claim 2, characterized in that When the reliability check result satisfies a preset low reliability condition, applying the perception information, the vehicle state information, and the control parameter value of the initial control strategy to a game model to obtain the target control strategy, including: When the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information and a control parameter value of the initial control strategy; The benefit parameter value and the current rule weight value of the initial control strategy are applied to a dynamic adjustment model to obtain the target control strategy.

4. The method according to claim 3, characterized in that When the reliability check result satisfies a preset low reliability condition, obtaining a benefit parameter value based on the perception information, the vehicle state information, and the control parameter value of the initial control strategy includes: When the reliability check result satisfies a preset low reliability condition, applying the vehicle state information and the control parameter value of the initial control strategy to a benefit measurement model of the target deep learning model to obtain a model benefit value; The perception information, the vehicle state information, the preset traffic safety rules and the control parameter value of the initial control strategy are applied to a rule-constrained benefit model to obtain a rule-constrained benefit value.

5. The method according to claim 1, characterized in that After applying the perception information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy, the method further includes: Performing a risk assessment on a control parameter value of the target control strategy based on the perception information to obtain a risk assessment result; When the risk assessment result is danger information, the target control strategy is replaced in sequence according to the order of the initial control strategy, the preset rule control strategy and the preset braking strategy.

6. The method according to claim 5, characterized in that The method further comprises: When the risk assessment result is safety information, in response to executing the target control strategy, obtaining feedback perception information and vehicle status information; Based on the feedback perception information and vehicle state information, the execution effect of the target control strategy is evaluated to obtain an effect evaluation value; When the effect evaluation value is greater than or equal to a preset good effect threshold, the training parameter values ​​of the target deep learning model and the performance parameter values ​​in the hierarchical decision tree are weighted based on the effect evaluation value.

7. The method according to claim 2, characterized in that: After performing a hard rule check on the control parameter value of the initial control strategy based on the preset hard rule and obtaining the hard rule check result, the method further includes: When the hard rule check result is abnormal, the target control strategy is determined to be a preset hard rule control strategy.

8. The method according to claim 2, characterized in that: When the hard rule check result is normal, performing a reliability check on the control parameter value of the initial control strategy, and after obtaining the reliability check result, the method further includes: When the reliability check result meets the preset high reliability condition, the target control strategy is determined to be the initial control strategy.

9. A decision-making device for an autonomous driving vehicle, characterized in that: include: An acquisition unit, used to acquire sensing information around the moving vehicle and vehicle status information; An application unit, configured to apply the perception information and the vehicle state information to a target deep learning model to obtain control parameter values ​​of an initial control strategy; A decision unit is used to apply the perception information, the vehicle state information and the control parameter value of the initial control strategy to a hierarchical decision tree to obtain a target control strategy.

10. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 8.

Images