Human-like decision making method, and human-like decision model construction method and system

Abstract

The application discloses a kind of method and system for constructing intelligent automobile class human decision model, comprising: analysis host car in current traffic environment's cognitive task, determine the key elements of host car in traffic scene and cognitive link is relevant, based on this, analysis the representation method of multi-agent safety between host car and traffic environment, construct class human understanding model;Analysis long-term safety risk represented by key elements, and according to long-term safety risk evaluation result and destination planning result, the future space-time range of traffic situation is predicted, and heuristic pre-judgment model is constructed, further combined with class human understanding model, form class human understanding and pre-judgment decision model;According to the optimization evaluation parameter except multi-agent safety, the decision result obtained by class human understanding and pre-judgment decision model is optimized.The application can make decision result keep logical continuity in space-time scale, and improve the explainability and scalability of decision model by reasonable connection between models.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent automobile application, in particular to a human-like decision-making method for intelligent automobile, and a method and system for constructing a human-like decision-making model of intelligent automobile. BACKGROUND

[0002] Before the large-scale application of autonomous vehicles, intelligent vehicles and human-driven vehicles will share the traffic road for a long time. As a new type of traffic participant, the traffic behavior of intelligent vehicles needs to be predictable and interpretable to avoid new driving accidents, which puts higher requirements on the situation awareness ability of intelligent vehicles.

[0003] In terms of improving the predictability and interpretability of intelligent vehicles, researchers have proposed various human-like decision-making methods, mainly including the following categories: establishing a human driving trajectory library, matching the current trajectory to be selected according to the scene, so that the intelligent vehicle has a human-like driving trajectory; using NDD to calibrate the target function for decision-making, so that the AV decision-making reflects the human driving characteristics; referring to the game theory to formulate the decision-making rules and corresponding thresholds offline to obtain a decision-making method with interactive characteristics. The above human-like decision-making, on the one hand, only focuses on the human-like degree of the decision-making result, and since the database, target function calibration, and game rules and other information are all set offline, when applied to a time-continuous scene, only discrete judgment results can be executed, resulting in a jump in the decision-making result at some critical values, thereby making the output of the decision-making result lack of maneuverability, flexibility, and poor scalability. SUMMARY

[0004] The purpose of the present application is to provide a model construction scheme for human-like decision-making to solve the problems in the background art through the application of the constructed model.

[0005] To solve the above technical problems, the embodiment of the present application provides a method for constructing a human-like decision-making model of intelligent vehicle, comprising: analyzing the cognitive task of the host vehicle in the current traffic environment to determine the key elements related to the cognitive link of the host vehicle in the traffic scene, based on which, analyzing the representation method of multi-agent safety between the host vehicle and the traffic environment, and constructing a human-like understanding model; analyzing the long-term safety risk represented by the key elements, and predicting the future space-time range of the traffic situation according to the long-term safety risk evaluation result and the destination planning result, constructing a heuristic prediction model, and further combining the human-like understanding model to form a human-like understanding and prediction decision-making model; optimizing the decision-making result obtained by the human-like understanding and prediction decision-making model according to the optimization evaluation parameters other than the multi-agent safety.

[0006] Preferably, in the process of constructing the heuristic pre-judgment model, the following steps are included: determining heuristic parameters capable of reflecting long-term driving safety trends according to the key elements and the description of the traffic situation, and analyzing potential driving risks under the future development trend of the heuristic parameters to establish a long-term safety evaluation model for the heuristic parameters; and evaluating the continuous time range and spatial range of safe driving according to the destination planning result and the long-term safety evaluation model, and establishing the heuristic pre-judgment model.

[0007] Preferably, the long-term safety evaluation model is established by analyzing potential driving risks with the destination planning result as a constraint condition, and a judgment result of multi-agent safety in a certain spatio-temporal range is generated, so as to generate the heuristic pre-judgment model, and a decision result is obtained by applying the heuristic pre-judgment model in the traffic scene according to the multi-agent safety output by the model.

[0008] Preferably, in the step of optimizing the decision result obtained by the human-like understanding and pre-judgment decision model according to the optimization evaluation parameters other than multi-agent safety, the following steps are included: evaluating the performance level of the driving decision result output by the human-like understanding and pre-judgment decision model on the optimization evaluation parameters according to the optimization evaluation parameters and the expected level of the corresponding evaluation parameters; and optimizing the decision result output by the human-like understanding and pre-judgment decision model based on the optimization parameter performance level.

[0009] Preferably, in the step of evaluating the influence of the driving decision result output by the human-like understanding and pre-judgment decision model on the optimization evaluation parameters according to the optimization evaluation parameters and the expected level of the corresponding evaluation parameters, the following steps are included: analyzing the influence of the driving decision result on each of the optimization evaluation parameters, generating a reward and punishment function for each of the optimization evaluation parameters according to the expected level, and the optimization evaluation parameters include but are not limited to driving efficiency, prosociality and driving comfort; and quantitatively evaluating the influence of the driving decision result on each of the optimization evaluation parameters according to the reward and punishment function.

[0010] Preferably, in the step of determining the key elements related to the cognitive link of the host vehicle in the traffic scene, the following steps are included: establishing a typical traffic scene model representing the current driving environment state of the host vehicle according to the perception information; analyzing the cognitive tasks that need to be completed by the host vehicle when the kinematics and dynamics changes, traffic rules and road conditions between the host vehicle and multi-agents have an impact on multi-agent safety based on the interaction of the host vehicle in the typical traffic scene model; and determining the key elements in the traffic scene according to the analysis result of the cognitive tasks.

[0011] Preferably, the method further includes simulating and verifying the reliability and validity of the human-like understanding model by setting a test scene.

[0012] Preferably, a plurality of test scenes under a series of combined conditions of different numbers of side cars, different combinations of main car motion states and different side car motion states, different road conditions and different traffic rules are set.

[0013] In another aspect, the embodiment of the present application provides a human-like decision-making method for intelligent vehicles, which applies the human-like understanding and pre-judgment decision-making model constructed as described above.

[0014] In addition, the embodiment of the present application also provides a system for constructing a human-like decision-making model for intelligent vehicles, comprising: a human-like understanding model construction module configured to analyze the cognitive tasks of a host vehicle in a current traffic environment to determine the key elements related to cognitive links of the host vehicle in a traffic scene, based on which, a representation method of multi-agent safety between the host vehicle and the traffic environment is analyzed, and a human-like understanding model is constructed; a human-like understanding and pre-judgment decision-making model construction module configured to analyze long-term safety risks represented by the key elements, and according to the long-term safety risk evaluation results and the destination planning results, to predict the future space-time range of the traffic situation, construct a heuristic pre-judgment model, and further combine the human-like understanding model to form a human-like understanding and pre-judgment decision-making model; and a decision result optimization module configured to optimize the decision results obtained by the human-like understanding and pre-judgment decision-making model according to optimization evaluation parameters other than multi-agent safety.

[0015] Compared with the prior art, one or more embodiments in the above scheme can have the following advantages or beneficial effects:

[0016] The present application proposes a method and system for constructing a human-like decision-making model for intelligent vehicles, and a human-like decision-making method applying such a human-like decision-making model. Specifically, by establishing a human-like understanding model, a heuristic pre-judgment model and a decision result evaluation model, a human-like understanding-pre-judgment decision-making model is obtained, which improves the understanding level of intelligent vehicles for traffic situations, avoids the occurrence of driving behaviors that human traffic participants cannot react to in time and correctly, and ultimately improves the predictability of intelligent vehicles. On the basis of ensuring multi-agent safety, an extended space for further optimizing decisions is given, which is conducive to the integration of intelligent vehicles into the traffic system with higher social acceptance.

[0017] In summary, the present application has the following advantages and positive effects:

[0018] (1) Referring to the human decision-making process, the automatic driving decision-making is divided into three processes of traffic situation understanding, traffic situation pre-judgment and decision result optimization, which makes the decision-making have temporal and spatial continuity, and avoids traffic accidents such as rear-end collisions caused by sudden changes in decision results;

[0019] (2) The establishment of a human-like understanding model enables intelligent vehicles to perceive safety risks in a multi-agent traffic environment, and compared with two-by-two game decision analysis, it can ensure multi-agent safety in more complex traffic scenarios;

[0020] (3) The establishment of a heuristic pre-judgment model gives intelligent vehicles the ability to perceive long-term safety, avoiding difficult-to-predict behaviors such as emergency braking from the source;

[0021] (4) By constructing a human-like understanding-prediction decision model, the transition mechanism between discrete decision results is revealed, making the basis followed by each decision result remain spatially and temporally continuous and understandable to other human traffic participants;

[0022] (5) Considering the decision performance of a higher level of intelligence, a decision result evaluation model is established to provide an extension interface for optimizing decision strategies to achieve a higher level of intelligent decision-making.

[0023] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0024] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate embodiments of the present application, and are used to explain the present application together with the written description. The drawings are not intended to limit the present application, and in the drawings:

[0025] Figure 1 A step diagram of the method for constructing the human-like decision model of the intelligent vehicle according to the embodiment of the present application.

[0026] Figure 2 A principle schematic diagram of the method for constructing the human-like decision model of the intelligent vehicle according to the embodiment of the present application.

[0027] Figure 3 An example diagram of a typical traffic scene model in the method for constructing the human-like decision model of the intelligent vehicle according to the embodiment of the present application.

[0028] Figure 4 A module block diagram of the system for constructing the human-like decision model of the intelligent vehicle according to the embodiment of the present application. DETAILED DESCRIPTION

[0029] The embodiments of the present application will be described in detail below with reference to the accompanying drawings and embodiments, so that how the present application applies technical means to solve technical problems and achieves technical effects can be fully understood and implemented. It should be noted that, as long as there is no conflict, each embodiment in the present application and each feature in each embodiment can be combined with each other, and the formed technical solutions are within the protection scope of the present application.

[0030] In addition, the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions. Moreover, although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0031] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the example embodiments. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0032] Before the large-scale application of autonomous vehicles, intelligent vehicles and human-driven vehicles will share the traffic road for a long time. As a new type of traffic participant, the traffic behavior of intelligent vehicles needs to be predictable and interpretable to avoid the occurrence of new driving accidents, which puts higher requirements on the situation awareness ability of intelligent vehicles.

[0033] In order to improve the predictability and interpretability of intelligent vehicles, researchers have proposed various human-like decision-making methods, mainly including the following categories: establishing a human driving trajectory library, matching the current trajectory to be selected according to the scene, so that the intelligent vehicle has a human-like driving trajectory; using NDD to calibrate the target function for decision-making, so that the AV decision-making reflects the human-like driving characteristics; referring to the game theory to formulate the decision-making rules and corresponding thresholds offline to obtain a decision-making method with interactive characteristics. The above human-like decision-making, on the one hand, only focuses on the human-like degree of the decision-making result, and since the database, target function calibration, and game rules and other information are all set offline, when applied to a time-continuous scene, only discrete judgment results can be executed, resulting in a jump in the decision-making result at some critical values, and thus the output of the decision-making result lacks maneuverability, flexibility, and poor scalability.

[0034] To address the problems existing in the prior art, this application proposes a method and system for constructing a human-like decision-making model for intelligent vehicles. This method and system incorporates situational understanding and prediction processes between real-time perception information and decision output, ensuring logical continuity of decision results across time and space. Furthermore, the reasonable connection between models enhances the interpretability and scalability of the decision-making model.

[0035] Figure 1 This diagram illustrates the steps of a method for constructing a human-like decision-making model for an intelligent vehicle, as described in an embodiment of this application. Figure 1 As shown in the embodiment of the present invention, the method for constructing a human-like decision-making model for intelligent vehicles (hereinafter referred to as the "human-like decision-making model construction method") includes the following steps: Step S110: Based on the cognitive task analysis of the master vehicle (current intelligent vehicle) in the current traffic environment, determine the key elements related to the cognitive link of the master vehicle in the traffic scenario, and based on the key element information, analyze the traffic situation representing the multi-agent safety between the master vehicle and the traffic environment (including but not limited to: other vehicles, roads, traffic regulations, etc.), and construct a human-like understanding model; Step S120: Analyze the long-term safety risks represented by the key elements obtained in Step S110, and based on the long-term safety risk assessment results and destination planning results, predict the future spatiotemporal range of the traffic situation, construct a heuristic prediction model, and further combine it with the human-like understanding model constructed in Step S110 to form a human-like understanding and prediction decision-making model; Finally, Step S130: Optimize the decision results obtained by the human-like understanding and prediction decision-making model based on optimization evaluation parameters other than multi-agent safety.

[0036] It should be noted that, in this embodiment of the invention, since the human decision-making process includes cognition, understanding, prediction, decision-making and evaluation, the cognition mentioned in step S110 refers to a link in the decision-making process, and the key elements identified refer to the information input to the "understanding" link.

[0037] Figure 2 This is a schematic diagram illustrating the principle of a method for constructing a human-like decision-making model for intelligent vehicles, as described in an embodiment of this application. The following is in conjunction with... Figure 1 and Figure 2 The method for constructing a human-like decision-making model according to the embodiments of the present invention will be described in detail below.

[0038] In step S110, it is first necessary to analyze the influencing factors of multi-agent safety in the traffic scenario based on the perception information, so as to determine the key elements of the current intelligent vehicle (master vehicle) in the traffic scenario.

[0039] In one embodiment, a typical traffic scenario model representing the driving environment state of the current ego vehicle is established according to the perception information representing the driving environment state of the current ego vehicle detected by the current ego vehicle; then, the kinematic and dynamic changes between the ego vehicle and the background vehicles, the influence of the traffic environment such as the road traffic regulations and the road conditions on the safety of the multi-agent, and the like are analyzed based on the driving state of the ego vehicle in the current typical traffic scenario model, so as to determine the cognitive task to be completed by the ego vehicle; and the key elements in the traffic scenario are determined according to the analysis result of the current cognitive task.

[0040] Specifically, the perception information of the surrounding driving environment state detected by the current ego vehicle is acquired, wherein the environment state can include the kinematic and dynamic information of the ego vehicle, the driving position of the ego vehicle, the number of the background vehicles, the kinematic and dynamic information such as the driving speed, the position, the acceleration, and the direction of the background vehicles, and the traffic regulations and the road conditions of the driving area. Figure 3 An example diagram of the typical traffic scenario model in the method for constructing the human-like decision model of the intelligent vehicle according to the embodiments of the present application is shown. Figure 3 The diagram shows the typical traffic scenario model constructed according to the perception information, in which the position where the ego vehicle (EV) is currently driving is shown, i.e., the three-lane in the same direction, and there are six background vehicles (BVs) around the ego vehicle which can affect the ego vehicle. Then, the cognitive task analysis of the ego vehicle needs to be performed in this scenario model, and the key elements are determined according to the analysis result.

[0041] The cognitive task analysis of the ego vehicle refers to the cognitive task to be completed by the ego vehicle when the influence of the kinematic and dynamic changes between the ego vehicle and the background vehicles in the lateral and longitudinal directions on the driving safety and the environmental factors such as the road traffic regulations and the road conditions are considered, wherein an example of the cognitive analysis result is shown in the diagram, i.e., the deceleration of BV1 will compress the safe driving space of the ego vehicle. Figure 3 The key elements to be further understood by the ego vehicle in the traffic scenario can be determined according to the cognitive task analysis result, wherein the key elements include but are not limited to the position, the speed, the acceleration, and the driving direction of each background vehicle.

[0042] Therefore, the establishment of the typical traffic scenario model is completed, and the key elements to be further understood by the ego vehicle in the traffic scenario, i.e., the input of the human-like comprehension model (HCM), are determined.

[0043] Further, after the key elements of the traffic scenario are determined, the human-like comprehension model is constructed according to the determined key elements of the traffic scenario and the traffic situation.

[0044] Further, according to the key elements of the traffic scene that the host vehicle needs to further understand, the embodiment of the present application performs analysis on the cognitive tasks existing between the host vehicle and the traffic environment, specifically analyzes the multi-agent safety impact that may be generated when the host vehicle and the side vehicle change in the horizontal and longitudinal kinematics, and thus generates an HCM model for describing the traffic situation including the driving risk and the driving conflict risk. The HCM model is a model (for example, description is realized by functions, expressions, etc.) that describes the traffic situation faced by the current host vehicle by taking the key elements of the traffic scene as input, and the output of the human-like understanding model is the traffic situation understanding result. The human-like understanding model can reflect the ability of the host vehicle to perceive the traffic situation, and is used to realize the understanding of the traffic situation including any number of side vehicles around the current host vehicle, the side vehicles in any motion state, the regulations of the traffic area where the host vehicle is located, road signs, etc. After the perception information is input and the key elements are determined, the corresponding traffic situation understanding result can be output.

[0045] As shown in Figure 3 , the embodiment of the present application can analyze the cognitive tasks existing between the EV and BV1-BV6 and the traffic impact factors such as traffic rules, and thus construct an HCM model.

[0046] In addition, in order to guarantee the accuracy of the currently constructed HCM model, the step S110 also verifies the currently constructed human-like understanding model. Further, a test scene is set to simulate and verify the reliability and validity of the human-like understanding model.

[0047] Specifically, when setting the test scene, the embodiment of the present application sets a plurality of test scenes under a series of combination conditions such as different numbers of side vehicles, different combinations of host vehicle motion states and side vehicle motion states, different road conditions, and different combinations of traffic rules, and thus simulates and verifies the reliability and validity of the human-like understanding model through the plurality of test scenes. In the case of insufficient reliability and / or validity, the key elements of the traffic scene and / or the description of the traffic situation understanding can be adjusted.

[0048] After obtaining the verified human-like understanding model, the step S120 is entered.

[0049] In the step S120, the embodiment of the present application first performs long-term safety risk analysis on the key elements obtained in the step S110, and then performs prediction based on the future space-time range on the transient traffic situation understanding result obtained in the step S110 according to the destination planning result and the long-term safety risk evaluation result, and thus constructs a heuristic pre-judgment model.

[0050] Further, according to the description of the key elements and the transient traffic situation understanding, the heuristic parameters capable of reflecting long-term safety trends are determined, and the potential driving risks of the current heuristic parameters under future development trends are analyzed, thereby establishing a long-term safety evaluation model for the heuristic parameters; then, according to the destination planning result and the long-term safety evaluation model for each heuristic parameter, the continuous time range and the spatial range of safe driving are evaluated, and a heuristic prediction model is established.

[0051] Specifically, according to the description of the potential driving risks (such as including potential driving risks and potential driving conflict risks) by the optimized human-like understanding model, the heuristic parameters (including but not limited to: the jerk of the host vehicle and each side vehicle, the speed of the host vehicle and each side vehicle, the driving direction of the host vehicle and each side vehicle, etc.) capable of reflecting long-term safety trends are selected from the key elements, and the potential driving risk safety analysis is carried out on the development trend of the current heuristic parameters (in an embodiment, for each heuristic parameter, the influence of each piece of prediction data on the safety of the host vehicle driving is specifically analyzed, and the unsafe data segment that has a potential dangerous influence on the safe driving of the host vehicle and the safe data segment that has no potential dangerous influence on the safe driving of the host vehicle are screened out), thereby establishing a long-term safety evaluation model for the heuristic parameters, and completing the future space-time prediction of the traffic situation.

[0052] In the embodiment of the present application, the long-term safety evaluation model can describe the unsafe data segment and the safe data segment in the continuous development trend of each heuristic parameter.

[0053] Next, according to the destination planning result (for example: the current vehicle needs to drive to destination A) and the long-term safety evaluation model, the continuous time range and the spatial range of safe driving are evaluated, and a heuristic prediction model is established. In an embodiment, the destination planning result is taken as a constraint condition, the potential driving risks are analyzed, the long-term safety evaluation model is established, and the judgment result of the safety of multiple agents in a certain space-time range is generated, thereby generating the heuristic prediction model. The corresponding driving decision result is obtained by applying the current heuristic prediction model in the traffic scene according to the safety of multiple agents in a certain space-time range output by the heuristic prediction model.

[0054] In an embodiment, the destination planning result can be a navigation result according to a to-be-completed driving task. For example: the current road section needs to turn right at the front 100m, and it is required to change to the rightmost lane in the area where lane changing is allowed to complete the driving task, which makes an upper limit to the decision of the host vehicle.

[0055] In one embodiment, the time and space range of safe driving can be, for example: taking the current state of the host vehicle and the traffic environment as initial variables, assuming that the adjacent side vehicle driving on the rightmost lane will continue to drive at the current speed in front of the right front of the host vehicle at this time according to the current traffic situation, and the host vehicle will collide at the (x, y) position in the future s seconds when changing lanes at the current speed, then the safe time range is from the current time to the s seconds, and the safe space range is the space enveloped by the current position to (x, y). It is noted that the safe time range can be adjusted according to actual application requirements, for example, a safety margin m is added, and it is considered that the safe time range is from the current time to the s-m seconds.

[0056] Further, the heuristic prediction model aims to output a future space-time range that can guarantee the safety of the host vehicle multi-agent, and provide a reference for obtaining the final decision result. In the embodiment of the present application, the destination planning result limits the host vehicle to complete the lane changing behavior to the rightmost lane within a certain space range, and the continuous space-time range of driving safety limits the timing of lane changing and the selection of driving strategies such as speed and acceleration. The destination planning result and the safe space-time range are combined to limit the future driving safety in space-time range, thereby obtaining the heuristic prediction model.

[0057] Further, in the embodiment of the present application, the heuristic prediction model is a model (for example: realized by functions, expressions, etc.) that describes the safe driving space-time range of the current host vehicle in the face of the current traffic situation by taking the traffic situation understanding result as input, wherein the output of the heuristic prediction model is the time range and the space range of safe driving. The heuristic prediction model can reflect the perception ability of the host vehicle to the traffic situation, and is used to obtain a decision result with time-space continuity and higher situation perception level.

[0058] After obtaining the heuristic prediction model, step S120 also associates the (verified) human-like understanding model generated in step S110 and the currently generated heuristic prediction model to obtain a human-like understanding and prediction decision model of associated time-space safety range.

[0059] Meanwhile, the embodiment of the present application can also directly obtain the corresponding driving strategy result using the human-like understanding and prediction decision model according to the key elements of the traffic scene obtained from the perception information detected by the current host vehicle.

[0060] In the embodiment of the present application, the human-like understanding and pre-judgment decision model takes the key elements of the traffic scene as input, considers the transient safety of multi-agent traffic situation understanding and the decision-making of coping with long-term safety risks of multi-agent, so as to take the driving space-time range ensuring the safety of multi-agent as the output of the model, and according to the space-time limit of safe driving as a constraint, the safe driving decision result (for example, right lane changing operation with a certain range of acceleration at the current speed can ensure the safety of multi-agent) can be directly obtained after the current human-like understanding and pre-judgment decision model is applied to the test traffic scene or the actual driving traffic scene.

[0061] Further, after obtaining the human-like understanding and pre-judgment decision model, the embodiment of the present application enters step S130. In step S130, the decision result obtained by the human-like understanding and pre-judgment decision model in step S120 is optimized according to the optimization evaluation parameters other than multi-agent safety, so as to obtain a complete human-like decision model.

[0062] In one embodiment, according to the optimization evaluation parameters and the expected level of the corresponding evaluation parameters, the performance level of the driving decision result output by the human-like understanding and pre-judgment decision model on each optimization evaluation parameter is evaluated, so as to establish a decision result evaluation model; then, based on the evaluation result of the optimization parameter performance level (that is, based on the decision result evaluation model), the decision result obtained by the human-like understanding and pre-judgment decision model considering only the safety of multi-agent is optimized, so as to generate a human-like decision model.

[0063] Specifically, first, one or more optimization evaluation parameters and the expected target level of each optimization evaluation parameter (that is, the expected level of each optimization evaluation parameter) are obtained. In the embodiment of the present application, the optimization evaluation parameters represent a class of parameters with low correlation with the safety driving of the host vehicle, including but not limited to: driving efficiency (the time required for the host vehicle to complete the driving task from A to B, the shorter the time, the higher the driving efficiency), prosociality and driving comfort, etc. The expected level is a parameter representing the expected intelligent quantitative level of the current intelligent driving vehicle in the corresponding optimization evaluation parameter.

[0064] When evaluating the influence of the driving decision result on each optimization evaluation parameter, it is necessary to first analyze the influence of the current driving decision result on each optimization evaluation parameter, generate a reward and punishment function for each optimization evaluation parameter according to the expected level of the corresponding optimization evaluation parameter, and then according to the reward and punishment function of each optimization evaluation parameter, the influence of the driving decision result on each optimization evaluation parameter is quantitatively evaluated, so as to quantitatively evaluate the rationality of the driving decision result.

[0065] Next, the application step S130 will also be based on the above driving decision results on each optimization evaluation parameter performance level, according to the understanding and prediction of the decision model obtained only considering multi-agent safety, the decision results are optimized, thus, the application embodiment utilizes step S130 from the decision result evaluation aspect to optimize the understanding and prediction of the decision model constructed by step S120, forming a human-like decision model.

[0066] Next, taking the optimization evaluation parameter of driving efficiency as an example, the optimization process of the human-like understanding and prediction decision model is described, specifically:

[0067] 1) Analyze the positive or negative influence of the current driving decision result on the EV driving efficiency, and construct a reward and punishment function according to the expected driving efficiency;

[0068] 2) According to the reward and punishment function, the influence of the decision result on the driving efficiency is quantitatively evaluated;

[0069] 3) According to the evaluation result, the optimization direction of the human-like understanding-prediction decision model is obtained, so as to optimize the description of the decision output part of the human-like understanding-prediction decision model by using the current optimization direction, thereby completing the optimization of the human-like understanding-prediction decision model for the optimization evaluation parameter of driving efficiency, and obtaining a human-like decision model.

[0070] In this way, the embodiment of the application can also continue to optimize the human-like understanding-prediction decision model with other optimization evaluation parameters according to the processes in steps 1)-3) above, so as to construct a model for realizing a higher intelligent level of human-like decision, and obtain a human-like understanding-prediction decision model optimized by optimization evaluation parameters.

[0071] On the other hand, based on the above-mentioned human-like decision model construction method, the embodiment of the application also provides a system for constructing a human-like decision model of an intelligent vehicle (also referred to as a "human-like decision model construction system"). Figure 4 The module block diagram of the system for constructing a human-like decision model of an intelligent vehicle according to the embodiment of the application is shown in FIG. 4. Figure 4 As shown in FIG. 4, the human-like decision model construction system according to the embodiment of the application comprises a human-like understanding model construction module 41, a human-like understanding and prediction decision model construction module 42, and a decision result optimization module 43.

[0072] Specifically, the human-like understanding model construction module 41 is implemented according to the above step S110, and is configured to analyze the cognitive task of the host vehicle in the current traffic environment to determine the key elements related to the cognitive link of the host vehicle in the traffic scene, and based on this, analyze the representation method of multi-agent safety between the host vehicle and the traffic environment, and construct the human-like understanding model; the human-like understanding and pre-judgment decision model construction module 42 is implemented according to the above step S120, and is configured to analyze the long-term safety risk represented by the key elements, and according to the long-term safety risk evaluation result and the destination planning result, predict the traffic situation in the future time and space range, construct the heuristic pre-judgment model, and further combine the human-like understanding model to form the human-like understanding and pre-judgment decision model; the decision result optimization module 43 is implemented according to the above step S130, and is configured to optimize the decision result obtained by the human-like understanding and pre-judgment decision model according to the optimization evaluation parameters other than the multi-agent safety.

[0073] In addition, based on the above human-like decision model construction method, the embodiment of the present application further provides a human-like decision method for intelligent vehicles. The human-like decision method applies the human-like decision model generated by the human-like decision model construction method.

[0074] The present application discloses a method and system for constructing an intelligent vehicle human-like decision model, and a human-like decision method realized by applying the human-like decision model. Specifically, by establishing a human-like understanding model, a heuristic pre-judgment model and a decision result evaluation model, a human-like understanding-pre-judgment decision model is obtained, which improves the understanding level of the intelligent vehicle for the traffic situation, avoids the driving behavior of the intelligent vehicle that is difficult for human traffic participants to react in time and correctly, and ultimately improves the predictability of the intelligent vehicle. On the basis of ensuring multi-agent safety, an extension space for further optimizing the decision is given, which is beneficial to the intelligent vehicle to be integrated into the traffic system with higher social acceptance.

[0075] In summary, the present application has the following advantages and positive effects:

[0076] (1) Referring to the human decision-making process, the automatic driving decision is divided into three processes of traffic situation understanding, traffic situation pre-judgment and decision result optimization, so that the decision has time and space continuity, and the traffic accidents such as rear-end collision caused by sudden change of decision result are avoided;

[0077] (2) The human-like understanding model is established, so that the intelligent vehicle can perceive the safety risk in the multi-agent traffic environment, and compared with the two-by-two game decision analysis, the multi-agent safety can be ensured in a more complex traffic scene;

[0078] (3) The heuristic pre-judgment model is established, and the intelligent vehicle is given the perception ability of long-term safety, so that the behavior of the intelligent vehicle that is difficult to be predicted such as emergency braking is avoided from the source;

[0079] (4) By constructing a human-like understanding-prediction decision-making model, the transition mechanism between discrete decision results is revealed, so that the basis followed by each decision result remains spatial and temporal continuity and can be understood by other human road participants;

[0080] (5) Considering the decision performance of a higher intelligent level, a decision result evaluation model is established, which provides an extension interface for optimizing the decision strategy to realize a higher intelligent level of decision.

[0081] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0082] In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more; the orientation or position relationship indicated by the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail" and the like is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third" and the like are only for descriptive purposes, and cannot be understood as indicating or implying relative importance.

[0083] In the description of the present application, it should be noted that, unless otherwise specified and limited, the terms "connected", "connected" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0084] It should be understood that the embodiments disclosed in the present application are not limited to the specific structure, processing steps or materials disclosed herein, but should extend to equivalent alternatives of these features understood by those skilled in the related art. It should also be understood that the terms used herein are only for the purpose of describing the specific embodiments and do not mean limitation.

[0085] The phrase "one embodiment" or "an embodiment" appearing in the specification means that the specific features, structures or characteristics described in connection with the embodiment are included in at least one embodiment of the present application. Therefore, the phrase "one embodiment" or "an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment.

[0086] Although the present application has been described with reference to the above embodiments, the contents described are only the embodiments adopted for facilitating the understanding of the present application, and are not intended to limit the present application. Any modification and change in the form and details of the present application can be made by any person skilled in the art without departing from the spirit and scope of the present application, and the patent protection scope of the present application shall be subject to the scope defined by the appended claims.

Claims

1. A method for constructing a human-like decision-making model for intelligent vehicles, characterized in that, include: The analysis of the cognitive tasks of the main vehicle in the current traffic environment determines the key elements related to the cognitive process of the main vehicle in the traffic scenario. Based on this, the analysis of the multi-agent safety representation method between the main vehicle and the traffic environment is carried out, and a human-like understanding model is constructed. The cognitive task analysis of the main vehicle refers to the cognitive task that the main vehicle needs to complete when considering the impact of the lateral and longitudinal kinematic changes and dynamic characteristics between the main vehicle and all adjacent vehicles on factors including driving safety, road traffic regulations, and road surface conditions. The human-like understanding model is a model that uses the key elements of the traffic scenario as input to describe the understanding results of the traffic situation currently faced by the main vehicle. This paper analyzes the long-term safety risks represented by the key elements, and based on the long-term safety risk assessment results and destination planning results, predicts the future spatiotemporal range of traffic conditions, constructs a heuristic prediction model, and further combines it with the human-like understanding model to form a human-like understanding and prediction decision-making model. This model includes: based on the descriptions of the key elements and traffic conditions, selecting heuristic parameters that reflect long-term driving safety trends from the key elements, and using destination planning results as constraints to analyze the potential driving risks of these heuristic parameters under future development trends. A long-term safety assessment model is then established for these heuristic parameters, which is used to assess the continuous development trends of the various heuristic parameters. The safety data segment and the safety data segment are described. Then, based on the destination planning results and the long-term safety evaluation model, the continuous time range and spatial range of safe driving are evaluated, and the heuristic prediction model is established. The heuristic prediction model is a model that uses traffic situation understanding results as input to describe the time range and spatial range of the safe driving decision results of the current master vehicle when facing the current traffic situation. The human-like understanding and prediction decision model uses key elements of the traffic scene as input, considers the transient safety of traffic situation understanding under the multi-agent system and the decision to deal with long-term safety risks in the dynamic traffic environment of multi-agent systems, and takes the driving spatiotemporal range as the output model to ensure the spatiotemporal continuity of multi-agent safety and decision-making. Based on optimization evaluation parameters other than multi-agent safety, the decision results obtained from the human-like understanding and prediction decision-making model are optimized.

2. The method according to claim 1, characterized in that, Using the destination planning results as constraints, potential driving risks are analyzed, a long-term safety evaluation model is established, and a judgment result on the safety of multiple agents within a certain time and space range is generated, thereby generating the heuristic prediction model. The decision result is obtained based on the multi-agent safety output by the model by applying the heuristic prediction model in the traffic scenario.

3. The method according to claim 1 or 2, characterized in that, The step of optimizing the decision results obtained from the human-like understanding and prediction decision-making model based on optimization evaluation parameters other than multi-agent safety includes: Based on the optimized evaluation parameters and the expected levels of the corresponding evaluation parameters, evaluate the performance level of the driving decision results output by the human-like understanding and prediction decision-making model on the optimized evaluation parameters; Based on the performance level of the optimized parameters, the decision results output by the human-like understanding and prediction decision-making model are optimized.

4. The method according to claim 3, characterized in that, The step of evaluating the influence of the driving decision output by the human-like understanding and prediction decision-making model on the optimized evaluation parameters based on the optimized evaluation parameters and the expected levels of the corresponding evaluation parameters includes: The impact of the driving decision results on each of the optimization evaluation parameters is analyzed, and a reward / penalty function is generated for each of the optimization evaluation parameters based on the expected level. The optimization evaluation parameters include, but are not limited to, driving efficiency, prosociality, and driving comfort. Based on the reward and punishment function, the impact of the driving decision result on each of the optimization evaluation parameters is quantitatively evaluated.

5. The method according to claim 1 or 2, characterized in that, The steps for identifying key elements of the primary vehicle in a traffic scenario that are relevant to the cognitive process include: Based on the perceived information, a typical traffic scenario model representing the current driving environment of the main vehicle is established; Based on the interaction between the master vehicle and the multi-agent, the cognitive tasks that the master vehicle needs to complete when the kinematic and dynamic changes between the master vehicle and the multi-agent, and the impact of traffic regulations and road conditions on the safety of the multi-agent are analyzed. Based on the results of the cognitive task analysis, the key elements in the traffic scenario are identified.

6. The method according to claim 1 or 2, characterized in that, The method further includes: simulating and validating the reliability and validity of the human-like understanding model by setting up test scenarios.

7. The method according to claim 6, characterized in that, The test scenarios are designed with a series of combined conditions, including different numbers of vehicles alongside the vehicle, different combinations of the main vehicle's movement state and the movement states of the vehicles alongside the vehicle, different road conditions, and different traffic rules.

8. A human-like decision-making method for intelligent vehicles, characterized in that, The human-like decision-making method is implemented using the human-like understanding and predictive decision-making model constructed as described in any one of claims 1 to 7.

9. A system for constructing a human-like decision-making model for intelligent vehicles, characterized in that, include: The human-like understanding model construction module is configured to analyze the cognitive tasks of the main vehicle in the current traffic environment, determine the key elements related to the cognitive process of the main vehicle in the traffic scenario, and based on this, analyze the multi-agent safety representation method between the main vehicle and the traffic environment, and construct a human-like understanding model. The main vehicle cognitive task analysis refers to the cognitive tasks that the main vehicle needs to complete when considering the impact of the lateral and longitudinal kinematic changes and dynamic characteristics between the main vehicle and all adjacent vehicles on factors including driving safety, road traffic regulations, and road surface conditions. The human-like understanding model is a model that uses key elements of the traffic scenario as input to describe the understanding results of the traffic situation currently faced by the main vehicle. The human-like understanding and predictive decision-making model construction module is configured to analyze the long-term safety risks represented by the key elements, and based on the long-term safety risk assessment results and destination planning results, predict the future spatiotemporal range of traffic conditions, construct a heuristic predictive model, and further combine it with the human-like understanding model to form a human-like understanding and predictive decision-making model. This includes: based on the descriptions of the key elements and traffic conditions, selecting heuristic parameters that can reflect long-term driving safety trends from the key elements, and using destination planning results as constraints to analyze the potential driving risks of the heuristic parameters under future development trends, establishing a long-term safety assessment model for the heuristic parameters. The long-term safety assessment model is used to evaluate the various heuristic parameters. The unsafe and safe data segments in the continuous development trend are described. Then, based on the destination planning results and the long-term safety evaluation model, the continuous time and space range of safe driving is evaluated, and the heuristic prediction model is established. The heuristic prediction model is a model that uses traffic situation understanding results as input to describe the time and space range of the safe driving decision results of the current master vehicle when facing the current traffic situation. The human-like understanding and prediction decision model uses key elements of the traffic scene as input, considers the transient safety of traffic situation understanding under the multi-agent system and the decision to deal with long-term safety risks in the dynamic traffic environment of multi-agent systems, and takes the driving spatiotemporal range that ensures the safety of multi-agents and the spatiotemporal continuity of decision-making as the output model. The decision result optimization module is configured to optimize the decision results obtained by the human-like understanding and prediction decision model based on optimization evaluation parameters other than multi-agent safety.

Images