Method for evaluating confluence position of vehicles, and model generation method and device for confluence position evaluation

Abstract

To accurately evaluate a confluence position with a relatively light load.SOLUTION: A confluence position evaluation method includes the steps of: simulating the behavior of the own vehicle and a group of other vehicles after various initial conditions taken as departure points; calculating, for each gap, an evaluation value (Em) related to the success or failure of a confluence operation of the own vehicle breaking into a gap of the group of other vehicles, on the basis of the result of the simulation; generating a regression model (M2) on the basis of an aggregate of data sets including the calculated evaluation value (Em) and an initial condition corresponding thereto; and acquiring a prescribed feature amount for each gap and inputting the acquired feature amount to the regression model (M2) as an explanatory variable, thereby identifying the evaluation value (Em) of each gap. The feature amount includes a predictive feature amount (Qp) of the same kind as an evaluation value (Em) that is predicated under a given condition from the state at confluence start time.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a method for evaluating a merging position when a vehicle merges from a branch line onto a main line on which a group of other vehicles is traveling, and a method and device for generating a model used for evaluating the merging position. [Background technology]

[0002] In recent years, as part of research into vehicle driving assistance or autonomous driving, assistance or automation of a merging operation of a vehicle merging from a branch line onto a main line has been considered. For example, Patent Document 1 below discloses that in a behavior plan generation device that generates a behavior plan for a vehicle merging onto a main line, a learning model generated from driving data obtained through a simulator function that allows a user to experience simulated vehicle driving is used to calculate a yielding degree that indicates the likelihood that other vehicles traveling on the main line will yield to the user's vehicle, and a behavior plan for the user's vehicle merging onto the main line is generated based on the calculated yielding degree. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2023-146576 Summary of the Invention [Problem to be solved by the invention]

[0004] Here, when a vehicle merges onto a main lane, it is necessary to determine one of a plurality of gaps (vehicle gaps) in a group of other vehicles traveling on the main lane as a target merging position. However, in the above-mentioned Patent Document 1, the determination of whether or not it is actually possible to merge at the target merging position, i.e., the evaluation of the merging position, is performed based on the output value (degree of yielding) of a learning model obtained from repeated experiments by subjects using a simulator function, which results in the problem that establishing the evaluation logic is time-consuming and labor-intensive. In particular, in order to ensure sufficient evaluation accuracy, measures such as sufficiently increasing the number of experiments are required, making it difficult to achieve both ensuring accuracy and reducing the amount of work.

[0005] The present invention has been made in consideration of the above circumstances, and aims to provide a vehicle merging position evaluation device and method that can accurately evaluate merging positions with a relatively light load. [Means for solving the problem]

[0006] In order to solve the above problem, a method for evaluating a merging position for a vehicle according to one aspect of the present invention is a method for evaluating a merging position when a host vehicle merges from a branch line onto a main line on which a group of other vehicles is traveling, the method comprising: a simulation step of using a driving behavior model that predicts the behavior of the host vehicle and the group of other vehicles to simulate subsequent behavior of the host vehicle and the group of other vehicles starting from various initial conditions with different traffic environments; an evaluation value calculation step of calculating, for each gap based on the results of the simulation, an evaluation value regarding the success or failure of a merging operation of the host vehicle to cut into a gap in the group of other vehicles; and a correlation between the calculated evaluation value and a corresponding previous evaluation value. and a merging position evaluation step of acquiring the feature values ​​for each gap and inputting the acquired feature values ​​as the explanatory variables into the regression model to identify the evaluation value of each gap, and determining which gap should be set as a merging target based on the identified evaluation value, wherein the feature values ​​include predicted feature values ​​of the same type as the evaluation value, which are predicted under certain conditions from the state at the start of merging.

[0007] According to the present invention, a simulation is performed using a predetermined driving behavior model to predict behavior at merging, starting from various initial conditions, and a regression model for evaluating merging positions is generated based on a set of data sets of initial conditions and evaluation values ​​obtained from the simulation results. This makes it possible to generate a regression model relatively easily, without undergoing complex processing such as reinforcement learning.

[0008] Furthermore, in an actual merging scene, a unique feature value corresponding to each gap in a group of other vehicles is acquired, and the feature value is input into the previously generated regression model to identify an evaluation value for each gap. In this way, an evaluation value for each gap can be obtained simply by inputting the feature value for each gap into the regression model. Therefore, it is possible to determine in a short time which gap should be set as the merging target, and to appropriately determine an action plan even in an actual merging scene where it is difficult to secure sufficient calculation time.

[0009] Furthermore, in the present invention, a predicted feature of the same type as the evaluation value, predicted from the state at the start of the merge, is used as one of the features (explanatory variables) input to the regression model, thereby improving the regression accuracy. That is, research by the present inventors has shown that the regression accuracy of a regression model improves when a predicted feature representing a future state after some time has passed since the start of the merge is included as an explanatory variable. In particular, it has been shown that using an index of the same type as the evaluation value output by the regression model as a predicted feature leads to improved regression accuracy. Therefore, according to the present invention, in which a predicted feature of the same type as the evaluation value is used as an explanatory variable as described above, the regression accuracy of the regression model can be improved, and the evaluation value of each gap can be accurately identified.

[0010] Preferably, in the merging position evaluation step, the simulation is performed under conditions where model parameters of the driving behavior model are fixed, and the predicted feature amount is calculated for each gap from a result of the simulation.

[0011] In this aspect, the predicted feature amount can be easily calculated by a simple simulation using the same model as the driving behavior model prepared for generating the regression model.

[0012] Preferably, in the evaluation value calculation step, a merging margin, which is the remaining distance or remaining time to the end of the branch line at the time when the vehicle has completed merging or is expected to complete merging, is calculated for each gap, and the evaluation value for each gap is calculated based on the merging margin.

[0013] The merging margin expressed as distance or time is a value that can change continuously depending on the situation, so it can be used not only as an indicator for determining success / failure, but also as an indicator for estimating how much margin is available for merging. Therefore, according to this mode of evaluating each gap using an evaluation value based on such merging margin, it is possible to appropriately determine how much margin is expected when each gap is set as a merging target, and to appropriately determine a merging target from among each gap.

[0014] In the above aspect, more preferably, the predicted characteristic amount is the merging margin predicted from a state at the start of merging.

[0015] The merging margin as a predicted feature can be considered to be an index of the same kind as the evaluation value calculated from the merging margin. Therefore, when such a merging margin is used as a predicted feature, the regression accuracy can be improved as described above.

[0016] Preferably, the feature amounts include, in addition to the predicted feature amounts, basic feature amounts calculated directly from a state at the start of merging without going through prediction.

[0017] According to research by the inventors of the present application, it has been found that the regression accuracy of the regression model is improved by including, as explanatory variables, not only the predicted feature quantities described above but also basic feature quantities that directly represent the state at the time of merging. Therefore, according to this aspect, which uses a combination of predicted feature quantities and basic feature quantities, it is possible to further improve the regression accuracy.

[0018] Preferably, the driving behavior model includes a following driving model that predicts the behavior of the group of other vehicles and outputs acceleration required to drive while maintaining an appropriate inter-vehicle distance from the preceding vehicle.

[0019] Furthermore, the driving behavior model includes a proportional model that outputs acceleration proportional to the difference between a predetermined target position and a predetermined target speed as a model for predicting the behavior of the host vehicle.

[0020] In these aspects, the future behavior of the other vehicle group and the host vehicle can be appropriately simulated through calculations with a relatively light load using specified calculation formulas.

[0021] A model generation method according to another aspect of the present invention is a method for generating a regression model for evaluating a merging position when a host vehicle merges from a branch lane onto a main lane on which a group of other vehicles is traveling, the method including: a simulation step of simulating the subsequent behavior of the host vehicle and the group of other vehicles, starting from various initial conditions that differ in traffic environments, using a driving behavior model that predicts the behavior of the host vehicle and the group of other vehicles; an evaluation value calculation step of calculating, for each gap, an evaluation value regarding the success or failure of the host vehicle's merging operation to cut into a gap in the group of other vehicles based on the results of the simulation; and a regression model generation step of generating a regression model in which feature amounts specific to each gap are used as explanatory variables and the evaluation value of each gap is used as a response variable, based on a set of data sets including the evaluation values ​​calculated for each gap and the initial conditions corresponding to the evaluation values, wherein the feature amounts include prediction feature amounts of the same type as the evaluation values, which are predicted under certain conditions from the state at the start of merging.

[0022] A model generation device according to yet another aspect of the present invention is a device for generating a regression model for evaluating a merging position when a host vehicle merges from a branch lane onto a main lane on which a group of other vehicles is traveling, the model generation device including: a simulation unit that uses a driving behavior model that predicts the behavior of the host vehicle and the group of other vehicles to simulate the subsequent behavior of the host vehicle and the group of other vehicles, starting from various initial conditions that differ in traffic environments; an evaluation value calculation unit that calculates, for each gap, an evaluation value regarding the success or failure of the host vehicle's merging operation to cut into a gap in the group of other vehicles based on the results of the simulation; and a regression analysis unit that generates a regression model in which feature amounts specific to each gap are used as explanatory variables and the evaluation value of each gap is used as a response variable, based on a set of data sets including the evaluation values ​​calculated for each gap and the corresponding initial conditions, the feature amounts including prediction feature amounts of the same type as the evaluation values ​​that are predicted under certain conditions from the state at the start of merging.

[0023] According to these model generation methods and devices, it is possible to generate a highly accurate regression model for evaluating a merging position relatively easily. [Effects of the Invention]

[0024] As described above, according to the present invention, the merge position can be evaluated with high accuracy and with a relatively light load. [Brief explanation of the drawings]

[0025] [Figure 1] 1 is a functional block diagram showing the configuration of a vehicle control system to which a merging position evaluation method according to an embodiment of the present invention is applied. [Figure 2] FIG. 1 is a schematic diagram illustrating an example of a merging scene in which a vehicle merges from a branch line onto a main line. [Figure 3] FIG. 1 is a functional block diagram showing the configuration of a model generation device that generates a regression model for merging position evaluation. [Figure 4A] FIG. 10 is a schematic diagram for explaining a situation in which the merging margin distance is positive. [Figure 4B]FIG. 10 is a schematic diagram for explaining a situation in which the merging margin distance becomes negative. [Figure 5] FIG. 2 is a schematic diagram showing the input / output relationship of the regression model. [Figure 6] FIG. 2 is a schematic diagram showing the contents of basic feature amounts input as explanatory variables to the regression model. [Figure 7] 10 is an equation showing the contents of a driving behavior model that predicts the behavior of the host vehicle and other vehicles when merging. [Figure 8] 10 is a flowchart showing a procedure for generating the regression model. [Figure 9] 9 is a subroutine showing details of step S4 in FIG. 8. [Figure 10] 10 is a subroutine showing details of step S13 in FIG. 9. [Figure 11A] 3A and 3B are schematic diagrams showing examples of setting a target position and a target speed of a host vehicle; [Figure 11B] 10A and 10B are schematic diagrams showing other examples of setting the target position and target speed of the host vehicle. [Figure 12] 1 is a graph showing the relationship between the value of a model parameter and the frequency of occurrence thereof; [Figure 13] 10 is a graph showing the relationship between the merging margin distance calculated by the above simulation and the frequency of occurrence thereof. [Figure 14] 10 is a graph for explaining that the margin evaluation value calculated from the merging margin distance changes depending on a coefficient. [Figure 15] 10 is a flowchart showing details of control performed by a controller of the host vehicle in an actual merging scene. [Figure 16] 10 is a bar graph illustrating the distribution of margin evaluation values ​​for each gap. DETAILED DESCRIPTION OF THE INVENTION

[0026] A preferred embodiment of the vehicle merging position evaluation method of the present invention will be described in detail below with reference to the drawings. The vehicle to which the present invention is applied may be any of an engine vehicle powered by an internal combustion engine, an electric vehicle powered by an electric motor, and a hybrid vehicle powered by both an internal combustion engine and an electric motor. The merging position evaluation device and method may evaluate a merging position during a merging operation of an autonomous vehicle that does not require driver operation, or may evaluate a merging position during a merging operation of a vehicle with a driving assistance function that assists the driver in driving.

[0027] (1) Control system Fig. 1 is a functional block diagram showing the configuration of a vehicle control system to which a merging position evaluation method according to one embodiment of the present invention is applied. As shown in the diagram, the control system includes a controller 10 that controls the vehicle, a group of internal sensors 20 that detect the vehicle's driving state, a group of external sensors 21 that detect the vehicle's surrounding conditions, a GNSS receiver 22 that identifies the vehicle's position, and a group of actuators 25 that drive the vehicle. The control system is mounted on a vehicle 1 shown in Fig. 2. Hereinafter, the vehicle 1 may be referred to as the host vehicle 1 to distinguish it from another vehicle 2.

[0028] The controller 10 is configured with a microcomputer as its main components, including an arithmetic unit 11 including a processor such as a CPU, a memory unit 17 including ROM, RAM, and an HDD, and peripheral circuits not shown in the figure such as an I / O interface.

[0029] The controller 10 is electrically connected to the internal sensor group 20, the external sensor group 21, the GNSS receiver 22, and the actuator group 25. That is, the controller 10 receives various information output from the sensor groups 20, 21 and the GNSS receiver 22, and outputs control signals to the actuator group 25.

[0030] The internal sensor group 20 is a collective term for sensors for detecting various information representing the traveling state of the host vehicle 1. The internal sensor group 20 includes a vehicle speed sensor for detecting the speed (traveling speed) of the host vehicle 1 and an acceleration sensor for detecting the acceleration of the host vehicle 1.

[0031] The GNSS receiver 22 is a receiver for determining the vehicle's own position using the Global Navigation Satellite System (GNSS). The GNSS receiver 22 determines the position of the vehicle 1 on the Earth by receiving and processing radio signals transmitted from GNSS satellites.

[0032] The external sensor group 21 is a collective term for sensors for recognizing targets around the host vehicle 1, and includes at least one of an optical sensor and an imaging sensor. Examples of optical sensors include radar such as millimeter-wave radar or a ranging device using LiDAR (light detection and ranging). Examples of imaging sensors include cameras including imaging elements such as CCDs and CMOSs. Alternatively, the external sensor group 21 may be a sensor fusion system including multiple types of sensors including optical sensors and imaging sensors and integration means for integrating and processing signals obtained from the sensors. Note that at least a portion of the external sensor group 21 may be optical sensors or imaging sensors installed on the road on which the host vehicle 1 travels. In this case, the controller 10 of the host vehicle 1 acquires information detected by the sensors on the road via appropriate communication means.

[0033] The actuator group 25 is a general term for actuators for driving that realize driving of the host vehicle 1. The actuator group 25 includes, for example, a power source such as an engine or a motor that generates power for driving, and a motor (steering motor) that drives a steering mechanism for changing the direction of travel.

[0034] The memory unit 17 stores various data necessary for driving control of the host vehicle 1. The data stored in the memory unit 17 include road map data M1, a regression model M2, and a driving behavior model M3. The road map data M1 is, for example, high-precision three-dimensional map data (HD map). In this case, the road map data M1 includes, for example, information on the position, shape, and gradient of the road, position information on traffic lights and signs on the road, and information on the width and position of lanes. The regression model M2 is a model used to evaluate the merging position of the host vehicle 1 in a merging scene (FIG. 2) described later. The driving behavior model M3 is a mathematical model for predicting the behavior of the host vehicle 1 and the other vehicle 2. At least one of the road map data M1, the regression model M2, and the driving behavior model M3 does not have to be stored in the memory unit 17 in advance, but may be input from an external server via a communication device (not shown) and temporarily stored in the memory unit 17.

[0035] The calculation unit 11 functionally includes a recognition unit 12, an evaluation unit 13, a route generation unit 14, and a travel control unit 15.

[0036] The recognition unit 12 is a module that recognizes the state of the host vehicle 1 and the state of the surroundings of the host vehicle 1, including the other vehicle 2. For example, the recognition unit 12 acquires the speed and acceleration of the host vehicle 1 and the position of the host vehicle 1 on the road based on information input from the internal sensor group 20 and the GNSS receiver 22. The recognition unit 12 also acquires the position and speed of the other vehicle 2 based on information input from the external sensor group 21. Furthermore, the recognition unit 12 acquires the shape of the road around the host vehicle 1, etc. from the road map data M1.

[0037] The evaluation unit 13 is a module that performs evaluation to determine a merging position for the host vehicle 1 at the time of merging shown in FIG. 2 , that is, when the host vehicle 1 passes through the branch line 102 and merges onto the main line 101. The evaluation unit 13 uses a regression model M2 for this evaluation. That is, when the host vehicle 1 merges onto the main line 101, the evaluation unit 13 uses the regression model M2 to evaluate which gap 3 in the group of other vehicles 2 traveling on the main line 101 the host vehicle 1 should cut into, that is, which gap 3 is appropriate as a merging position for the host vehicle 1.

[0038] In the example of FIG. 2, the host vehicle 1 is traveling on a branch line 102 including a parallel section that extends parallel to the main line 101 in order to merge onto the main line 101. Furthermore, a plurality of other vehicles 2 (hereinafter also referred to as the other vehicle group 2) are traveling on the main line 101. A gap 3 is formed between each of the vehicles adjacent to each other in the fore-and-aft direction (travel direction) in the other vehicle group 2. When merging onto the main line 101, the host vehicle 1 determines which gap 3 in the other vehicle group 2 the host vehicle 1 should cut into, and moves toward the determined gap 3. In order to determine in advance the merging operation of the host vehicle 1, the evaluation unit 13 evaluates each gap 3 using a regression model M2. That is, the evaluation unit 13 regards each of the plurality of gaps 3 that the host vehicle 1 may cut into as candidate merging positions, and performs evaluation using the regression model M2 to compare the merits and demerits of each candidate.

[0039] As will be described in detail later, the regression model M2 receives an explanatory variable consisting of a predetermined feature amount corresponding to each gap 3, and outputs, as a response variable, a predetermined evaluation value (a margin evaluation value Em described later) that indicates the relative merits of each gap 3 as a merging position. The evaluation unit 13 evaluates each gap 3 based on the evaluation value output by the regression model M2.

[0040] For example, as shown in FIG. 2, a group of other vehicles 2 is assumed to consist of a first vehicle 2A, a second vehicle 2B, a third vehicle 2C, and a fourth vehicle 2D traveling on the main road 101 in this order from the front in the traveling direction. In this case, three gaps 3A to 3C are evaluated. That is, the first gap 3A between the first vehicle 2A and the second vehicle 2B, the second gap 3B between the second vehicle 2B and the third vehicle 2C, and the third gap 3C between the third vehicle 2C and the fourth vehicle 2D are evaluated. The evaluation unit 13 assumes a case in which the host vehicle 1 will cut into each of the first to third gaps 3A to 3C, and inputs explanatory variables corresponding to each of the gaps 3A to 3C into the regression model M2. The regression model M2 then outputs an evaluation value for each of the gaps 3A to 3C as a response variable. The evaluation unit 13 evaluates which of the gaps 3A to 3C is a better merging position based on the evaluation value of each of the gaps 3A to 3C obtained in this manner.

[0041] The route generation unit 14 is a module that generates a movement route for the host vehicle 1 to a target merging position, i.e., a merging target. That is, the route generation unit 14 determines one gap 3 with a high evaluation as the merging target based on the evaluation of each gap 3 by the evaluation unit 13. Then, assuming that the host vehicle 1 will move to the determined merging target, the route generation unit 14 generates a movement route for the host vehicle 1 required for this purpose.

[0042] The traveling control unit 15 is a module that controls the movement of the host vehicle 1 along the travel route generated by the route generation unit 14. That is, the traveling control unit 15 controls the actuator group 25 that causes the host vehicle 1 to travel, thereby moving the host vehicle 1 along the travel route and leading it to the merging target.

[0043] (2) Regression model generator The regression model M2 described above is generated by a model generating device 30 shown in Fig. 3. The model generating device 30 includes a calculation unit 31 including a processor such as a CPU, a storage unit 37 including a ROM, a RAM, an HDD, and peripheral circuits not shown in the figure such as an I / O interface. The model generating device 30 is typically a personal computer or a server located away from the host vehicle 1, but may also be mounted on the host vehicle 1.

[0044] The memory unit 37 stores various data required for generating the regression model M2. The data stored in the memory unit 37 includes a driving behavior model M3 and an initial condition database M4. The driving behavior model M3 is a mathematical model for predicting the behavior of the host vehicle 1 and other vehicles 2, which is the same as that of the vehicle controller 10 (FIG. 1). However, in this case, the driving behavior model M3 is used for simulation purposes to create training data that forms the basis of the regression model M2, that is, to create training data by simulating the behavior of the host vehicle 1 and other vehicles 2 when merging. The initial condition database M4 is a database for setting initial conditions for the simulation.

[0045] The initial condition database M4 stores various initial conditions for different traffic environments such as road shapes, vehicle placement patterns, etc. That is, the initial condition data stored in the initial condition database M4 includes a large number of road shape data that differ in the length of the branch line 102 (particularly the length of the section that runs parallel to the main line 101), etc., and a large number of vehicle placement data that differ in the positions and speeds of the host vehicle 1 and the other vehicle group 2, the number of vehicles in the other vehicle group 2, etc.

[0046] The calculation unit 31 functionally includes an initial condition setting unit 32, a simulation unit 33, an evaluation value calculation unit 34, and a regression analysis unit 35.

[0047] The initial condition setting unit 32 is a module that sets the initial conditions of the simulation. That is, the initial condition setting unit 32 sequentially reads road shape data and vehicle location data from the initial condition database M4 and sets the read data as the initial conditions of the simulation. Furthermore, in this embodiment, the initial condition setting unit 32 also adds noise to the data read from the initial condition database M4 to make small changes. This is to further increase the randomness of the data.

[0048] The simulation unit 33 is a module that simulates the behavior of the host vehicle 1 and the other vehicle group 2 when merging using the driving behavior model M3 in order to create training data for the regression model M2. That is, the simulation unit 33 starts from various initial conditions set by the initial condition setting unit 32 and repeatedly simulates the subsequent behavior of the host vehicle 1 and the other vehicle group 2. Each simulation simulates a merging operation in which the host vehicle 1 cuts into any gap 3 in the other vehicle group 2. The results of this reproduction (simulation) are used as training data for generating the regression model M2.

[0049] The evaluation value calculation unit 34 is a module that calculates an evaluation value regarding the success or failure of the merging operation based on the result of the simulation performed by the simulation unit 33. In this embodiment, the evaluation value that is calculated is a margin evaluation value Em (see formula (1) described later) that indicates how much margin is available for merging.

[0050] Specifically, the margin evaluation value Em is a value calculated based on a merging margin, which is the remaining distance or remaining time to the terminus 102a of the branch line 102 (hereinafter also referred to as the branch line terminus 102a) at the time when the host vehicle 1 has completed merging or is expected to complete merging. As shown in FIG. 4A, in this embodiment, the merging margin is calculated as a merging margin, which is the distance from the host vehicle 1 to the branch line terminus 102a at the time when merging completion is expected. Here, the time when merging completion is expected is the time when merging will be completed if the host vehicle 1 simply changes lanes, moving laterally from the branch line 102 to the main line 101. As will be described in detail later, in this embodiment, it is determined that merging completion is expected when merging possible conditions are met, including conditions that the size of the gap 3 (second gap 3B in the example of FIG. 4A) at the cut-in destination is larger than a predetermined value and the host vehicle 1 is traveling near the gap 3 at a speed close to that of the other vehicle 2.

[0051] The merging margin distance Lm can be calculated as a negative value. In other words, in the above simulation, even if the merging condition is not yet met when the host vehicle 1 approaches the branch line terminal end 102a and merging completion is not expected, in other words, even if it is determined that merging will fail, the calculation using the driving behavior model M3 continues, assuming that there is another lane beyond the branch line terminal end 102a. The calculation continues until the merging condition is met (merging completion is expected), and the negative merging margin distance Lm shown in FIG. 4B is calculated. In other words, the situation in FIG. 4B is a situation in which merging is not possible unless the host vehicle 1 moves beyond the branch line terminal end 102a, which does not actually exist, so the merging margin distance Lm is negative. Conversely, when the merging condition is met just before the branch line terminal end 102a as shown in FIG. 4A, the merging margin distance Lm is calculated as a positive value. Thus, the merging margin distance Lm can be calculated regardless of whether merging is successful or not, and is an evaluation value that can continuously change from positive to negative.

[0052] The evaluation value calculation unit 34 performs an evaluation for each gap 3 based on the merging margin distance Lm calculated as described above. The larger the merging margin distance Lm, the more margin there is for merging. Therefore, roughly speaking, the evaluation of each gap 3 can be performed in such a way that the larger the merging margin distance Lm, the higher the evaluation of the gap 3. However, in this embodiment, a simulation is repeatedly performed for each gap 3 to calculate the merging margin distance Lm each time, and each gap 3 is statistically evaluated from the distribution (mean μ and standard deviation σ) of the merging margin distance Lm obtained from the simulation results. In other words, a margin evaluation value Em, which is a statistical evaluation value, is calculated from the distribution of the merging margin distance Lm, and each gap 3 is evaluated based on the margin evaluation value Em. This will be described in detail later.

[0053] The regression analysis unit 35 is a module that generates a regression model M2 from a collection of data sets of the calculated margin evaluation values ​​Em and their corresponding initial conditions. That is, the regression analysis unit 35 collects data sets including the initial conditions of each simulation performed by the simulation unit 33 and the margin evaluation values ​​Em calculated by the evaluation value calculation unit 34 from the results of the simulation, and generates the regression model M2 by analyzing training data, which is a collection of the data sets. The generated regression model M2 is a model that uses feature quantities specific to each gap 3 as explanatory variables and the margin evaluation values ​​Em of each gap 3 as objective variables.

[0054] FIG. 5 is a schematic diagram showing the input / output relationship of the regression model M2. As shown in this figure, the explanatory variables of the regression model M2 are broadly divided into two types of feature quantities Q and Qp. That is, in this embodiment, the feature quantities (explanatory variables) input to the regression model M2 include a basic feature quantity Q obtained from the state at the start of merging, which corresponds to the initial condition, and a predicted feature quantity Qp obtained from a future state predicted based on the state at the start of merging. The regression analysis unit 35 uses multiple regression analysis to examine the correlation between the basic feature quantity Q and the predicted feature quantity Qp and the headroom evaluation value Em from the set of data described above, and identifies a regression equation that derives the headroom evaluation value Em from both feature quantities Q and Qp. In other words, the regression model M2 is a model that outputs a headroom evaluation value Em corresponding to the input feature quantities Q and Qp through the regression equation.

[0055] The basic feature amount Q and the predicted feature amount Qp are values ​​obtained for each gap 3 of the other vehicle group 2. In other words, if the basic feature amount Q and the predicted feature amount Qp for a certain gap 3 are known, the headroom evaluation value Em for that gap 3 can be found by inputting them into the regression model M2.

[0056] In this embodiment, the predicted feature quantity Qp is the merging margin distance Lm calculated (predicted) for each gap 3. The merging margin distance Lm in this case is calculated by a simulation using the driving behavior model M3, similar to the calculation of the margin evaluation value Em described above. That is, the behaviors of the host vehicle 1 and the other vehicle group 2 after the start of merging are simulated using the driving behavior model M3 to predict a future state in which the merging condition is met, and the merging margin distance Lm (FIGS. 4A and 4B) is calculated based on that state. As will be described in detail later, the merging margin distance Lm as the predicted feature quantity Qp is simply calculated by a one-time simulation. Furthermore, the merging margin distance Lm as the predicted feature quantity Qp can be considered to be an index of the same kind as the margin evaluation value Em, which is also an evaluation value based on the distribution of the merging margin distance Lm. In other words, the predicted feature quantity Qp in this embodiment is an index of the same kind as the evaluation value (margin evaluation value Em) predicted under certain conditions from the state at the start of merging.

[0057] The contents of the basic feature quantity Q will be explained using FIG. 6. FIG. 6 shows the layout of the host vehicle 1 and the other vehicle group 2 at the time when the virtual merging in the above simulation starts (i.e., the time corresponding to the initial conditions). As shown in this figure, the basic feature quantity Q includes five feature quantities consisting of a tentative inter-vehicle distance Q1, a tentative inter-vehicle time Q2, a risk perception index Q3, a rear inter-vehicle distance Q4, and a remaining time Q5 to the branch line end 102a. As will be described later, these five feature quantities Q1 to Q5 are calculated from the positions or speeds of the host vehicle 1 and the other vehicle group 2 at the time when the merging starts, the position of the gap 3 in the other vehicle group 2, and the road shape. In other words, the basic feature quantities Q (Q1 to Q5) are values ​​calculated directly from the state at the time when the merging starts without going through prediction.

[0058] Each of the feature quantities Q1 to Q5 is calculated for each gap 3. As an example, FIG. 6 shows definitional equations that define each of the feature quantities Q1 to Q5 corresponding to the first gap 3A between the first vehicle 2A and the second vehicle 2B. FIG. 6 also shows a group of parameters related to each definitional equation, i.e., a group of parameters used to calculate each of the feature quantities Q1 to Q5. These are seven parameters: p0, p1, p2, p3, pe, v0, and v2. p1 represents the position of the first vehicle 2A, p2 represents the position of the second vehicle 2B, p3 represents the position of the third vehicle 2C, and pe represents the position of the branch line end 102a. v0 represents the speed of the host vehicle 1, and v2 represents the speed of the second vehicle 2B. Furthermore, p0 represents a tentative merging position assuming that the host vehicle 1 has merged into the first gap 3A at this time. The tentative merging position p0 is designated as an appropriate position within the first gap 3A, and for example, the tentative merging position p0 may be near the center between the positions p1 and p2 of the other vehicles 2A and 2B traveling before and after the gap 3A.

[0059] The five feature amounts Q1 to Q5 corresponding to the first gap 3A are each calculated from two or more parameters selected from p0, p1, p2, p3, pe, v0, and v2.

[0060] That is, the tentative inter-vehicle distance Q1 is the inter-vehicle distance between the host vehicle 1 and the other vehicle 2 (following vehicle) behind it when it is assumed that the host vehicle 1 has entered one of the gaps 3 at the moment when merging has started. For example, in the case of FIG. 6 where the host vehicle 1 enters the first gap 3A, the tentative inter-vehicle distance Q1 is defined as the difference (p0-p2) between the tentative merging position p0 specified within the first gap 3A and the position p2 of the second vehicle 2B, which is the following vehicle.

[0061] The tentative inter-vehicle time Q2 is the inter-vehicle time between the host vehicle 1 and the following vehicle when it is assumed that the host vehicle 1 has currently entered one of the gaps 3. For example, in the case of FIG. 6 where the host vehicle 1 has entered the first gap 3A, the tentative inter-vehicle time Q2 is defined as the difference (p0-p2) between the tentative merging position p0 and the position p2 of the second vehicle 2B divided by the speed v2 of the second vehicle 2B.

[0062] The risk perception index Q3 is an index that represents the sense of risk from the perspective of the following vehicle when it is assumed that the host vehicle 1 has currently entered one of the gaps 3. For example, in the case of FIG. 6 where the host vehicle 1 is entering the first gap 3A, the risk perception index Q3 is defined by the equation shown in the figure, using the tentative merging position p0, the position p2 and speed v2 of the second vehicle 2B, and the speed v0 of the host vehicle 1.

[0063] The rear inter-vehicle distance Q4 is the distance between the vehicle following behind the host vehicle 1 and the vehicle following behind that vehicle, assuming that the host vehicle 1 has currently entered one of the gaps 3. For example, in the case of FIG. 6 where the host vehicle 1 has entered the first gap 3A, the rear inter-vehicle distance Q4 is defined as the difference (p2-p3) between the position p2 of the second vehicle 2B and the position p3 of the third vehicle 2C.

[0064] The remaining time Q5 to the branch line end 102a is the time required for the host vehicle 1 to travel to the branch line end 102a, assuming that the host vehicle 1 has currently entered one of the gaps 3. For example, in the case of FIG. 6 where the host vehicle 1 enters the first gap 3A, the remaining time Q5 is defined as the difference (pe-p0) between the position pe of the branch line end 102a and the tentative merging position p0 within the first gap 3A, divided by the speed v0 of the host vehicle 1.

[0065] 6 shows the definition formulas for the feature quantities Q1 to Q5 when the cut-in destination of the vehicle 1, i.e., the merging position, is the first gap 3A, but the feature quantities Q1 to Q5 are calculated using the same definition formulas when the merging position is another gap 3B or 3C. However, the vehicle and gap corresponding to each parameter of the definition formula changes according to the change in the merging position.

[0066] That is, when the first gap 3A is the target, as shown in FIG. 6, the feature quantities Q1 to Q5 are calculated using parameters including the position (p1) of the first vehicle 2A, the position (p2) of the second vehicle 2B, the position (p3) of the third vehicle 2C, the position (p0) of the first gap 3A, and the speed (v2) of the second vehicle 2B. In contrast, when the second gap 3B is the target, the vehicles and gaps corresponding to each parameter are shifted backward by one position. That is, the feature quantities Q1 to Q5 of the second gap 3B can be calculated using parameters including the positions of the second vehicle 2B, the third vehicle 2C, the fourth vehicle 2D, the position of the second gap 3B, and the speed of the third vehicle 2C.

[0067] Furthermore, when targeting the third gap 3C, the vehicles and gaps corresponding to each parameter can be shifted two positions backward from the case of FIG. 6. In this case, however, position information for a fifth vehicle (not shown) located further behind the fourth vehicle 2D is required. Note that it is possible that the fifth vehicle does not exist, in which case the position of the fifth vehicle can be set under appropriate assumptions. For example, it is conceivable to position the fifth vehicle behind the fourth vehicle in a statistically most likely relationship based on the positioning pattern of the group of vehicles ahead of the fifth vehicle (first to fourth vehicles 2A to 2D).

[0068] (3) Driving behavior model 7 shows the equations of the driving behavior model M3 used in the above simulation. The driving behavior model M3 shown in this figure includes an own vehicle model equation F1 that predicts the behavior of the own vehicle 1, and an other vehicle model equation F2 that predicts the behavior of the other vehicle 2.

[0069] The vehicle model formula F1 is a model formula that outputs acceleration ACCx proportional to the difference between the target position and the target speed, and expresses a proportional model such as a spring mass model. In this vehicle model formula F1, Δx is the difference between the target position and the current position, and Δv is the difference between the target speed and the current speed. These Δx and Δv are variables that change with time. Also, a max is the upper limit acceleration, a min is the lower limit acceleration, and c1 and c2 are weighting coefficients. max ,a min , c1, and c2 are parameters that are determined in advance. Note that the weighting coefficients c1 and c2 may be changed as appropriate depending on the situation of the host vehicle 1 or the other vehicle 2, etc.

[0070] The acceleration ACCx of the host vehicle 1 calculated by such host vehicle model formula F1 is basically calculated as the sum of a value proportional to the difference (Δx) between the target position and the current position and a value proportional to the difference (Δv) between the target speed and the current vehicle speed. However, the acceleration ACCx includes a lower limit acceleration a min not fall below the upper limit acceleration a max A limiter is applied to prevent it from exceeding this.

[0071] The other vehicle model formula F2 is a model formula that outputs the acceleration ACCy required to travel while maintaining an appropriate distance from the preceding vehicle, and represents a following driving model generally known as the IDM (Intelligent Driver Model). In this other vehicle model formula F2, s is the distance from the preceding vehicle, v is the speed, and Δv is the relative speed from the preceding vehicle. These s, v, and Δv are variables that change over time. Also, v0 is the desired speed, s0 is the safe distance from the preceding vehicle (in other words, the distance between vehicles when stopped), T is the minimum time between vehicles (reaction time), a is the maximum acceleration, b is the maximum deceleration, and δ is the power of the acceleration term (smoothness of acceleration). These v0, s0, T, a, b, and δ are characteristic parameters that can change depending on the driver's characteristics and the vehicle's performance.

[0072] Simply put, this other vehicle model formula F2 (follow-up driving model) calculates the appropriate inter-vehicle distance (s * ) and calculates the acceleration ACCy from the calculated inter-vehicle distance. As a result, the acceleration of the other vehicle 2 is adjusted so as to maintain an appropriate inter-vehicle distance from the preceding vehicle.

[0073] (4) Details of regression model generation Next, a procedure for the model generation device 30 to generate the regression model M2 will be described. FIG. 8 is a flowchart showing the procedure for generating the regression model M2. When the control shown in this figure starts, the initial condition setting unit 32 of the model generation device 30 selects initial conditions for the simulation (step S1). The initial conditions selected here are initial conditions that represent the traffic environment (road shape and vehicle placement pattern) at the start of a simulation that predicts the behavior of the host vehicle 1 and the other vehicle group 2 (S13 in FIG. 9, which will be described later), and are selected from the initial condition database M4. That is, the initial condition setting unit 32 reads out an appropriate set of data from various road shape data and vehicle placement data stored as initial conditions in the initial condition database M4, and selects the read data as the initial conditions for the simulation.

[0074] Next, the initial condition setting unit 32 adds noise to the initial conditions selected in step S1 (step S2). That is, the initial condition setting unit 32 adds noise to each of the road shape data and the vehicle location data read from the initial condition database M4. The targets of the noise addition are, for example, the length of the branch line 102 included in the road shape data and the positions and velocities of the host vehicle 1 and the other vehicle group 2 included in the vehicle location data. The method of adding noise is not particularly limited, but it is possible to add noise by multiplying the values ​​by a coefficient that randomly changes within a predetermined range using an appropriate random number generator. By adding noise using this method, the initial condition setting unit 32 slightly changes the length of the branch line 102 and the positions and velocities of the host vehicle 1 and the other vehicle group 2 from the values ​​stored in the initial condition database M4.

[0075] The noise addition may be performed multiple times. That is, the noise addition in step S2 may be performed multiple times for the initial condition selected in step S1. In this way, a set of multiple initial conditions that vary within an appropriate range around the selected initial condition can be created.

[0076] Next, the simulation unit 33 of the model generation device 30 specifies one gap 3 in the group of other vehicles 2 based on the vehicle arrangement pattern of the initial conditions set through steps S1 and S2 (step S3). For example, as shown in FIG. 2, in a pattern in which four other vehicles 2 consisting of first to fourth vehicles 2A to 2D are arranged, the gaps that are candidates for designation are the first gap 3A between the first vehicle 2A and the second vehicle 2B, the second gap 3B between the second vehicle 2B and the third vehicle 2C, and the third gap 3C between the third vehicle 2C and the fourth vehicle 2D. In this case, the simulation unit 33 specifies an appropriate gap 3 from these three gaps 3A to 3C.

[0077] Next, the simulation unit 33 and evaluation value calculation unit 34 of the model generation device 30 evaluate the designated gap 3, which is the gap 3 specified in step S3 (step S4). That is, the simulation unit 33 and evaluation value calculation unit 34 simulate the behavior of the host vehicle 1 and the other vehicle group 2 when the host vehicle 1 cuts into the designated gap 3 using the driving behavior model M3, and evaluate the designated gap 3 based on the results of the simulation.

[0078] 9 is a subroutine showing the detailed procedure for evaluating the specified gap 3 in step S4 above. When the control shown in this figure starts, the simulation unit 33 sets a target position and target speed of the host vehicle 1 (step S11). The target position and target speed are the position and speed that the host vehicle 1 aims to achieve in the host vehicle model formula F1 of the driving behavior model M3 described above, and are determined according to the specified gap 3 specified in step S3 above (FIG. 8).

[0079] For example, as shown in FIG. 11A, if the designated gap 3 is a first gap 3A between a first vehicle 2A and a second vehicle 2B, the target position is set to the side of the first gap 3A. Also, as shown in FIG. 11B, if the designated gap 3 is a third gap 3C between a third vehicle 2C and a fourth vehicle 2D, the target position is set to the side of the third gap 3C. In the former case (FIG. 11A), the target position is set in front of the host vehicle 1, so the host vehicle 1 moves toward the target position while accelerating. On the other hand, in the latter case (FIG. 11B), the target position is set behind the host vehicle 1, so the host vehicle 1 moves toward the target position while suppressing acceleration (or decelerating in some cases).

[0080] The target position of the vehicle 1 may be generally to the side of the designated gap 3 designated in step S3, and various specific methods for setting the target position are conceivable. For example, the target position may be to the side of the center of the designated gap 3, or to the side of another vehicle 2 (leading vehicle) located in front of the designated gap 3.

[0081] The target speed of the host vehicle 1 is set based on the speeds of the other vehicles 2 that make up the designated gap 3, in other words, the speeds of the other vehicles 2 before and after the designated gap 3. For example, if the designated gap 3 is a first gap 3A as shown in FIG. 11A, the target speed of the host vehicle 1 is set based on the speeds of the first vehicle 2A in front of the first gap 3A and the second vehicle 2B behind the first gap 3A. There are various specific setting methods, but for example, the average speed of the first vehicle 2A and the second vehicle 2B may be set as the target speed of the host vehicle 1. This is also true when the designated gap 3 is another gap (the second gap 3B or the third gap 3C).

[0082] Next, the simulation unit 33 randomly sets the model parameters of the other vehicle model formula F2 described above that predicts the behavior of the other vehicle 2 (step S12). Specifically, in this embodiment, four parameters included in the other vehicle model formula F2, namely, the desired speed (v0), the minimum inter-vehicle time (T), the maximum acceleration (a), and the maximum deceleration (b), are randomly set based on statistical occurrence frequency distribution data. On the other hand, the remaining parameters, namely, the exponent (δ) of the acceleration term and the minimum inter-vehicle distance (s0), are maintained at fixed values. This is because it is known that sufficient accuracy can be obtained even if these two parameters are treated as fixed values.

[0083] Here, the statistical occurrence frequency distribution data that forms the basis for random setting is statistical data that defines the relationship between parameter values ​​and occurrence frequencies, as shown in FIG. 12, and is obtained in advance by observing actual vehicles traveling at merging sections on expressways, etc. The four model parameters described above depend on the characteristics of the driver of the other vehicle 2 and the performance of the other vehicle 2, and therefore cannot be known by the own vehicle 1. Therefore, in step S14, the model parameters are randomly set based on the occurrence frequency distribution data of FIG. 12 obtained from observations of actual traffic. This is equivalent to statistically inferring the driver characteristics of the other vehicle 2 from the results of observations of actual traffic. Note that there are already several publicly available occurrence frequency distribution data (statistical data) such as that shown in FIG. 12, and it is possible to obtain them without independently observing actual traffic.

[0084] For example, suppose the occurrence frequency distribution data in FIG. 12 is data on the occurrence frequency of each parameter in the range from W1 to W2, and the most frequent value is Wx. In this case, the simulation unit 33 randomly sets the model parameters within the range from W1 to W2 and varies the occurrence probability based on the distribution curve. That is, the model parameter value generated by the random setting is most likely to be Wx. Conversely, the further a value is from Wx, the less likely it is to be set as a model parameter. Thus, in step S12, the simulation unit 33 sets the model parameters with probabilities according to the distribution curve in FIG. 12. Therefore, when step S12 is repeated and model parameters are randomly set multiple times, the occurrence frequency of the model parameters will be higher as they are closer to the most frequent value Wx in FIG. 12 and lower as they are further from the most frequent value Wx.

[0085] Furthermore, the random setting of the model parameters in step S12 above is performed individually for each other vehicle 2. That is, the random setting of the model parameters for one other vehicle 2 is performed independently of the random setting of the model parameters for the other other vehicles 2. For this reason, in most cases, different model parameters will be set for each other vehicle 2.

[0086] As will be described later, the random setting of the model parameters in step S12 is repeated until a predetermined number of sampling times N1 (S14) is reached, and the model parameters set each time are used in the next merging simulation (S13). In other words, the values ​​of the model parameters used in the merging simulation vary as the random setting in step S12 is repeated. From this, it can be said that the processing in step S12 is processing for imparting fluctuations to the model parameters based on statistical occurrence frequency distribution data.

[0087] When the random setting of the model parameters of the other vehicle model formula F2 is completed as described above, the simulation unit 33 performs a merging simulation using the set model parameters (step S13). The merging simulation is a process of model predicting the future behavior of each vehicle 1, 2 in a merging scene where the host vehicle 1 cuts in to a group of other vehicles 2. In this merging simulation, a simulation is performed using the driving behavior model M3 including the host vehicle model formula F1 and the other vehicle model formula F2 described above. At this time, the parameters randomly set in step S12 above are used as the model parameters (v0, T, a, b) of the other vehicle model formula F2.

[0088] 10 is a subroutine showing the detailed procedure for performing the merging simulation in step S13. When the control shown in this figure starts, the simulation unit 33 determines the variables of each of the model formulas F1 and F2 from the initial positions and initial velocities of the host vehicle 1 and the other vehicle group 2, and the target position of the host vehicle 1 (step S21).

[0089] Specifically, the simulation unit 33 determines, for each other vehicle 2, the values ​​of the variables of the other vehicle model formula F2, namely, the speed (v), the relative speed (Δv) from the preceding vehicle, and the inter-vehicle distance (s) from the preceding vehicle, from the initial position and initial speed of the other vehicle group 2, i.e., the positions and speeds of the other vehicle group 2 set as initial conditions in steps S1 and S2 (FIG. 8). For example, for the second vehicle 2B shown in FIG. 2, the simulation unit 33 determines the speed of the second vehicle 2B set in steps S1 and S2 as the speed (v) of the other vehicle model formula F2. The simulation unit 33 also calculates the relative speed and inter-vehicle distance of the second vehicle 2B with respect to the first vehicle 2A from the positions and speeds of the first vehicle 2A and second vehicle 2B set in steps S1 and S2, and determines the calculated values ​​as the relative speed (Δv) and inter-vehicle distance (s) of the other vehicle model formula F2. The simulation unit 33 performs similar processing for the other other vehicles 2. Note that for the first vehicle 2A, the above variables can be determined using information on other vehicles (not shown) located further ahead. However, if there are no other vehicles ahead of the first vehicle 2A, the above variables can be determined under an appropriate assumption, such as, for example, that there is another vehicle sufficiently ahead of the first vehicle 2A traveling at the same speed as the desired speed (v0) of the first vehicle 2A.

[0090] Furthermore, the simulation unit 33 determines the difference (Δx) between the target position and the current position and the difference (Δv) between the target speed and the current speed, which are variables of the host vehicle model formula F1, based on the position (initial position) and speed (initial speed) of the host vehicle 1 set as initial conditions in steps S1 and S2 above, and the target position and target speed of the host vehicle 1 set in step S11 (FIGS. 11A and 11B). For example, the simulation unit 33 determines the difference (Δx) between the target position and the current position in the host vehicle model formula F1 based on the position of the host vehicle 1 set in steps S1 and S2 above, and the target position of the host vehicle set in step S11. Furthermore, the simulation unit 33 determines the difference (Δv) between the target speed and the current speed in the host vehicle model formula F1 based on the speed of the host vehicle 1 set in steps S1 and S2 above, and the target speed of the host vehicle set in step S11.

[0091] Next, the simulation unit 33 calculates the acceleration ACCy of the other vehicle group 2 from the other vehicle model formula F2 (step S22). That is, the simulation unit 33 performs calculations for each other vehicle 2 using the other vehicle model formula F2 using the variables (v, Δv, s) determined for each other vehicle 2 in step S21 above, thereby calculating the acceleration ACCy of each other vehicle 2. At this time, the values ​​randomly set for each other vehicle 2 in step S14 above are used as the model parameters (v0, T, a, b) of the other vehicle model formula F2.

[0092] Next, the simulation unit 33 calculates the position and speed of each of the other vehicles 2 at the time when the small time Δt has elapsed, using the acceleration ACCy calculated for each other vehicle 2 in step S22 (step S23). That is, the simulation unit 33 calculates the position and speed of each other vehicle 2 after Δt has elapsed, based on the current position and speed of each other vehicle 2 obtained in the previous step and the acceleration ACCy of each other vehicle 2 calculated in step S22.

[0093] Next, the simulation unit 33 calculates the acceleration ACCx of the other vehicle group 2 from the host vehicle model formula F1 (step S24). That is, the simulation unit 33 performs calculations using the host vehicle model formula F1 using the variables (Δx, Δv) determined for the host vehicle 1 in step S21, thereby calculating the acceleration ACCx of the host vehicle 1.

[0094] Next, the simulation unit 33 calculates the position and speed of the host vehicle 1 at the time when the small time Δt has elapsed, using the acceleration ACCx of the host vehicle 1 calculated in step S24 (step S25). That is, the simulation unit 33 calculates the position and speed of the host vehicle 1 after Δt has elapsed, based on the current position and speed of the host vehicle 1 obtained in the previous step and the acceleration ACCx of the host vehicle 1 calculated in step S24.

[0095] Next, the simulation unit 33 updates the positions and velocities of the host vehicle 1 and the other vehicle group 2, as well as the target position and target speed of the host vehicle 1 (step S26). That is, the simulation unit 33 updates the positions and velocities of the host vehicle 1 and the other vehicle group 2 to the positions and velocities after Δt has elapsed, calculated in steps S23 and S25 above. The simulation unit 33 also updates the target position and target speed of the host vehicle 1 based on the updated position and speed of the other vehicle group 2.

[0096] Next, the simulation unit 33 determines whether or not the merging possible condition is satisfied (step S27). The merging possible condition is a condition under which the vehicle 1 can start changing lanes, moving laterally from the branch lane 102 to the main lane 101. Once the merging possible condition is satisfied, the merging can be completed by simply changing lanes, and therefore, it can be considered that the completion of the merging is expected at this point. In other words, determining whether or not the merging possible condition is satisfied is equivalent to determining whether or not the completion of the merging is expected.

[0097] Specifically, the confluence enabling conditions may include, for example, the following requirements (i) to (iv). (i) The vehicle 1 is located near the target position set on the side of the designated gap 3. For example, the vehicle 1 is located within a range of ±5 m of the target position. (ii) The vehicle 1 is traveling at a speed close to the target speed set so that it can travel alongside other vehicles 2 before and after the designated gap 3. For example, the vehicle 1 is traveling within a range of ±5 km / h of the target speed. (iii) The above requirements (i) and (ii) continue for a predetermined period of time (for example, approximately 3 seconds) or more. (iv) The distance between vehicles in designated gap 3 is sufficient. For example, the distance between vehicles in designated gap 3 is a distance equivalent to a time gap of 1.5 seconds or more.

[0098] If the determination in step S27 above is NO, confirming that the merging condition is not met, the simulation unit 33 updates the variables of the model formulas F1 and F2 described above (step S28). That is, the simulation unit 33 calculates the variables (v, Δv, s) of the other vehicle model formula F2 and the variables (Δx, Δv) of the host vehicle model formula F1, respectively, from the positions and velocities of the host vehicle 1 and the other vehicle group 2 updated in step S26 above and the target position and target velocity of the host vehicle 1, in the same manner as in step S21 above. Then, using the calculated values ​​as the updated variables, the simulation unit 33 performs the calculations in steps S22 to S25 above to calculate the positions and velocities of the host vehicle 1 and the other vehicle group 2 after Δt has elapsed. The simulation unit 33 repeats this calculation to calculate the position and velocity of each vehicle every Δt until the merging condition in step S27 above is met.

[0099] It should be noted that, even if the position of the vehicle 1 approaches the branch line terminal end 102a during the repetition of the above calculations, the calculations continue unless the merging condition (S27) is met at that point. In other words, the calculations indicate that the vehicle 1 may move beyond the branch line terminal end 102a. However, if the merging condition is not met even after the vehicle 1 has significantly passed the branch line terminal end 102a, the calculations may be terminated at that point and the merging condition may be deemed to be met.

[0100] If the determination in step S27 above is YES and it is confirmed that the merging condition is met, the simulation unit 33 ends the above calculations for determining the position and speed of each vehicle every Δt (step S29), and calculates the merging margin distance Lm using the conditions at the time of completion (step S30). That is, the simulation unit 33 calculates the merging margin distance Lm (FIGS. 4A and 4B), which is the distance from the host vehicle 1 to the branch line end 102a, based on the position of the host vehicle 1 on the branch line 102 when the merging condition is met, i.e., the position of the host vehicle 1 on the branch line 102 calculated in the last of the repeated calculations.

[0101] The merging margin distance Lm calculated here can be considered the remaining distance to the branch line terminal end 102a at the time when merging is expected to be completed, and as described above, is a value that can be either positive or negative. That is, if the position of the host vehicle 1 at the time when the merging condition is met is before the branch line terminal end 102a, a positive merging margin distance Lm is calculated. On the other hand, if the position of the host vehicle 1 at the time when the merging condition is met is beyond the branch line terminal end 102a, a negative merging margin distance Lm is calculated. The larger the positive merging margin distance Lm, the more leeway there is for merging. Furthermore, a negative merging margin distance Lm means that merging is not possible using behavior according to the model formula.

[0102] When the calculation of the merging margin distance Lm is completed in the above manner, the simulation unit 33 returns to the flow of FIG. 9 and determines whether the number of times the model parameters of the other vehicle model formula F2 have been randomly set has reached a predetermined number of sampling times N1 (step S14).

[0103] If the determination in step S14 above is NO and it is confirmed that the number of random settings has not reached the sampling number N1, the simulation unit 33 returns to step S12 above and randomly sets new model parameters for the other vehicle group 2. Then, using the randomly set model parameters here, a merging simulation (S13) is performed and the merging margin distance Lm is calculated. The simulation unit 33 repeats this merging simulation (calculation of the merging margin distance Lm) for each condition of the model parameters until the number of random settings reaches the sampling number N1.

[0104] When it is determined as YES in the above step S14 and it is confirmed that the number of times of random setting has reached the sampling number N1, the evaluation value calculation unit 34 obtains the distribution of a large number of merging margin distances Lm obtained by repeating the merging simulation in the above step S13, that is, the distribution of the merging margin distances Lm corresponding to various conditions in which the model parameters are varied (step S15). Specifically, the evaluation value calculation unit 34 obtains the average μ and the standard deviation σ shown in FIG. 13 as the distribution of the merging margin distance Lm. As is well known, the average μ is a value obtained by dividing the sum of the obtained data of the merging margin distances Lm by the number of data. The standard deviation σ is a value indicating the degree of variation with respect to the average value μ, and is represented by the square root of the value (variance) obtained by dividing the sum of the squares of the differences (deviations) between the values of the target data and the average μ by the number of data.

[0105] Next, the evaluation value calculation unit 34 calculates a margin evaluation value Em for the specified gap 3 from the distribution of the merging margin distances Lm obtained in the above step S15, that is, the average μ and the standard deviation σ (step S16). In the present embodiment, the margin evaluation value Em is a value defined by the following formula (1) using the average μ and the standard deviation σ of the merging margin distance Lm.

[0106] Em = μ + k×σ ··· (1) Here, k is a coefficient of -1 or more and +1 or less.

[0107] The coefficient k in the above formula (1) may be appropriately set based on values such as the value to be emphasized, and may be a constant value regardless of conditions, or may be a value that changes depending on conditions. In the latter case, as an example, a method of changing the coefficient k according to the degree of margin can be considered. For example, in a condition where careful judgment is required, such as when the average μ of the merging margin distance Lm is relatively small, the coefficient k is made smaller than zero (-1 ≦ k <0), and in a condition where an optimistic judgment is possible, such as when the average μ of the merging margin distance Lm is relatively large, the coefficient k is made larger than zero (0 <k ≦ +1).

[0108] 14 is a graph showing the range of the margin evaluation value Em obtained by the above formula (1). As shown in this figure, the margin evaluation value Em is maximum Emb (= μ + σ) when the coefficient k is +1, and is minimum Ema (= μ - σ) when the coefficient k is -1. In this way, the margin evaluation value Em is calculated so as to vary within the range from Ema to Emb depending on the setting of the coefficient k.

[0109] When the calculation of the margin evaluation value Em is completed in the above manner, the evaluation value calculation unit 34 returns to the flow of Fig. 8 and determines whether or not the evaluation of all gaps 3 is completed (step S5). For example, if there are a total of three gaps 3 in the other vehicle group 2 (first to third gaps 3A to 3C) as in the example of Fig. 2, the evaluation value calculation unit 34 determines whether or not the calculation of the margin evaluation value Em (Fig. 9) for all of these three gaps 3A to 3C is completed.

[0110] If the determination in step S5 above is NO and it is confirmed that unevaluated gaps 3 remain, the evaluation value calculation unit 34 designates the next gap 3 to be evaluated (step S6). Then, the designated gap (designated gap) 3 is newly evaluated (S4) to determine the margin evaluation value Em of the designated gap 3. The evaluation value calculation unit 34 repeats the designation and evaluation of gaps 3 (S4, S6) until such evaluation (calculation of margin evaluation values ​​Em) has been performed for all gaps 3.

[0111] If the determination in step S5 above is YES and it is confirmed that the evaluation of all gaps 3 has been completed, the initial condition setting unit 32 determines whether or not the selection of all initial conditions has been completed (step S7). That is, the initial condition setting unit 32 determines whether or not all initial conditions stored in the above-mentioned initial condition database M4 have been selected as initial conditions for the merging simulation.

[0112] If the determination in step S7 above is NO and it is confirmed that unselected initial conditions remain, the initial condition setting unit 32 selects the next initial condition (step S8). The selected initial condition is used as the initial condition for a new merging simulation, and the results of the merging simulation are used to evaluate each gap 3 (calculate the margin evaluation value Em). The initial condition setting unit 32 repeats the selection of initial conditions in the initial condition database M4 until the simulation and evaluation have been performed starting from all the initial conditions in the initial condition database M4.

[0113] If the determination in step S7 above is YES and it is confirmed that the selection of all initial conditions has been completed, the regression analysis unit 35 of the model generation device 30 extracts a data set of the initial conditions and the headroom evaluation values ​​Em (step S9). That is, the regression analysis unit 35 extracts all of the many headroom evaluation values ​​Em (S16) calculated through the repeated merging simulations (S13) in association with the initial conditions (S1, S2) of each merging simulation, thereby creating a data set that collects data pairs (data sets) of headroom evaluation values ​​Em and their corresponding initial conditions, equal to the number of initial conditions.

[0114] Next, the regression analysis unit 35 generates a regression model M2 by performing multiple regression analysis using the set of data extracted in step S9 as training data (step S10). As already explained with reference to Fig. 5, the regression model M2 is a model that uses the feature quantities Q1 to Q5, Qp specific to each gap 3 as explanatory variables and the margin evaluation value Em of each gap 3 as a response variable.

[0115] Through the above-described procedure, a regression model M2 is generated for evaluating the relative merits of each gap 3 as a merging position. The generated regression model M2 is stored in the storage unit 17 of the controller 10 of the host vehicle 1.

[0116] (5) Control during driving Next, details of the control performed by the controller 10 in an actual merging scene in which the host vehicle 1 cuts in to a group of other vehicles 2 will be described. FIG. 15 is a flowchart showing a control procedure performed by the controller 10 when merging. The control shown in this figure starts assuming that the host vehicle 1 is traveling on the branch line 102 (FIG. 2) and that multiple other vehicles 2 are traveling within the range of the main line 101 that the host vehicle 1 can recognize. When this control starts, the recognition unit 12 of the controller 10 acquires the traveling state of the host vehicle 1 and the traffic environment around the host vehicle 1 (step S41). That is, the recognition unit 12 acquires information including the position and speed of the host vehicle 1 as information representing the traveling state of the host vehicle 1. The recognition unit 12 also acquires information including the road shape around the host vehicle 1 and the position and speed of the other vehicles 2 as information representing the traffic environment around the host vehicle 1. This information is determined from the detection values ​​of the internal sensor group 20 and the external sensor group 21, the road map data M1 stored in the memory unit 17, and the received signal of the GNSS receiver 22.

[0117] More specifically, the recognition unit 12 acquires information including the speed of the vehicle 1 and the position of the vehicle 1 on the branch line 102 based on the detection values ​​of the internal sensor group 20 and the received signal of the GNSS receiver 22. The recognition unit 12 also acquires information about the road shape around the vehicle 1 based on the acquired position (current position) of the vehicle 1 and detailed road map data M1 stored in the memory unit 17. This information may include information about the acceleration section length, which is the distance from the current position of the vehicle 1 traveling on the branch line 102 to the branch line end 102a, and information about lane change prohibited sections, which are sections where lane changes from the branch line 102 to the main line 101 are prohibited. If a lane change prohibited section exists, the position of the branch line end 102a may be set to the end of the branch line 102 excluding the lane change prohibited section, in other words, the end of the section where lane changes from the branch line 102 to the main line 101 are permitted. Furthermore, the recognition unit 12 acquires information including the position and speed of each of a plurality of other vehicles 2 (other vehicle group) traveling on the main road 101 in a range from diagonally ahead to diagonally behind the host vehicle 1, based on the detection values ​​of the external sensor group 21. In this case, the acquired positions of the other vehicle group 2 are relative positions with the host vehicle 1 as the reference.

[0118] Next, the recognition unit 12 extracts gaps 3 that are candidates for a cut-in destination for the host vehicle 1, i.e., a merging position, based on the information on the other vehicle group 2 acquired in step S41 (step S42). For example, if the other vehicle group 2 recognized in step S41 is four vehicles, first to fourth vehicles 2A to 2D, as shown in Fig. 2, the recognition unit 12 extracts three gaps, first to third gaps 3A to 3C, as candidates for the merging position.

[0119] Next, the evaluation unit 13 of the controller 10 calculates a basic feature quantity Q of the regression model M2 (FIG. 5) for each gap 3 extracted in step S42 (step S43). As described above, the basic feature quantity Q used in this embodiment is a value calculated directly from the state at the start of merging without going through prediction, and includes five feature quantities Q1 to Q5 shown in FIG. 6. That is, based on the information about the host vehicle 1 and the group of other vehicles 2 and the information about the road shape acquired in step S41 above as the traffic environment (initial condition) at the start of merging, the evaluation unit 13 calculates the tentative inter-vehicle distance Q1, the tentative inter-vehicle time Q2, the risk perception index Q3, the inter-vehicle distance Q4, and the remaining time Q5 to the branch line end 102a for each gap 3, and adopts each calculated value as the basic feature quantity Q.

[0120] For example, when calculating the basic feature quantity Q of the first gap 3A as in the case of Fig. 6, the evaluation unit 13 identifies the positions p1, p2, and p3 of the first to third vehicles 2A to 2C, the position pe of the branch line terminal 102a, the speed v0 of the host vehicle 1, and the speed v2 of the second vehicle 2B based on the various information at the start of merging acquired in step S41. Then, based on the identified values, the evaluation unit 13 calculates the above-mentioned feature quantities Q1 to Q5 (basic feature quantities Q) related to the first gap 3A. The evaluation unit 13 also calculates the feature quantities Q1 to Q5 in the same manner for the other gaps 3 (the second gap 3B and the third gap 3C).

[0121] Next, the evaluation unit 13 calculates a predicted feature value Qp of the regression model M2 for each gap 3 extracted in step S42 (step S44). As described above, the predicted feature value Qp used in this embodiment is the merging margin distance Lm (FIGS. 4A and 4B) calculated based on a future state predicted through a simulation using the driving behavior model M3. That is, the evaluation unit 13 predicts the behavior of the host vehicle 1 and the other vehicle 2 when the host vehicle 1 cuts into each gap 3 through a simulation using the driving behavior model M3 stored in the storage unit 17. Then, the evaluation unit 13 calculates a merging margin distance Lm from the predicted state and uses it as the predicted feature value Qp.

[0122] For example, when calculating the predicted feature quantity Qp of the first gap 3A as in the case of Fig. 6, the evaluation unit 13 starts from the state at the start of merging acquired in step S41 and simulates the behavior of the host vehicle 1 and the other vehicle group 2 when the host vehicle 1 enters the first gap 3A using the driving behavior model M3. At this time, the evaluation unit 13 adopts statistically most likely values, i.e., mode values, such as the parameter value Wx shown in Fig. 12, as the model parameters (v0, T, a, b) included in the other vehicle model formula F2 of the driving behavior model M3 (Fig. 7). Then, a one-time simulation is performed under the conditions adopting the most mode model parameters in this manner, and based on the result, the evaluation unit 13 calculates the merging margin distance Lm when the host vehicle 1 enters the first gap 3A, that is, the merging margin distance Lm, which is the remaining distance to the branch line end 102a when the merging possible condition is met (when merging is expected to be completed). The evaluation unit 13 also performs the same calculation of the merging margin distance Lm for the other gaps 3 (the second gap 3B and the third gap 3C).

[0123] As described above, the evaluation unit 13 performs a simulation once for each gap 3 under the condition that the model parameters of the driving behavior model M3 are fixed, and adopts the merging margin distance Lm calculated for each gap 3 from the result as the prediction feature quantity Qp. In other words, the prediction feature quantity Qp (merging margin distance Lm) of each gap 3 is an index of the same type as the evaluation value (margin evaluation value Em), which is calculated from the future state predicted by a single simulation under certain conditions from the state at the start of merging (specifically, the positional relationship between the host vehicle 1 and the branch end 102a when the merging enable condition is satisfied).

[0124] Next, the evaluation unit 13 identifies the margin evaluation value Em of each gap 3 based on the basic feature quantity Q and the prediction feature quantity Qp calculated in the above steps S43 and S44 (step S45). That is, the evaluation unit 13 inputs the basic feature quantity Q (Q1 to Q5) and the prediction feature quantity Qp calculated for each gap 3 in the above steps S43 and S44 into the regression model M2 as explanatory variables. In response to this, the regression model M2 outputs the margin evaluation value Em of each gap 3 as the target variable. The evaluation unit 13 identifies the margin evaluation value Em of each gap 3 through the analysis using such a regression model M2.

[0125] When the identification of the margin evaluation value Em of each gap 3 is completed as described above, the evaluation unit 13 specifies the gap 3 with the highest margin evaluation value Em (step S46).

[0126] FIG. 16 is a graph showing, as an example, the difference in the margin evaluation value Em when there are three gaps 3 that are candidates for the merging position, namely candidates 1 to 3. When the margin evaluation value Em of candidate 1 is Em1, the margin evaluation value Em of candidate 2 is Em2, and the margin evaluation value Em of candidate 3 is Em3, in the example of FIG. 16, the relationship Em1 < Em3 < Em2 holds. In this case, the evaluation unit 13 specifies candidate 2 corresponding to the highest margin evaluation value Em2.

[0127] When the graph in Figure 16 is associated with the example layout of each vehicle shown in Figure 2, candidate 1 may correspond to the first gap 3A, candidate 2 may correspond to the second gap 3B, and candidate 3 may correspond to the third gap 3C. Of course, other correspondences are possible depending on the situation.

[0128] Furthermore, in step S46, if there are two or more candidates with the highest headroom evaluation value Em, one of the candidates may be narrowed down by an appropriate method and identified as the candidate with the highest evaluation. Various methods for narrowing down the candidates may be considered, for example, a method of narrowing down the candidates to those closest to the vehicle 1 at the time of evaluation may be considered.

[0129] Next, the route generation unit 14 determines the gap 3 identified in step S46 above, i.e., the gap 3 with the highest margin evaluation value Em, as the merging target for the host vehicle 1 (step S47). For example, if a gap of candidate 2 is identified in step S46 above and corresponds to the second gap 3B shown in Fig. 2, the route generation unit 14 determines the second gap 3B as the merging target. As a result, the host vehicle 1 will travel along the travel route to the second gap 3B and will enter the second gap 3B.

[0130] (6) Effects As described above, in this embodiment, the regression model M2 is used as a tool for evaluating the merits and demerits of gaps 3 that are candidate merging locations. The regression model M2 is generated in advance through a simulation using the driving behavior model M3. Specifically, a margin evaluation value Em, which indicates how much margin a vehicle has available for merging, is calculated each time through a simulation starting from various initial conditions with different traffic environments (road shapes and vehicle placement patterns). The regression model M2 is generated based on a set of data sets of the calculated margin evaluation values ​​Em and the corresponding initial conditions. Furthermore, in an actual merging scene, multiple feature values ​​Q and Qp, which are explanatory variables of the regression model M2, are acquired for each gap 3. The acquired feature values ​​Q and Qp for each gap 3 are input into the regression model M2 to identify the margin evaluation value Em for each gap 3. Then, based on the identified margin evaluation value Em, it is determined which gap 3 should be used as a merging target. In particular, in this embodiment, one of the feature quantities (explanatory variables) input to the regression model M2 is a predicted feature quantity Qp, which is an index (merging margin distance Lm) similar to the margin evaluation value Em, predicted under certain conditions from the state at the start of merging. This configuration has the advantage of enabling accurate evaluation of each gap 3 using a relatively simple method with little processing load.

[0131] That is, in this embodiment, a simulation for predicting behavior at the time of merging is performed using a predetermined driving behavior model M3 starting from various initial conditions, and a regression model M2 for evaluating the merging position is generated based on a set of data sets of the initial conditions and evaluation values ​​(margin evaluation values ​​Em) obtained from the results. Therefore, the regression model M2 can be generated relatively easily without undergoing complex processing such as reinforcement learning.

[0132] Furthermore, in an actual merging scene, the unique feature quantities Q and Qp corresponding to each gap 3 of the group of other vehicles 2 are acquired, and the feature quantities Q and Qp are input into the regression model M2 generated in advance to identify a margin evaluation value Em that indicates the suitability of each gap 3 as a merging location. In this way, the margin evaluation value Em for each gap 3 can be obtained simply by inputting the feature quantities Q and Qp for each gap 3 into the regression model M2. Therefore, it is possible to determine in a short time which gap 3 should be set as the merging target, and to appropriately determine an action plan even in an actual merging scene where it is difficult to secure sufficient calculation time.

[0133] Furthermore, in this embodiment, a predicted feature Qp, which is the same type as the margin evaluation value Em and is predicted from the state at the start of merging, is used as one of the feature quantities (explanatory variables) input to the regression model M2, thereby improving the regression accuracy. Specifically, research by the present inventors has shown that the regression accuracy of the regression model M2 is improved when the predicted feature Qp, which represents the future state some time after the start of merging, is included as an explanatory variable. In particular, it has been shown that using an index similar to the evaluation value output by the regression model M2 as the predicted feature Qp leads to improved regression accuracy. Therefore, according to this embodiment, in which the predicted feature Qp, which is an index similar to the margin evaluation value Em (merging margin distance Lm) as described above, is used as an explanatory variable, the regression accuracy of the regression model M2 can be improved, and the margin evaluation value Em for each gap 3 can be accurately identified.

[0134] In this embodiment, the above simulation for generating the regression model M2 calculates the merging margin distance Lm, which is the remaining distance to the branch line end 102a at the time when the host vehicle 1 is expected to complete merging (when the merging condition is met), and calculates the margin evaluation value Em based on the distribution (mean μ and standard deviation σ) of the merging margin distance Lm obtained through repeated simulations. With this configuration, an appropriate gap 3 that allows merging with ample margin can be selected as the merging target.

[0135] The merging margin distance Lm, which is the remaining distance to the branch line end 102a, is a value that can continuously change depending on the situation, and therefore can be used not only as an indicator for determining success / failure, but also as an indicator for estimating how much margin is available for merging. Therefore, according to this embodiment, in which each gap 3 is evaluated using a margin evaluation value Em based on such merging margin distance Lm, it is possible to appropriately determine how much margin is expected when each gap 3 is set as a merging target, and to appropriately determine a merging target from among each gap 3. Moreover, because the margin evaluation value Em is calculated based on the distribution of the merging margin distance Lm, a statistically valid gap 3 can be determined as the merging target.

[0136] Furthermore, in this embodiment, a merging margin distance Lm is calculated as the predicted feature value Qp to be input to the regression model M2 at the time of merging. Specifically, a simulation is performed under certain conditions in which the model parameters (v0, T, a, b) included in the other vehicle model formula F2 of the driving behavior model M3 are fixed, and the merging margin distance Lm is calculated for each gap 3 based on the results. With this configuration, the predicted feature value Qp can be easily calculated by a simple simulation using the same model as the driving behavior model M3 prepared for generating the regression model M2.

[0137] Moreover, the merging margin distance Lm used as the predicted feature quantity Qp in this case can be said to be an index of the same kind as the margin evaluation value Em, which is an evaluation value based on the distribution of the merging margin distance Lm. Therefore, according to this embodiment, which employs the predicted feature quantity Qp with such properties as one of the explanatory variables, it is possible to improve the regression accuracy as described above and more appropriately evaluate each gap 3.

[0138] Furthermore, in this embodiment, in addition to the predicted feature quantity Qp described above, a basic feature quantity Q, which is calculated directly from the state at the start of merging without going through prediction, is input to the regression model M2. Research by the inventors of the present application has shown that using a combination of the predicted feature quantity Qp, which represents the future state, and the basic feature quantity Q, which represents the state at the start of merging, as explanatory variables is more effective in improving regression accuracy. Therefore, according to this embodiment, which uses a combination of the predicted feature quantity Qp and the basic feature quantity Q, it is possible to further improve regression accuracy.

[0139] Furthermore, in this embodiment, a plurality of feature quantities Q1 to Q5 are calculated as the basic feature quantity Q from the positions or speeds of the host vehicle 1 and the other vehicle group 2 at the start of merging, the position of the gap 3 in the other vehicle group 2, and the road shape. With this configuration, the basic feature quantity Q (Q1 to Q5) specific to each gap 3 can be appropriately calculated from the state of each vehicle at the start of merging, etc.

[0140] In this embodiment, the driving behavior model M3 includes an other vehicle model formula F2 that predicts the behavior of the other vehicle group 2, and a host vehicle model formula F1 that predicts the behavior of the host vehicle 1. The other vehicle model formula F2 corresponds to a following driving model that outputs acceleration required to drive while maintaining an appropriate inter-vehicle distance from the preceding vehicle, and the host vehicle model formula F1 corresponds to a proportional model that outputs acceleration proportional to the difference between a predetermined target position and a predetermined target speed. With this configuration, the future behavior of the other vehicle group 2 and the host vehicle 1 can be appropriately simulated through relatively light calculations using specified calculation formulas.

[0141] (7) Variations Although the preferred embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and various modifications are possible without departing from the spirit of the present invention.

[0142] For example, in the above embodiment, in order to create training data for regression model M2, the simulation was continued until a predetermined merging condition was met and merging completion was expected, and the distance from the vehicle 1 to the end of the branch line 102a at that time, in other words, the remaining distance to the end of the branch line 102a, was calculated as the merging margin distance Lm. However, the simulation may also be continued until the vehicle 1 actually enters the specified gap 3 of the group of other vehicles 2 and merging is completed, and the remaining distance to the end of the branch line 102a at that time (the time of merging completion) may be calculated as the merging margin distance Lm.

[0143] Furthermore, instead of the merging margin distance Lm as described above, a merging margin time, which is the remaining time until the branch line end 102a at the time when merging has been completed or is expected to be completed, may be calculated by simulation, and each gap 3 may be evaluated based on the merging margin time. In other words, when a value indicating how much margin the vehicle 1 has to merge, that is, a merging margin, is used as the evaluation value of each gap 3, this merging margin may be a value expressed in terms of either distance or time.

[0144] Furthermore, the evaluation value for evaluating the merits of each gap 3 need only be some value related to the success or failure of merging, and is not limited to the merging margin (merging margin distance or time). For example, an evaluation value may be used that integrates multiple index values, including a first index value that indicates whether merging is possible before the branch line end 102a, and a second index value that indicates the degree of burden on the driver, such as the smoothness of acceleration and deceleration.

[0145] Furthermore, in the above embodiment, the merging margin distance Lm was used as the predicted feature quantity Qp, which is one of the explanatory variables input to the regression model M2. However, the predicted feature quantity Qp is not limited to the merging margin distance Lm and may be an index of the same type as the evaluation value output by the regression model M2. For example, if the evaluation value output by the regression model M2 is an evaluation value (margin evaluation value Em) based on the merging margin (merging margin distance Lm) as in the above embodiment, the predicted feature quantity may be an index other than the merging margin distance Lm as long as it indicates how much margin a vehicle has to merge. For example, instead of the merging margin distance, a merging margin time may be used, or some index indicating the amount of margin using a value other than distance or time may be used.

[0146] In the above embodiment, when repeatedly simulating the future behavior of the host vehicle 1 and the other vehicle group 2, variations based on statistical occurrence frequency distribution data are imparted to the model parameters of the driving behavior model M3 (other vehicle model formula F2), but the method of imparting variations to the model parameters is not limited to this. For example, normal distribution data that approximates occurrence frequency distribution data obtained from observations of actual traffic may be prepared, and variations may be imparted based on the normal distribution data. Alternatively, an occurrence frequency table that discretely expresses the relationship between parameter values ​​and occurrence frequencies may be prepared, and interpolation using appropriate random numbers may be performed so that parameter values ​​appear evenly within each discrete section.

[0147] Furthermore, when vehicle type information such as standard vehicles, large vehicles, and motorcycles is available for the other vehicle group 2, the appearance frequency distribution data may be used separately for each vehicle type. In this way, parameter setting that is more in line with reality can be realized.

[0148] In the above embodiment, a following-traveling model (other vehicle model formula F2) is used as a model for predicting the behavior of the other vehicle 2, and a proportional model (subject vehicle model formula F1) is used as a model for predicting the behavior of the host vehicle 1, but the behavior of the host vehicle 1 and the other vehicle 2 may be predicted using the same model. For example, the behavior of both the host vehicle 1 and the other vehicle 2 may be predicted using a following-traveling model. Furthermore, since various models with a similar purpose have already been proposed as the following-traveling model (IDM), it goes without saying that a following-traveling model other than the model formula (other vehicle model formula F2) shown in FIG. 4 can be used as appropriate. [Explanation of symbols]

[0149] 1. Your vehicle 2 Other vehicles (groups of other vehicles) 3. Gap 33 Simulation Department 34 Evaluation value calculation unit 35 Regression Analysis Section 101 Main Line 102 branch line 102a (branch line) termination Em Margin evaluation value (evaluation value) Lm Merging margin distance (merging margin) M2 regression model M3 driving behavior model Q Basic feature Qp predicted feature

Claims

1. A method for evaluating a merging position when a vehicle merges from a branch line into a main line on which a group of other vehicles is traveling, comprising: a simulation step of simulating subsequent behaviors of the host vehicle and the other vehicle group, starting from various initial conditions with different traffic environments, using a driving behavior model that predicts behaviors of the host vehicle and the other vehicle group; an evaluation value calculation step of calculating, for each gap, an evaluation value relating to whether or not the host vehicle will be able to merge into a gap between the other vehicles based on a result of the simulation; a regression model generation step of generating a regression model in which a feature quantity specific to each gap is used as an explanatory variable and the evaluation value of each gap is used as a response variable, based on a set of data sets including the calculated evaluation values ​​and the corresponding initial conditions; a merging position evaluation step of acquiring the feature amount for each of the gaps, inputting the acquired feature amount as the explanatory variable into the regression model, thereby identifying the evaluation value for each of the gaps, and determining which of the gaps should be set as a merging target based on the identified evaluation value, The method for evaluating a merging position for a vehicle, wherein the feature amount includes a predicted feature amount of the same type as the evaluation value, which is predicted under certain conditions from the state at the start of merging.

2. 2. The method for evaluating a merging position for a vehicle according to claim 1, the merging position evaluation step performs the simulation under conditions in which model parameters of the driving behavior model are fixed, and calculates the predicted feature amount for each gap from a result of the simulation.

3. 2. The method for evaluating a merging position for a vehicle according to claim 1, In the evaluation value calculation step, a merging margin, which is the remaining distance or remaining time to the end of the branch line at the time when the vehicle has completed merging or is expected to complete merging, is calculated for each gap, and the evaluation value for each gap is calculated based on the merging margin.

4. 4. The method for evaluating a merging position for a vehicle according to claim 3, The method for evaluating a merging position for a vehicle, wherein the predicted feature amount is the merging margin predicted from a state at the start of merging.

5. The vehicle merging position evaluation method according to any one of claims 1 to 4, The feature amount includes, in addition to the predicted feature amount, a basic feature amount that is calculated directly from a state at the start of merging without going through prediction.

6. The vehicle merging position evaluation method according to any one of claims 1 to 4, A method for evaluating a vehicle's merging position, wherein the driving behavior model includes a following driving model that outputs the acceleration required to drive while maintaining an appropriate inter-vehicle distance from the preceding vehicle as a model for predicting the behavior of the group of other vehicles.

7. 7. The method for evaluating a merging position for a vehicle according to claim 6, The method for evaluating a merging position of a vehicle, wherein the driving behavior model includes a proportional model that outputs acceleration proportional to the difference between a predetermined target position and a predetermined target speed as a model for predicting the behavior of the vehicle.

8. A method for generating a regression model for evaluating a merging position when a vehicle merges from a branch line into a main line on which a group of other vehicles is traveling, comprising: a simulation step of simulating subsequent behaviors of the host vehicle and the other vehicle group, starting from various initial conditions with different traffic environments, using a driving behavior model that predicts behaviors of the host vehicle and the other vehicle group; an evaluation value calculation step of calculating, for each gap, an evaluation value relating to whether or not the host vehicle will be able to merge into a gap between the other vehicles based on a result of the simulation; a regression model generation step of generating a regression model in which a feature amount specific to each gap is used as an explanatory variable and the evaluation value of each gap is used as a response variable, based on a set of data sets including the evaluation value calculated for each gap and the corresponding initial conditions, A model generation method for merging position evaluation, wherein the feature amount includes a predicted feature amount of the same type as the evaluation value, which is predicted under certain conditions from a state at the start of merging.

9. An apparatus for generating a regression model for evaluating a merging position when a vehicle merges from a branch line into a main line on which a group of other vehicles is traveling, comprising: a simulation unit that uses a driving behavior model that predicts the behavior of the host vehicle and the other vehicle group, starting from various initial conditions that differ in traffic environments, and simulates the subsequent behavior of the host vehicle and the other vehicle group; an evaluation value calculation unit that calculates, for each gap, an evaluation value regarding whether or not the host vehicle will be able to merge into a gap between the other vehicles based on a result of the simulation; a regression analysis unit that generates a regression model using a feature amount specific to each gap as an explanatory variable and the evaluation value of each gap as a response variable based on a set of data sets including the evaluation value calculated for each gap and the corresponding initial conditions, The feature quantity includes a predicted feature quantity of the same type as the evaluation value, which is predicted under certain conditions from the state at the start of merging.

Images