Travel route management system and travel route management method

Abstract

To generate a travel route that matches the driving tendencies of each user.SOLUTION: A travel route management device 1 includes a first learning unit 11 that performs adversarial learning of a generative model and having a generator 111 that uses first route information on which a vehicle 2 has traveled in the past as true route information and generates pseudo route information similar to the true route information, and a discriminator 112 that discriminates between the pseudo route information generated by the generator 111 and the true route information, a generation unit 12 that generates pseudo route information using a trained generator 111' constructed by the first learning unit 11, a second learning unit 14 that learns the relationship between second route information for reaching a destination point at the minimum cost and the pseudo route information generated by the trained generator 111' using a discriminative model, and a first storage unit 13 that stores the trained discriminative model constructed by the second learning unit 14.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a travel route management system and a travel route management method. [Background technology]

[0002] Car navigation systems mounted on vehicles and the like have been known for some time. For example, Patent Document 1 discloses a technology for providing route guidance for a vehicle by setting a route that provides the shortest distance from the vehicle's current location to a destination or a route that reaches the destination in the shortest time. Patent Document 1 also sets a route that allows the vehicle to reach the destination more efficiently based on the presence or absence of traffic lights, the presence or absence of intersections, ease of vehicle entry into facilities, and the like.

[0003] Here, a user who is a driver of a vehicle may frequently use shortcuts that differ from the guided roads when traveling along a uniformly set guided route based on the shortest time or distance. However, the route setting by the car navigation system disclosed in Patent Document 1 was unable to generate an individualized driving route that was tailored to the driving tendencies of each user. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2012-220265 Summary of the Invention [Problem to be solved by the invention]

[0005] As described above, according to conventional techniques, it has been difficult to generate a driving route that matches the driving tendencies of each user.

[0006] The present invention has been made to solve the above-mentioned problems, and has an object to generate a driving route that matches the driving tendencies of each user. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, the driving route management device of the present invention includes a first acquisition unit configured to acquire first route information from a first departure point to a first destination point that a vehicle has previously traveled; a first learning unit configured to perform adversarial learning of a generative model having a generator that generates pseudo route information similar to the true route information using the first route information as true route information from the first departure point to the first destination point; and a discriminator that distinguishes between the pseudo route information generated by the generator and the true route information; a generation unit configured to generate the pseudo route information using the trained generator constructed by the first learning unit; a second learning unit configured to learn, using a discriminative model, the relationship between second route information that reaches the first destination point from the first departure point at the minimum cost and the pseudo route information generated by the trained generator; and a memory unit configured to store the trained discriminative model constructed by the second learning unit.

[0008] In addition, the driving route management device of the present invention may further include a second acquisition unit configured to acquire third route information for reaching a second destination point from a second departure point newly set in the vehicle at the minimum cost, and a determination unit configured to provide the third route information as an unknown input to the trained identification model, perform calculations on the trained identification model, output pseudo route information from the second departure point to the second destination point, and determine the output pseudo route information as recommended route information.

[0009] In addition, the driving route management device of the present invention may further include a guidance unit configured to guide the vehicle from the second departure point to the second destination point along the recommended route information determined by the determination unit.

[0010] In the travel route management device according to the present invention, the second route information and the third route information may be route information that allows the vehicle to travel from a departure point to a destination point in the shortest distance or in the shortest time.

[0011] In order to solve the above-mentioned problems, the driving route management method of the present invention includes a first acquisition step of acquiring first route information from a first departure point to a first destination point along which a vehicle has previously traveled; a first learning step of performing adversarial learning of a generative model having a generator that uses the first route information as true route information from the first departure point to the first destination point and generates pseudo route information similar to the true route information, and a discriminator that distinguishes between the pseudo route information generated by the generator and the true route information; a generation step of generating the pseudo route information using the trained generator constructed in the first learning step; a second learning step of learning, using a discriminative model, the relationship between second route information for reaching the first destination point from the first departure point with the least cost and the pseudo route information generated by the trained generator; and a storage step of storing the trained discriminative model constructed in the second learning step in a storage unit.

[0012] Furthermore, the driving route management method according to the present invention may further include a second acquisition step of acquiring third route information for reaching a second destination point from a second departure point newly set in the vehicle at the minimum cost, and a determination step of providing the third route information as an unknown input to the trained discrimination model, performing calculations on the trained discrimination model, outputting pseudo route information from the second departure point to the second destination point, and determining the output pseudo route information as recommended route information.

[0013] In addition, the driving route management method of the present invention may further include a guiding step of guiding the vehicle from the second departure point to the second destination point along the recommended route information determined in the determination step. [Effects of the Invention]

[0014] According to the present invention, the relationship between second route information for reaching the first destination point from the first starting point with the least cost and pseudo route information generated by a trained generator is learned using a discriminative model, so that a driving route tailored to the driving tendencies of each user can be generated. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is a block diagram showing the configuration of a travel route management system including a travel route management device according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram showing the hardware configuration of the travel route management device according to this embodiment. [Figure 3] FIG. 3 is a diagram for explaining the first learning unit included in the travel route management device according to this embodiment. [Figure 4] FIG. 4 is a diagram for explaining the first learning unit included in the travel route management device according to this embodiment. [Figure 5] FIG. 5 is a diagram for explaining the first learning unit included in the travel route management device according to this embodiment. [Figure 6] FIG. 6 is a diagram for explaining the second learning unit included in the travel route management device according to this embodiment. [Figure 7] FIG. 7 is a flowchart showing the operation of the travel route management device according to this embodiment. [Figure 8] FIG. 8 is a flowchart showing the operation of the travel route management device according to this embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0016] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will now be described in detail with reference to FIGS.

[0017] [Configuration of the driving route management system] First, an overview of a travel route management system including a travel route management device 1 according to an embodiment of the present invention will be described with reference to Fig. 1. The travel route management system manages the travel route of a vehicle 2 by generating an optimal travel route that reflects the tendency of the travel route taken by the vehicle 2.

[0018] The travel route management system is provided, for example, in an in-vehicle navigation system of the vehicle 2. The vehicle 2 and the travel route management device 1 are connected via an in-vehicle network NW.

[0019] The vehicle 2 includes automobiles, motor vehicles, motorcycles, etc. that are driven by a driver. The vehicle 2 is equipped with an ECU (Electronic Control Unit) (not shown) and a navigation system with a conventional route search function, and is linked to the driving route management device 1 via an in-vehicle network NW. The navigation system with a conventional route search function calculates the minimum-cost driving route from a set departure point to a destination point. The departure point can be GPS location information indicating the current location of the vehicle 2.

[0020] The minimum-cost driving route to the destination is calculated using conventional route search techniques such as the well-known Dijkstra algorithm. Specifically, as shown in Figure 1, based on the nodes N and links L that make up the map data, links L, which are roads from the starting point "START" to the destination point "GOAL," are searched for, and the cost, which is the evaluation value, is calculated. The sequence of consecutive links L that minimizes the cost to the destination is calculated as minimum-cost route information (second route information, third route information). The minimum-cost route information can also be calculated taking into account factors such as traffic conditions, road type, and speed limits.

[0021] The map data consists of a road network in which roads on a map are represented by links L, and can be provided as section data in which roads are divided into multiple sections based on pre-set nodes N such as intersections and branching points. Each node N has GPS location information consisting of longitude and latitude. Furthermore, the links L that make up each section can include GPS location information consisting of the longitude and latitude of the start and end points, as well as data such as length, road width, and type.

[0022] The route information shown in Figure 1, which is indicated by a dashed line passing through node N on the map data in the order of "node 1 - node 2 - node 3 - node n", is calculated as the driving route to reach the destination point with the minimum cost. In addition to defining the cost as the shortest distance to the destination point, it is also possible to define the cost as the shortest time and calculate the driving route with the shortest time.

[0023] On the other hand, the route information shown by the solid line in Fig. 1, which passes through node N in the order of "node 1 - node 2' - node 3' - node 4' - node 5' - node n", is route information generated by the travel route management system according to this embodiment. The route information generated in this embodiment is route information that reflects the driving tendencies of the driver of vehicle 2. Therefore, the route information shown by the solid line is not necessarily a travel route that reaches the destination in the shortest time or the shortest distance.

[0024] The driving route management system of this embodiment constructs a generator 111 that performs adversarial learning of a GAN (Generative Adversarial Network) using past route information that vehicle 2 has actually traveled from the starting point set in vehicle 2 to the destination point as true route information, and generates pseudo route information that is similar to the true route information.

[0025] Furthermore, using training data in which the pseudo route information generated by the constructed trained generator 111' is used as a training label, the system uses a discriminative model to learn the relationship between the pseudo route information and route information for getting from the departure point to the destination point at the minimum cost, calculated by a navigation system with a conventional route search function, for example. The driving route management system provides the trained discriminative model with the minimum-cost route information from the departure point to the destination point, which is newly set in the vehicle 2, as an unknown input, and determines the output pseudo route information as the recommended route information.

[0026] [Function block of the driving route management device] Next, functional blocks of the driving route management device 1 according to this embodiment will be described with reference to the block diagram of Fig. 1. As shown in Fig. 1, the driving route management device 1 includes a first acquisition unit 10, a first learning unit 11, a generation unit 12, a first memory unit (memory unit) 13, a second learning unit 14, a second acquisition unit 15, a determination unit 16, a guiding unit 17, and a second memory unit 18.

[0027] The first acquisition unit 10 acquires first route information from a starting point (first starting point) to a destination point (first destination point) along which the vehicle 2 has traveled in the past. More specifically, the first acquisition unit 10 acquires, from the history of the travel route of the vehicle 2, information on a plurality of nodes N that the vehicle 2 has passed through from a previously set starting point to the destination point, as the first route information. The first route information is information in which GPS position information is associated with identifiers of a plurality of consecutive nodes N. The first route information is a history of various route information along which the vehicle 2 has actually traveled in the past, and is stored in the second storage unit 18. The first acquisition unit 10 can acquire the first route information from the second storage unit 18, and can also use the history of travel routes for an arbitrarily set period, such as the most recent year, as the first route information.

[0028] The first learning unit 11 performs adversarial learning of a generative model having a generator 111 that generates pseudo route information similar to the true route information, using the first route information as true route information from the departure point (first departure point) to the destination point (first destination point), and a classifier 112 that distinguishes between the pseudo route information generated by the generator 111 and the true route information.

[0029] 4, the first learning unit 11 adversarially learns a GAN having a generator 111 and a classifier 112. Through the learning by the first learning unit 11, a trained generator 111′ is constructed.

[0030] 4 and 5 are diagrams schematically illustrating the neural network configuration of the generator 111 and the classifier 112 of the GAN used by the first learning unit 11. As shown in FIG. 4, the generator 111 is configured as a neural network having an input layer, a hidden layer, and an output layer. The generator 111 is a model that generates pseudo route information from random noise. For example, m randomly sampled Gaussian noise vectors (z1 to z m ).

[0031] The generator 111 performs a product-sum operation on the input and weight parameters and threshold processing using an activation function to output an output G(z). The output G(z) from the generator 111 is pseudo-route information that is similar to the first route information that the vehicle 2 actually traveled from a predetermined departure point to a destination point. A CNN or a ResNet can be used as the neural network that constitutes the generator 111.

[0032] The classifier 112 shown in Fig. 5 is configured with a neural network having an input layer, a hidden layer, and an output layer. In the example of Fig. 5, first route information x, acquired by the first acquisition unit 10, of the vehicle 2 traveling in the past from a predetermined starting point to a destination point, is given as input of training data. For example, the GPS position information of node 1, the starting point, is input to x1 shown in Fig. 5, the GPS position information of node 2, the second node passed through, is input to x2, and nThe GPS position information of the destination node n is input to

[0033] The classifier 112 performs a product-sum operation on the input and weight parameters and threshold processing using an activation function, and outputs a binary output of 1 or 0. When the classifier 112 correctly identifies the training data related to the input true route information as true route information, it outputs an output y=1. On the other hand, when the classifier 112 correctly identifies the training data related to the input pseudo route information as pseudo route information, it outputs an output y=0. In this way, the classifier 112 is a model that distinguishes the model distribution generated by the generator 111 from the data distribution of the training data, which is the true distribution. A CNN can be used as the neural network that constitutes the classifier 112.

[0034] FIG. 3 is a block diagram for explaining the adversarial learning of GAN by the first learning unit 11. The generator 111 of the GAN adopted by the first learning unit 11 is represented as a function G, and the discriminator 112 is represented as a function D. Furthermore, the true root information is represented as x, the predicted value output by the discriminator 112 is represented as y, and the correct label is represented as t. The correct label t is set to 1 for the true root information and 0 for the pseudo root information generated by the generator 111. At this time, the discriminator 112 calculates the cross entropy E CE It can be expressed as:

[0035]

number

[0036] The first term in the brace of the above equation (1) represents t n lny n In this case, the predicted value y n is the correct label of the true route information, t n = 1. On the other hand, the second term in the braces represents (1-t n )ln(1-y n ), the predicted value y n The value of the correct label (1-t n) = 0. In this way, the cross entropy E CE is the maximum value when the predicted value matches the correct label value.

[0037] Here, the generator 111 constituting the GAN has parameters w G ,θ G and the function G(w G ,θ G ) The classifier 112 uses the parameter w D ,θ D and function D(w D ,θ D ) The cross entropy E in the above equation (1) CE The objective function E of the GAN including the generator 111 and the discriminator 112 based on the above can be expressed by the following equation (2).

number

[0038] The first term of the above equation (2) represents E D(x)=1 lnD(w D ,θ D ) is the expected value that the discriminator 112 will recognize the true route information as the true route information. D(x)=0 ln(1-D(G(w G ,θ G ),w D ,θ D )) is the expected value at which the discriminator 112 discriminates the pseudo route information generated by the generator 111 as pseudo route information. In GAN learning, the generator 111 and the discriminator 112 are trained in an adversarial manner through min-max optimization of the objective function E. Therefore, the generator 111 is trained to be able to generate pseudo route information that will deceive the discriminator 112, and the discriminator 112 is trained to discriminate the pseudo route information generated by the generator 111 as pseudo route information.

[0039] In the learning of the classifier 112, when true route information is given, the classifier 112 outputs an output close to y=1, thereby maximizing the first term of the objective function E in the above equation (2). On the other hand, when pseudo route information is given, the classifier 112 learns to output an output close to y=0, thereby maximizing the second term of the objective function E.

[0040] In the learning of the generator 111, D(G(w G ,θ G ),w D ,θ D ) (D(G(z)) in Figure 3) is close to 1. G ,θ G ) (G(z) in FIG. 3 ), thereby minimizing the objective function E. The first learning unit 11 uses a learning procedure that alternately updates the parameters of the generator 111 and the parameters of the classifier 112. Details of the learning procedure of the generator 111 and the classifier 112 by the first learning unit 11 will be described later.

[0041] When the objective function E of the GAN is optimized, the first learning unit 11 passes the trained generator 111′ to the generation unit 12.

[0042] The generation unit 12 generates pseudo route information similar to the true route information acquired by the first acquisition unit 10, using a trained generator 111′ constructed by GAN training.

[0043] The first storage unit 13 stores parameters of the trained reinforcement learning model and the trained generator 111′. The first learning unit 11 also stores a trained discrimination model trained by the second learning unit 14, which will be described later.

[0044] The second learning unit 14 uses a discriminative model to learn the relationship between second route information for arriving at a destination point (first destination point) from a departure point (first departure point) with the minimum cost and the pseudo route information generated by the generation unit 12. FIG. 6 is a schematic diagram showing the structure of a neural network employed as the discriminative model in this embodiment. The neural network can have a multi-layer structure consisting of an input layer X, a hidden layer H, and an output layer Y. Each input node of the input layer X is provided with second route information for arriving at a destination point from a departure point in the shortest distance or in the shortest time, calculated by a navigation system having a conventional route search function, for example. The second route information is also provided with information on a route from a departure point node N (=X1) to a destination node N (=X2). L ) is given by the GPS location information of each node N.

[0045] The neural network shown in Figure 6 applies an activation function to the weighted sum of inputs given to the input layer X, which is the PGS location information of each node N from the starting point to the destination, and passes the output determined by threshold processing to the output layer Y. As shown in Figure 6, the number of input nodes corresponds to L, which is the number of nodes N with the minimum cost from the starting point to the destination. Each output node in the output layer Y outputs the GPS location information of each of the multiple nodes N that make up the route information.

[0046] The second learning unit 14 adjusts the weights w of the connections between the nodes, which are neurons, so that when second route information that reaches the destination in the shortest distance or the shortest time, i.e., the GPS position information of the plurality of nodes N of the second route information, is given as an input, the output becomes the value of the pseudo route information of the correct answer label, i.e., the value of the GPS position information of the plurality of nodes N of the pseudo route information. c and bias θ c The second learning unit 14 uses, for example, backpropagation to compare the obtained output value with the given input value, examine the error of each weight, and perform backpropagation to determine optimized parameters. Through this learning process, the second learning unit 14 constructs a trained discrimination model.

[0047] The trained discrimination model constructed by the second training unit 14 is stored in the first storage unit 13.

[0048] The second acquisition unit 15 acquires third route information for reaching a destination point (second destination point) from a newly set departure point (second departure point) in the vehicle 2 with the least cost. More specifically, the second acquisition unit 15 can acquire, via the in-vehicle network NW, third route information based on the shortest distance or shortest time calculated by a navigation system having a conventional route search function provided in the vehicle 2. Furthermore, the navigation system of the vehicle 2 can calculate the third route information with the least cost based on the destination point set by accepting a user input, with the current position of the vehicle 2 as the departure point. The third route information is information in which the identifiers of multiple consecutive nodes N are associated with GPS position information.

[0049] The determination unit 16 provides the third route information, acquired by the second acquisition unit 15, which reaches the destination point at the minimum cost as an unknown input to the trained identification model, performs calculations on the trained identification model, and outputs pseudo route information from the set departure point (second departure point) to the destination point (second destination point), and determines the output pseudo route information as recommended route information.

[0050] More specifically, the determination unit 16 provides the trained discrimination model with third route information based on the shortest distance or shortest time calculated by a navigation system of the vehicle 2 having a conventional route search function, based on the starting point and destination set in the vehicle 2. The determination unit 16 outputs pseudo route information from the starting point to the destination through product-sum operations and threshold processing based on trained parameters such as weights. The recommended route information determined by the determination unit 16 is route information that is very similar to true route information that represents the driving tendencies unique to the driver of the vehicle 2.

[0051] The guiding unit 17 guides the vehicle 2 from the departure point to the destination point according to the recommended route information determined by the determining unit 16. More specifically, the guiding unit 17 can display the recommended route information on a map on the screen of the display device 109 (described later) so as to distinguish it from other routes. Furthermore, the guiding unit 17 can output audio guidance from the speaker 110 when the vehicle 2 reaches a predetermined position on the recommended driving route, such as the position of each node N, based on the GPS position information of the vehicle 2.

[0052] The second storage unit 18 can store the history of the route traveled by the vehicle 2. The route information stored in the second storage unit 18 is information in which identifiers of multiple consecutive nodes N that the vehicle 2 passed through from the departure point to the destination point are associated with GPS location information. The second storage unit 18 also stores map data. The map data includes, for example, road map data for all of Japan and facility data associated therewith, such as various facilities and stores, and includes information on preset nodes N and links L.

[0053] [Hardware configuration of the driving route management device] Next, an example of a hardware configuration for realizing the travel route management device 1 having the above-described functions will be described with reference to FIG.

[0054] 2, the driving route management device 1 can be realized by, for example, a computer including a processor 102, a main memory device 103, a communication interface 104, an auxiliary memory device 105, and an input / output (I / O) 106, all connected via a bus 101, and a program that controls these hardware resources. Furthermore, the driving route management device 1 includes a GPS receiver 107, a sensor 108, a display device 109, and a speaker 110.

[0055] The processor 102 is realized by a CPU, a GPU, an FPGA, an ASIC, or the like.

[0056] The main memory device 103 pre-stores programs for the processor 102 to perform various controls and calculations. The processor 102 and the main memory device 103 implement the functions of the driving route management device 1, such as the first acquisition unit 10, the first learning unit 11, the generation unit 12, the second learning unit 14, the determination unit 16, and the guidance unit 17 shown in FIG.

[0057] The communication interface 104 is an interface circuit for connecting the travel route management device 1 to various external electronic devices via a network.

[0058] The auxiliary storage device 105 is composed of a readable / writable storage medium and a drive for reading and writing various information such as programs and data from and to the storage medium. The auxiliary storage device 105 can use a semiconductor memory such as a hard disk or flash memory as the storage medium.

[0059] The auxiliary storage device 105 has a program storage area for storing a driving route management program for determining recommended route information and guiding the vehicle 2 along the determined route information. The auxiliary storage device 105 also has a program storage area for storing a first learning program for performing adversarial learning of the GAN executed by the driving route management device 1 and a second learning program for performing supervised learning of a discriminative model. The auxiliary storage device 105 realizes the first storage unit 13 and the second storage unit 18 described in FIG. 1. Furthermore, the auxiliary storage device 105 may have, for example, a backup area for backing up the above-mentioned data, programs, etc.

[0060] The input / output I / O 106 is an input / output device that inputs signals from external devices and outputs signals to external devices.

[0061] The GPS receiver 107 receives the absolute positions and time of the GPS satellites, and the current position of the vehicle 2 is determined based on this GPS position information.

[0062] The sensor 108 is realized by various sensors such as a geomagnetic sensor, a gyro sensor, an acceleration sensor, a vehicle speed sensor, a camera, LiDAR, etc. The sensor 108 detects the direction, tilt angle, travel distance, etc. of the vehicle 2, and these are used to identify the position of the vehicle 2.

[0063] The display device 109 is configured with an organic EL display, a liquid crystal display, etc. The display device 109 displays route information recommended in map data, as well as the vehicle's position and traveling direction on the screen. The display device 109 also functions as a touch panel and can accept destination setting from the user.

[0064] The speaker 110 converts the digital data of the voice guidance into analog data and outputs it.

[0065] [Operation of the driving route management device] Next, the operation of the travel route management device 1 having the above-described configuration will be described with reference to the flowcharts of FIGS.

[0066] As shown in FIG. 7, first, the first acquisition unit 10 acquires first route information on which the vehicle 2 has traveled in the past (step S1). The first acquisition unit 10 can acquire, for example, the history of the travel routes of the vehicle 2 stored in the second storage unit 18. Next, the first learning unit 11 performs a first learning process (step S2). In step S2, the first learning unit 11 performs adversarial learning of a GAN that uses the first route information as true route information and includes a generator 111 that generates pseudo route information similar to the true route information and a classifier 112 that distinguishes between the pseudo route information generated by the generator 111 and the true route information. Details of the first learning process performed by the first learning unit 11 will be described later.

[0067] Next, the generation unit 12 generates pseudo route information using the trained generator 111' constructed by the first learning unit 11 (step S3). Thereafter, the second learning unit 14 executes a second learning process (step S4). In step S4, the second learning unit 14 uses a discriminative model to learn the relationship between second route information that reaches the destination from the same starting point and destination as the first route information acquired in the first step S1 in the shortest distance or in the shortest time, and the first route information generated by the generation unit 12 in step S3, i.e., pseudo route information that is similar to the true route information.

[0068] In step S4, the second learning unit 14 performs supervised learning of an identification model having the neural network configuration shown in Figure 6 using training data consisting of GPS location information of each node N indicating the second route information and GPS location information of each node N indicating the pseudo route information generated by the trained generator 111'.

[0069] The second learning unit 14 receives as input the GPS position information of each node N, which represents second route information based on the shortest distance or shortest time set for the vehicle 2, calculated by a navigation system having a conventional route search function. Furthermore, the second learning unit 14 adjusts the weights w of the connections between the nodes, which are neurons, so that the output of the neural network when the second route information is received is the value of the GPS position information of each node N, which represents the pseudo route information, which is the correct label. c and bias θ c The second learning unit 14 uses, for example, backpropagation to compare the obtained output value with the given input value, examine the error of each weight, and perform backpropagation to determine the optimal parameters. A trained classifier is constructed by the second learning process in step S4.

[0070] Thereafter, the first storage unit 13 stores the trained classifier constructed in step S4 (step S5). Next, the second acquisition unit 15 acquires third route information for reaching the destination from the newly set departure point in the vehicle 2 in the shortest distance or the shortest time (step S6). Specifically, the second acquisition unit 15 can acquire the third route information from a navigation system having a conventional route search function via the in-vehicle network NW.

[0071] Next, the determination unit 16 provides the third route information acquired in step S6 as an unknown input to the trained classifier, performs calculations on the trained classifier, outputs pseudo route information from the newly set departure point to the destination point, and determines the output pseudo route information as recommended route information (step S7).

[0072] Thereafter, the guiding unit 17 guides the vehicle 2 to the destination point along the recommended route information determined in step S7 (step S8).

[0073] Next, the first learning process in step S2 described in FIG. 7 will be described with reference to FIG. 8. First, the first learning unit 11 inputs the true route information to the classifier 112 as training data 114, and adjusts the parameter w of the classifier 112 so that the true route information is distinguished from the true route information (y=1). D ,θ D As shown in the block diagram of the first learning unit 11 in FIG. 3, the true route information collected in step S1 in FIG. 7 is used as training data 114 to be input when training the classifier 112.

[0074] In step S20, the first learning unit 11 can cause the classifier 112 to learn the true route information using, for example, backpropagation. By step S20, the classifier 112 that can distinguish the true route information from the true route information is constructed in advance.

[0075] Next, the first learning unit 11 generates Gaussian noise and provides a random vector of the generated Gaussian noise as an input to the generator 111 (step S21). Subsequently, the generator 111 calculates a random vector of the input z and the weight parameter w based on the provided Gaussian noise. G ,θ G Then, a product-sum operation and a threshold process using an activation function are performed to generate pseudo route information G(z) (step S22).

[0076] Next, the first learning unit 11 performs learning of the classifier 112. The learning of the classifier 112 is performed by using the parameter w D ,θ D First, the first learning unit 11 provides the true route information acquired in step S1 of FIG. 7 as training data 114 to the classifier 112 as input. Then, the first learning unit 11 adjusts the parameter w by backpropagation or the like so that the objective function E in the above formula (2) is maximized. D ,θ D (Step S23). The label of the training data 114 is set to 1 (true route information).

[0077] Next, the first learning unit 11 provides the pseudo route information generated by the generator 111 in step S22 to the discriminator 112, and calculates the parameter w by backpropagation or the like so that the objective function E in the above equation (2) is maximized. D ,θ D (Step S24). That is, in steps S23 and S24, in order to maximize the objective function E in the above equation (2), the first term is updated as D(w D ,θ D )=1 is output, and the second term is D(G(w G ,θ G ),w D ,θ D )=0. In the training data 114, the label is set to 0 (pseudo route information).

[0078] The learning of the classifier 112 in steps S23 and S24 corresponds to the dashed arrows in the block diagram of the first learning unit 11 shown in FIG. 3 , which indicate that a classifier error is calculated in block 115 of the objective function E based on the output 113 from the classifier 112, and then the error is back-propagated to the classifier 112.

[0079] Next, the first learning unit 11 performs learning of the generator 111. The learning of the generator 111 is performed with the parameters of the discriminator 112 fixed. The first learning unit 11 trains the generator 111 so that pseudo route information is generated when random Gaussian noise is given to the generator 111. Specifically, the first learning unit 11 learns the parameters w G ,θ G is updated (step S25).

[0080] The learning in step S25 corresponds to the flow indicated by the dashed arrow indicating backpropagation of error to the generator 111 in the block diagram of the first learning unit 11 in Fig. 3. That is, step S25 corresponds to the flow indicated by the dashed arrow in which pseudo route information generated by the generator 111 in Fig. 3 is input to the discriminator 112, a generator error is calculated from its output 113 in the block 115 of the objective function E, and the error is further backpropagated to the generator 111.

[0081] Thereafter, learning of the discriminator 112 and the generator 111 from step S22 to step S25 is repeated until the value of the objective function E reaches a Nash equilibrium and converges (step S26: NO). On the other hand, if the value of the objective function E has converged (step S26: YES), the processing from step S20 to step S26 is repeated until learning of the generator 111 and the discriminator 112 is performed using the remaining M-1 pieces of true route information out of all M pieces of true route information (step S27: NO).

[0082] Thereafter, when the generator 111 and the discriminator 112 are trained using the remaining M-1 pieces of true root information (step S27: YES), the first training unit 11 stores the trained generator 111' in the first storage unit 13 (step S28). The trained generator 111' is constructed by the above-described processes from step S20 to step S28. After that, the process proceeds to step S3 in FIG. 7.

[0083] As described above, the driving route management device 1 according to this embodiment performs adversarial learning of the GAN using the history of past driving routes of the vehicle 2 as true route information, and generates pseudo route information using the generated trained generator 111'. Furthermore, the pseudo route information similar to the true route information generated by the trained generator 111' is used as training data to train a discriminative model on the relationship between the pseudo route information and minimum-cost route information calculated by, for example, a navigation system with a conventional route search function. This makes it possible to generate driving routes that match the driving tendencies of each driver.

[0084] The traveling route management device 1 according to the embodiment has been described as having a separate configuration from a navigation system having a conventional route search function that is provided in the vehicle 2. However, the traveling route management device 1 can be configured integrally with a navigation system having a conventional route search function. In this case, the traveling route management device 1 can display route information for the shortest time or shortest distance, as well as recommended route information that reflects the driver's driving tendencies, on a screen so that the driver can select from the displayed information, and can guide the vehicle 2 to the destination based on the selected route information.

[0085] The above describes embodiments of the driving route management device and driving route management method of the present invention, but the present invention is not limited to the described embodiments, and various modifications that a person skilled in the art can imagine are possible within the scope of the invention described in the claims. [Explanation of symbols]

[0086] 1...driving route management device, 2...vehicle, 10...first acquisition unit, 11...first learning unit, 12...generation unit, 13...first memory unit, 14...second learning unit, 15...second acquisition unit, 16...determination unit, 17...guidance unit, 18...second memory unit, 101...bus, 102...processor, 103...main memory unit, 104...communication interface, 105...auxiliary memory unit, 106...input / output I / O, 107...GPS receiver, 108...sensor, 109...display unit, 110...speaker, 111...generator, 112...classifier, 113...output, 114...training data, 115...block of objective function E, NW...in-vehicle network.

Claims

1. a first acquisition unit configured to acquire first route information from a first departure point to a first destination point, the first route information having been traveled by the vehicle in the past; a first learning unit configured to perform adversarial learning of a generative model including a generator that generates pseudo route information similar to the first route information, using the first route information as true route information from the first departure point to the first destination point, and a classifier that distinguishes between the pseudo route information generated by the generator and the true route information; a generation unit configured to generate the pseudo route information using a trained generator constructed by the first learning unit; a second learning unit configured to learn, using a discriminative model, a relationship between second route information for arriving at the first destination point from the first departure point at the minimum cost and the pseudo route information generated by the trained generator; a storage unit configured to store the trained discrimination model constructed by the second training unit; A driving route management device comprising:

2. The driving route management device according to claim 1, a second acquisition unit configured to acquire third route information for arriving at a second destination point from a second departure point newly set in the vehicle at a minimum cost; a determination unit configured to provide the third route information as an unknown input to the trained discriminant model, perform calculations on the trained discriminant model, output pseudo route information from the second departure point to the second destination point, and determine the output pseudo route information as recommended route information; A driving route management device comprising:

3. 3. The driving route management device according to claim 2, Further, a guiding unit configured to guide the vehicle from the second departure point to the second destination point along the recommended route information determined by the determining unit. A driving route management device characterized by:

4. 4. The driving route management device according to claim 2, wherein: The second route information and the third route information are route information for the vehicle to reach the destination from the departure point in the shortest distance or the shortest time. A driving route management device characterized by:

5. a first acquisition step of acquiring first route information from a first departure point to a first destination point along which the vehicle has traveled in the past; a first learning step of performing adversarial learning of a generative model including a generator that generates pseudo route information similar to the first route information, using the first route information as true route information from the first departure point to the first destination point, and a classifier that distinguishes between the pseudo route information generated by the generator and the true route information; a generating step of generating the pseudo route information using the trained generator constructed in the first learning step; a second learning step of learning, using a discriminative model, a relationship between second route information for arriving at the first destination point from the first departure point at the minimum cost and the pseudo route information generated by the trained generator; a storage step of storing the trained discrimination model constructed in the second learning step in a storage unit; A driving route management method comprising:

6. 6. The driving route management method according to claim 5, a second acquisition step of acquiring third route information for arriving at a second destination point from a second departure point newly set in the vehicle at a minimum cost; a determining step of providing the third route information as an unknown input to the trained discriminant model, performing an operation on the trained discriminant model, outputting pseudo route information from the second departure point to the second destination point, and determining the output pseudo route information as recommended route information; A driving route management method comprising:

7. 7. The driving route management method according to claim 6, The method further includes a guiding step of guiding the vehicle from the second departure point to the second destination point along the recommended route information determined in the determining step. A driving route management method comprising:

Images