JP2022038241A - Vehicle allocation system - Google Patents

Abstract

To suppress an inequality among passengers of vehicle allocation services.SOLUTION: In a vehicle allocation system 100 comprising a server 1 and a plurality of autonomous driving vehicles 2, and providing a vehicle allocation service using the autonomous driving vehicle 2, the server 1 is configured that the server selects a learning vehicle which needs to learn a learning model from the autonomous driving vehicles 2 which are waiting for a vehicle dispatch service based on a learning progress of the learning model used in the autonomous driving vehicle 2 received from the autonomous driving vehicle 2, and transmits a learning instruction to the learning vehicle; and upon receiving the learning instruction, the autonomous driving vehicle 2 is configured to perform traveling for learning the learning model while waiting for dispatch based on the learning instruction.SELECTED DRAWING: Figure 1

Description

The present invention relates to a vehicle allocation system.

In recent years, autonomous driving technology has been developed for the realization of mobility services such as taxis, buses, and ride sharing using autonomous vehicles capable of autonomous driving. As one of such mobility services, Patent Document 1 aims to move an autonomous driving vehicle to a boarding point desired by the user in response to a user's request for use, and to allow the user to ride on the autonomous driving vehicle. The technology related to the vehicle dispatch service that can be delivered to the ground is disclosed.

Patent Document 2 discloses that the temperature of the exhaust gas purification catalyst of an internal combustion engine is estimated by using a learned neural network trained by a server or a vehicle.

Japanese Unexamined Patent Publication No. 2017-182137 Japanese Unexamined Patent Publication No. 2019-183698

In the future, it is expected that various controls using the learning model will be carried out on each autonomous driving vehicle used for the vehicle allocation service, and that the learning model will be learned. Naturally, it is desirable that the learning model used in each autonomous driving vehicle is well-learned. However, the vehicle allocation status of each autonomous vehicle used for the vehicle allocation service varies from vehicle to vehicle. Therefore, depending on the vehicle allocation situation, it is conceivable that there will be a vehicle in which the learning of the learning model has not sufficiently progressed, that is, an autonomous driving vehicle in which the learning progress of the learning model is lower than that of other autonomous driving vehicles.

Then, for example, the quality of the vehicle dispatch service for the passengers of the vehicle dispatch service to which the self-driving vehicle in which the learning model has not been sufficiently learned is dispatched, and the self-driving vehicle in which the learning model has been sufficiently learned are dispatched. There is a risk of inequality among the passengers of the vehicle dispatch service, such as lowering the quality of the vehicle dispatch service to the passengers.

The present invention has been made by paying attention to such a problem, and an object of the present invention is to suppress the occurrence of inequality among the users of the vehicle dispatch service.

In order to solve the above problems, the vehicle allocation system according to a certain aspect of the present invention includes a server and a plurality of autonomous driving vehicles, and provides a vehicle allocation service using the autonomous driving vehicle. Based on the learning progress of the learning model used in the autonomous driving vehicle received from the autonomous driving vehicle, the server needs to learn the learning model from the autonomous driving vehicle waiting for the vehicle allocation service. It is configured to select a vehicle and send learning instructions to the learning vehicle. Upon receiving the learning instruction, the autonomous driving vehicle is configured to perform traveling for learning the learning model while waiting for dispatch based on the learning instruction.

According to this aspect of the present invention, a learning vehicle that requires learning of a learning model is selected from among the automatic driving vehicles waiting for a vehicle allocation service based on the learning progress of the learning model of each automatic driving vehicle. , The learning vehicle can be made to run for learning the learning model while waiting for the vehicle to be dispatched. Therefore, even if the vehicle allocation status of each autonomous driving vehicle varies from vehicle to vehicle, it is possible to suppress a difference in the learning progress of the learning model of each autonomous driving vehicle. Therefore, it is possible to suppress the occurrence of inequality among the users of the vehicle dispatch service.

FIG. 1 is a schematic configuration diagram of a vehicle allocation system according to an embodiment of the present invention. FIG. 2 is a schematic diagram showing a hardware configuration of an autonomous driving vehicle according to an embodiment of the present invention. FIG. 3 is a diagram showing an example of a neural network model. FIG. 4 is a flowchart illustrating a vehicle allocation process executed between a server, an autonomous vehicle, and a mobile terminal. FIG. 5 is a flowchart illustrating a learning progress update process executed between the server and the autonomous driving vehicle. FIG. 6 is a flowchart illustrating a learning driving process executed between the server and the autonomous driving vehicle.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In the following description, similar components are given the same reference numbers.

FIG. 1 is a schematic configuration diagram of a vehicle allocation system 100 according to an embodiment of the present invention.

The vehicle allocation system 100 according to the present embodiment includes a server 1, a plurality of autonomous driving vehicles 2 used for the vehicle allocation service, and a mobile terminal 3 such as a mobile phone or a tablet computer possessed by each user 4 of the vehicle allocation service. Be prepared. The vehicle allocation system 100 moves the autonomous driving vehicle 2 to the desired boarding point of the passenger 4 in response to the usage request of the passenger 4, and causes the autonomous driving vehicle 2 to board the passenger 4 to the destination of the passenger 4. It is a system to realize the delivery service.

The server 1, the self-driving vehicle 2, and the mobile terminal 3 can communicate with each other via a network 5 composed of an optical communication line or the like. The server 1 is connected to the network 5 via, for example, a gateway (not shown). The self-driving vehicle 2 and the mobile terminal 3 are connected to the network 5 via, for example, a radio base station 6.

As shown in FIG. 1, the server 1 includes a server communication unit 11, a server storage unit 12, and a server processing unit 13.

The server communication unit 11 has a communication interface circuit for connecting the server 1 to the network 5 via, for example, a gateway, and can communicate with each other with each autonomous driving vehicle 2 and each mobile terminal 3. It is configured as follows.

The server storage unit 12 has a storage medium such as an HDD (Hard Disk Drive), an optical recording medium, and a semiconductor memory, and stores various computer programs, data, and the like used for processing in the server processing unit 13.

The server processing unit 13 has one or more processors and peripheral circuits thereof. The server processing unit 13 executes various computer programs stored in the server storage unit 12 and comprehensively controls the overall operation of the server 1, for example, a CPU (Central Processing Unit). For example, when the server processing unit 13 receives a vehicle allocation request from the mobile terminal 3, it determines a vehicle allocation vehicle from among a plurality of autonomous driving vehicles 2 and transmits a vehicle allocation instruction to the vehicle allocation vehicle.

FIG. 2 is a schematic diagram showing a hardware configuration of the autonomous driving vehicle 2.

As shown in FIG. 2, the autonomous driving vehicle 2 includes a peripheral information detection device 21, a GPS receiver 22, a map database 23, an actuator 24, a state quantity detection device 25, a communication module 26, and an electronic control unit 20, and accelerates. It is configured to be able to automatically perform driving operations related to steering and braking. The peripheral information detection device 21, the GPS receiver 22, the map database 23, the actuator 24, the state quantity detection device 25, and the communication module 26 are electronically controlled units via an in-vehicle network 27 compliant with standards such as CAN (Controller Area Network). Connected to 20 communicably.

The peripheral information detection device 21 detects peripheral information of the autonomous driving vehicle 2. Peripheral information includes information such as white lines on roads, other vehicles, pedestrians, bicycles, buildings, signs, traffic lights, obstacles, and the like. The peripheral information detection device 21 includes, for example, an out-of-vehicle camera, a millimeter-wave radar, a lidar (LiDAR; Light Detection and Ranging), an ultrasonic sensor, and the like. The output of the peripheral information detection device 21 is transmitted to the electronic control unit 20.

The GPS receiver 22 receives signals from three or more GPS satellites and detects the current position of the autonomous driving vehicle 2 (for example, the latitude and longitude of the autonomous driving vehicle 2). The output of the GPS receiver 22 is transmitted to the electronic control unit 20.

The map database 23 stores map information. The electronic control unit 20 acquires map information from the map database 23.

The actuator 24 is various control components driven by the electronic control unit 20 for autonomous driving, and is, for example, a drive device (for example, at least one of an internal combustion engine and a motor) necessary for accelerating the autonomous driving vehicle 2. It also includes a braking device (for example, a brake actuator) necessary for braking the autonomous driving vehicle 2, a steering device (for example, a steering motor) necessary for steering the autonomous driving vehicle 2.

The state quantity detecting device 25 detects the state quantity of the self-driving vehicle 2, the internal combustion engine, the motor, the battery, the surrounding environment, and the like. For example, the state amount detecting device 25 includes a vehicle speed sensor, a yaw rate sensor, an accelerator pedal stroke sensor, an air flow meter, an air fuel ratio sensor, a crank angle sensor, a torque sensor, a voltage sensor, an outside temperature sensor, a NOx sensor and the like. The output of the state quantity detecting device 25 is transmitted to the electronic control unit 20.

The communication module 26 is an in-vehicle terminal having a wireless communication function. The communication module 26 connects to the network 5 via the radio base station 6 by accessing the radio base station 6 (see FIG. 1) connected to the network 5 (see FIG. 1) via a gateway (not shown) or the like. Will be done. As a result, mutual communication is performed with the server 1.

The electronic control unit 20 includes an in-vehicle communication interface 201, a vehicle storage unit 202, and a vehicle processing unit 203 connected to each other via a signal line.

The in-vehicle communication interface 201 is a communication interface circuit for connecting the electronic control unit 20 to the in-vehicle network 27.

The vehicle storage unit 202 has a storage medium such as an HDD (Hard Disk Drive), an optical recording medium, and a semiconductor memory, and stores various computer programs, data, and the like used for processing in the vehicle processing unit 203.

The vehicle processing unit 203 includes one or more processors and peripheral circuits thereof. The vehicle processing unit 203 executes various computer programs stored in the vehicle storage unit 202 and comprehensively controls various control components mounted on the autonomous driving vehicle 2, for example, a CPU. For example, when the vehicle processing unit 203 receives a vehicle allocation instruction from the server 1, the vehicle processing unit 203 moves the own vehicle toward the desired boarding point of the user 4 who has requested the vehicle allocation, and causes the user 4 to board the own vehicle. Deliver to 4 destinations.

In each autonomous driving vehicle 2, in controlling each autonomous driving vehicle 2, a learning model (artificial intelligence model) in which learning such as machine learning is performed is used as needed. In this embodiment, as a learning model, a neural network model (hereinafter referred to as “NN model”) using a deep neural network (DNN; Deep Neural Network), a convolutional neural network (CNN; Convolutional Neural Network), or the like is used. Deep learning is carried out for the NN model. Therefore, the learning model according to this embodiment can be said to be a trained NN model in which deep learning is performed. Deep learning is one of the representative machine learning methods of artificial intelligence (AI).

FIG. 3 is a diagram showing an example of the NN model.

The circles in FIG. 3 represent artificial neurons. Artificial neurons are commonly referred to as nodes or units (referred to herein as "nodes"). In FIG. 3, L = 1 indicates an input layer, L = 2 and L = 3 indicate a hidden layer, and L = 4 indicates an output layer. The hidden layer is also called the middle layer. Although FIG. 3 illustrates an NN model having two hidden layers, the number of hidden layers is not particularly limited, and the nodes of the input layer, the hidden layer, and the output layer are not particularly limited. The number is not particularly limited.

In FIG. 3, x ₁ and x ₂ indicate the node of the input layer (L = 1) and the output value from the node, and y indicates the node of the output layer (L = 4) and its output value. There is. Similarly, z ₁ ^{(L = 2)} _, z ₂ ^{(L = 2)} and z ₃ ^{(L = 2)} indicate each node of the hidden layer (L = 2) and the output value from that node, and z ₁ ^{(L = 3)} and z ₂ ^{(L = 3)} indicate each node of the hidden layer (L = 3) and the output value from that node.

The input is output as it is at each node of the input layer. On the other hand, the output values x1 and x2 of _each node of the input layer are input to each node of the hidden layer (L = ₂ ), and the corresponding weights w and each node of the hidden layer (L = 2) are input. The total input value u is calculated using the bias b. For example, in FIG. 3, the total input value _uk ^{(L = 2)} calculated at each node represented by z _k ^{(L = 2)} (k = 1, 2, 3) of the hidden layer (L = 2) is It becomes as follows (M is the number of nodes in the input layer).

Next, this total input value uk ^{(L = 2)} is converted by the activation function f, and the output value z _k ⁽ L = 2) is output from the node _represented by z _k ^{(L = 2)} of the hidden layer (L = 2). ²⁾ It is output as (= f ( _uk ^{(L = 2)} )). On the other hand, for each node of the hidden layer (L = 3), the output values of each node of the hidden layer (L = 2) z ₁ ^{(L = 2)} _, z ₂ ^{(L = 2)} and z ₃ ^{(L = 2} ). ⁾ Is input, and at each node of the hidden layer (L = 3), the total input value u (= Σz · w + b) is calculated using the corresponding weight w and bias b, respectively. This total input value u is similarly converted by the activation function, and is output as output values z ₁ ^{(L = 3)} and z ₂ ^{(L = 3)} from each node of the hidden layer (L = 3). The activation function is, for example, a sigmoid function σ.

Further, the output values z ₁ ^{(L = 3)} and z ₂ ^{(L = 3) of each node of the hidden layer (L = 3)} are input to the node of the output layer (L = 4), and the node of the output layer is input. Then, the total input value u (Σz · w + b) is calculated using the corresponding weights w and the bias b, respectively, or the total input value u (Σz · w) is calculated using only the corresponding weights w. Will be done. For example, the node of the output layer uses an identity function as the activation function. In this case, the total input value u calculated in the node of the output layer is directly output from the node of the output layer as the output value y.

As described above, the NN model includes an input layer, a hidden layer, and an output layer, and when one or more input parameters are input from the input layer, one or a plurality of output parameters corresponding to the input parameters are output. Output from the layer.

As an example of the input parameters, for example, when controlling the internal combustion engine mounted on the automatic driving vehicle 2 using the NN model, the engine rotation speed, the engine cooling water temperature, the fuel injection amount, the fuel injection timing, the fuel pressure, etc. The current values of various parameters indicating the operating state of the internal combustion engine, such as the amount of intake air, the intake temperature, the EGR rate, and the boost pressure, can be mentioned. Examples of output parameters corresponding to such input parameters include estimated values of various parameters representing the performance of the internal combustion engine such as NOx concentration in exhaust gas, concentration of other substances, and engine output torque. As a result, by inputting the current values of various parameters indicating the operating state of the internal combustion engine into the NN model as input parameters, the estimated values (current estimated value or future estimated value) of various parameters indicating the performance of the internal combustion engine can be input. Since it can be acquired as an output parameter, the internal combustion engine can be controlled so that the performance of the internal combustion engine approaches a desired performance, for example, based on the output parameter. Thereby, for example, the engine output torque can be appropriately controlled to control the acceleration in the front-rear direction and the lateral direction of the vehicle, and the vehicle behavior can be stabilized and the riding comfort can be improved. Further, when a sensor or the like for actually measuring the output parameter is provided, it is possible to determine the failure of the sensor or the like according to the difference between the measured value and the estimated value.

In order to improve the accuracy of the NN model, it is necessary to train the NN model. For learning the NN model, a large number of training data sets including the measured values of the input parameters and the measured values (correct answer data) of the output parameters corresponding to the measured values of the input parameters are used. By repeatedly updating the values of the weight w and the bias b in the neural network by a known error back propagation method using a large number of training data sets, the values of the weight w and the bias b are learned and a learning model is generated. To.

Here, the vehicle allocation status of each autonomous driving vehicle 2 used for the vehicle allocation service varies from vehicle to vehicle. Therefore, among the autonomous driving vehicles 2, some vehicles are evenly distributed within the service providing area of the vehicle allocation service, and some vehicles are distributed only to a part of the biased area within the service providing area. The service area of the vehicle dispatch service is not particularly limited, and may be the entire area within a prefecture or a wide area across prefectures, a part of the prefecture, or a smart city (connected city). It may be a relatively limited area such as inside.

In the case of the self-driving vehicle 2 evenly distributed in the service providing area, the NN model can be learned based on the training data set obtained by traveling the entire service providing area evenly, so that the service providing area can be learned. It is possible to create a learning model optimized for.

On the other hand, in the case of the self-driving vehicle 2 that is distributed only to a part of the service providing area in a biased area, the non-driving area or the area where the number of times of driving is small in the service providing area (hereinafter, "non-driving area, etc."" Since it is not possible to acquire a sufficient amount of training data set in (), there will be areas in the service providing area where the learning of the NN model is not sufficient (hereinafter referred to as “learning unlearned area”). Therefore, it is not possible to create a trained model optimized for the service providing area, and the NN model of the autonomous driving vehicle 2 distributed only to a part of the biased area in the service providing area is a non-driving area, etc. It becomes an NN model with insufficient learning in.

Therefore, if the autonomous driving vehicle 2 that is dispatched only to a part of the biased area in the service providing area travels in the non-driving area or the like while the vehicle dispatching service is being provided, the non-driving area or the like. The self-driving vehicle 2 will be controlled using the NN model for which learning is insufficient. As a result, the autonomous driving vehicle 2 will be controlled using an NN model that is not sufficiently suitable for the road traffic environment such as a non-driving area and the terrain. There is a risk that the quality of vehicle dispatch services for service users will deteriorate.

Therefore, in the present embodiment, it is decided to calculate the learning progress of the NN model of each autonomous driving vehicle 2 in the service providing area. In the present embodiment, the learning progress of the NN model is an index showing how much the learning of the NN model has progressed in the service providing area of the vehicle dispatch service, and is basically in the service providing area. Vehicles with many areas where learning is inadequate show lower values. Therefore, the self-driving vehicle 2 having a low learning progress of the NN model has insufficient learning in a specific area within the service providing area, that is, insufficient traveling in the specific area, and is for learning the NN model. It means that the training data set required for the training has not been sufficiently acquired.

Then, based on the learning progress of the NN model of each autonomous driving vehicle 2, the learning progress is relatively low among the autonomous driving vehicles 2 waiting for the vehicle allocation service, and the next vehicle allocation service is provided. Select the self-driving vehicle 2 that has enough time as the "learning vehicle", and drive the learning inexperienced area while waiting for the vehicle dispatch service in order to improve the learning progress of the own vehicle for the learning vehicle. And said.

As a result, the learning vehicle can travel in an unlearned area while waiting for the vehicle allocation service, acquire a training data set in the area, and learn an NN model based on the acquired training data set. .. Therefore, it is possible to improve the learning progress of each autonomous driving vehicle 2 while the vehicle dispatch service is on standby.

Hereinafter, various processes executed by the vehicle allocation system 100 will be described.

FIG. 4 is a flowchart illustrating a vehicle allocation process executed between the server 1, the autonomous driving vehicle 2, and the mobile terminal 3.

In step S1, the user 4 of the vehicle allocation service operates the mobile terminal 3 owned by the user 4 to transmit the vehicle allocation request to the server 1. The user 4 can send a vehicle allocation request to the server 1 by, for example, activating an application for using the vehicle allocation service on the mobile terminal 3. The vehicle allocation request includes, for example, identification information (personal information) of the passenger 4, information necessary for vehicle allocation such as desired boarding time, desired boarding position, desired disembarkation position, and planned travel area.

In step S2, the server 1 that has received the vehicle allocation request refers to the vehicle allocation status of each autonomous driving vehicle 2, the vehicle allocation reservation information, etc., selects a vehicle allocation vehicle from each autonomous driving vehicle 2, and also provides information on the vehicle allocation vehicle ( For example, the vehicle type, vehicle identification number, etc.) are transmitted to the mobile terminal 3 from which the vehicle allocation request is transmitted.

In step S3, the mobile terminal 3 that is the transmission source of the vehicle allocation request determines whether or not the information regarding the vehicle allocation vehicle has been received, and if the information regarding the vehicle allocation vehicle is received, proceeds to the process of step S4 and proceeds to the process of the vehicle allocation vehicle. If the information has not been received, after a predetermined time has passed, it is determined whether or not the information regarding the dispatched vehicle has been received again.

In step S4, the mobile terminal 3 that has received the information about the vehicle dispatching vehicle provides the information to the user 4 of the vehicle dispatching service, for example, by displaying the information on the screen of the mobile terminal 3 via an application.

In step S5, the server 1 transmits a vehicle allocation instruction to the vehicle dispatch vehicle. The vehicle allocation instruction includes various information included in the vehicle allocation request, that is, identification information (personal information) of the passenger 4, desired boarding time, desired boarding position, desired disembarkation position, planned travel area, and the like.

In step S6, the electronic control unit 20 of the autonomous driving vehicle 2 that has received the vehicle allocation instruction automatically drives the own vehicle according to the vehicle allocation instruction, and by the desired boarding time of the user 4 who has requested the vehicle allocation, the own vehicle is delivered. Arrive at the desired boarding position.

FIG. 5 is a flowchart illustrating a learning progress update process executed between the server 1 and the autonomous driving vehicle 2.

In step S11, the electronic control unit 20 of the self-driving vehicle 2 determines whether or not a predetermined period has elapsed since the learning progress of the NN model of the own vehicle was transmitted to the server 1 last time. The electronic control unit 20 proceeds to the process of step S12 if a predetermined period has elapsed since the learning progress of the NN model of the own vehicle was transmitted to the server 1 last time. On the other hand, the electronic control unit 20 ends this process if a predetermined period has not elapsed since the learning progress of the NN model of the own vehicle was transmitted to the server 1 last time.

In step S12, the electronic control unit 20 of the autonomous driving vehicle 2 calculates the learning progress of the NN model of the own vehicle.

In this embodiment, the service providing area is divided into a plurality of subregions, and the learning progress of the NN model for each subregion based on the number of training data sets acquired in each subregion and the acquisition time thereof. The degree is calculated. Specifically, in this embodiment, the learning progress of each subregion is calculated based on the following equation (1).
Learning progress = A x (number of training data sets) + F (training data set acquisition time) ... (1)

In the equation (1), A is a predetermined value, and is basically set to a value such that the learning progress increases as the number of training data sets increases. Further, F is a predetermined conversion function, and is a function that calculates a value such that the learning progress becomes smaller as the training data set acquisition time becomes older than the time when the learning progress is calculated. This is because the older the acquisition time of the training data set acquired in a certain subregion, the more the traffic environment and the like change, so that it becomes necessary to newly drive and learn in the subregion.

In this embodiment, the electronic control unit 20 acquires training data sets (measured values of input parameters and output parameters of the NN model) at any time while the vehicle is running.

In step S13, the electronic control unit 20 of the autonomous driving vehicle 2 transmits the calculated learning progress of the NN model of the own vehicle to the server 1 together with the identification information of the own vehicle.

In step S14, the server 1 that has received the learning progress of the NN model updates the learning progress database in which the learning progress of the NN model for each subregion of each autonomous driving vehicle 2 is stored. Specifically, the server 1 identifies the self-driving vehicle 2 that has transmitted the learning progress of the NN model based on the vehicle identification information, and updates the learning progress of the NN model in each subregion of the self-driving vehicle 2. do.

FIG. 6 is a flowchart illustrating a learning traveling process executed between the server 1 and the autonomous driving vehicle 2.

In step S21, the server 1 periodically refers to the learning progress database and extracts learning vehicle candidates from each autonomous driving vehicle 2. For example, the server 1 extracts from each autonomous driving vehicle 2 an autonomous driving vehicle 2 having a small area where the learning progress is equal to or less than a predetermined value as a learning vehicle candidate.

In step S22, the server 1 goes from the learning vehicle candidates to a small area where the learning progress is equal to or less than a predetermined value from the vehicle base (garage) during the waiting time for vehicle allocation, and trains in the small area. The self-driving vehicle 2 that can return to the depot after traveling for data set acquisition is selected as the learning vehicle. That is, the server 1 has a small area where the learning progress is relatively low among the autonomous driving vehicles 2 waiting for the vehicle allocation service, and the automatic operation has enough time until the next vehicle allocation service is provided. The driving vehicle 2 is selected as the learning vehicle.

In step S23, the server 1 transmits a learning instruction to the learning vehicle. The learning instructions include information about the subregions from which the training data set is acquired, that is, information about the subregions that need learning to improve the learning progress of the NN model (hereinafter referred to as "learning target subregions"). The return time to the vehicle base and is included.

In step S24, the electronic control unit 20 of the autonomous driving vehicle 2 determines whether or not the learning instruction has been received. If the electronic control unit 20 has received the learning instruction, the electronic control unit 20 proceeds to the process of step S25. On the other hand, if the electronic control unit 20 has not received the learning instruction, the electronic control unit 20 ends the current process.

In step S25, the electronic control unit 20 of the autonomous driving vehicle 2 automatically departs its own vehicle toward the learning target small area based on the learning instruction to acquire the training data set in the learning target small area. After driving, the vehicle is returned to the vehicle base by the return time. Then, the electronic control unit 20 executes learning of the NN model based on the training data set acquired while traveling in the training target subregion. In the present embodiment, the learning of this NN model is executed at any time while traveling within the learning target subregion, but it may be performed after returning to the depot.

The vehicle allocation system 100 according to the present embodiment described above includes a server 1 and a plurality of autonomous driving vehicles 2, and provides a vehicle allocation service using the autonomous driving vehicle 2. The server 1 learns a learning model from among the autonomous driving vehicles 2 waiting for the vehicle allocation service based on the learning progress of the learning model used in the autonomous driving vehicle 2 received from the autonomous driving vehicle 2. Is configured to select the required learning vehicle and send learning instructions to the learning vehicle. Upon receiving the learning instruction, the autonomous driving vehicle 2 is configured to perform traveling for learning the learning model while waiting for vehicle allocation based on the learning instruction.

As a result, a learning vehicle that requires learning of the learning model (for example, a vehicle with a low learning progress of the learning model) is selected from the automatic driving vehicles 2 waiting for the vehicle allocation service, and the learning vehicle is selected. , It is possible to carry out the running for learning the learning model while waiting for the vehicle to be dispatched. Therefore, it is possible to suppress a difference in the learning progress of the learning model used in each autonomous driving vehicle 2.

If there is a difference in the learning progress of the learning model used in each autonomous driving vehicle 2, for example, the quality of the vehicle allocation service for the passengers of the vehicle allocation service to which the vehicle for which the learning of the learning model has not been sufficiently advanced is sufficiently determined. There is a risk of inequality among the passengers of the vehicle dispatch service, for example, because the quality of the vehicle dispatch service for the passengers to whom the vehicle has been sufficiently trained in the learning model is lower than the quality of the vehicle dispatch service.

Therefore, by suppressing the difference in the learning progress of the learning model used in each autonomous driving vehicle 2 as in the present embodiment, it is possible to suppress the occurrence of inequality among the users of the vehicle allocation service. be able to.

Although the embodiments of the present invention have been described above, the above embodiments are only a part of the application examples of the present invention, and the technical scope of the present invention is limited to the specific configuration of the above embodiments. do not have.

For example, in the above embodiment, the server 1 that has received the vehicle allocation request refers to the vehicle allocation status of each autonomous driving vehicle 2 and the vehicle allocation reservation information, and selects a vehicle allocation vehicle from each autonomous driving vehicle 2. With reference to the learning progress of the NN model of each autonomous driving vehicle 2, a vehicle having a low learning progress may have a lower priority when selected as a vehicle to be dispatched as compared with a vehicle having a high learning progress. As a result, it becomes difficult for a vehicle having a low learning progress to be selected as a vehicle dispatch vehicle, so that there is a tendency to have time to run for learning the NN model while waiting for the vehicle dispatch service. As a result, it is possible to improve the learning progress by increasing the opportunity and frequency of running for learning the NN model while waiting for the vehicle dispatch service.

1 Server 2 Self-driving vehicle 100 Vehicle dispatch system

Claims (1)

It is a vehicle allocation system that includes a server and a plurality of autonomous driving vehicles and provides a vehicle allocation service using the autonomous driving vehicle.
Based on the learning progress of the learning model used in the autonomous driving vehicle received from the autonomous driving vehicle, the server selects the learning model from among the autonomous driving vehicles waiting for the vehicle allocation service. It is configured to select a learning vehicle that needs learning and send a learning instruction to the learning vehicle.
Upon receiving the learning instruction, the autonomous driving vehicle is configured to perform traveling for learning the learning model while waiting for vehicle allocation based on the learning instruction.
Vehicle dispatch system.

Images