Control device

Abstract

A control device according to one side of the present disclosure generates one or more paths by executing calculations of one or more trained control models from among a plurality of trained control models, monitors the behavior of a target user, and determines the behavior of the target user on the basis of a result obtained by monitoring the behavior of the target user. And a control unit that selects one path from among the generated one or more paths, and controls the movement of the moving body in accordance with the selected path. Each trained control model obtains, through machine learning, the ability to generate a path associated with a corresponding behavior of a user for controlling movement of a moving body. As a result, the accuracy of executing the control of the moving body in accordance with the behavior of the user is improved.

Description

Technical Field

[0001] The present disclosure relates to control technology for autonomous driving vehicles. Background Art

[0002] Patent Document 1 proposes a system for autonomous vehicle control, which is configured to use a trained neural network to determine vehicle commands based on a route, GPS data, and sensor data.

[0003] Prior art literature

[0004] Patent Document 1: Japanese Patent Application No. 2019-533810 Summary of the invention

[0005] Technical problem to be solved by the invention

[0006] One of the objects of the present disclosure is to provide a technology for achieving improved accuracy in executing control of a moving object that is suitable for the behavior of a user.

[0007] Technical solutions to solve problems

[0008] The control device involved in the first technical solution of the present disclosure comprises a storage unit for storing a plurality of trained control models and a control unit. The control unit is configured to execute: generating one or more paths by executing the operation of one or more trained control models among the plurality of trained control models; monitoring the behavior of the target user; selecting a path from the one or more generated paths according to the result obtained by monitoring the behavior of the target user; and controlling the movement of the mobile body according to the selected path. Each of the trained control models acquires the ability to generate a path for controlling the movement of the mobile body that is associated with the corresponding behavior of the user through machine learning. Each trained control model can be composed of a neural network, and deep learning can be used as the machine learning method.

[0009] Effects of the Invention

[0010] According to the present disclosure, it is possible to achieve improvement in the accuracy of executing control of a moving object suitable for the behavior of a user. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 An example of a scenario in which the present disclosure is applied is shown in a schematic manner.

[0012] Figure 2 An example of a control model structure of the present disclosure is shown in a schematic manner.

[0013] Figure 3 An example of the second control model structure of the present disclosure is schematically shown.

[0014] Figure 4 An example of the hardware configuration of the control device of the present disclosure is schematically shown.

[0015] Figure 5 An example of a processing procedure related to control performed by the control device of the present disclosure is shown.

[0016] Description of symbols

[0017] 1 control device; 11 control unit; 12 storage unit; 30, 35 trained control models; 50, 55 paths; M mobile body; OU target user; S sensor; MD monitoring device DETAILED DESCRIPTION

[0018] According to the previous methods of Patent Document 1, it is possible to construct an automatic driving system by using a trained machine learning model. However, the inventors of the present application have found that the previous methods have the following problems. For example, it is assumed that a specific mode of driving such as lane change and road shoulder avoidance is performed according to the user's behavior. The user's behavior is, for example, the operation of a turn signal, the driver's abnormality, etc. For collecting training data for each mode of driving based on such user behavior, the pattern is huge and difficult. In addition, the generation frequency may also be very different. It is assumed that under the condition of a mixture of such scenes with greatly different generation frequencies, the user's behavior is accepted as input to generate a trained machine learning model for generating a path suitable for each behavior. In this case, it may be difficult to perform the control of the vehicle suitable for the user's behavior with high accuracy in the obtained trained model. For example, the learning of the user's behavior with an extremely low generation frequency will become insufficient, which may lead to a decrease in the accuracy of the vehicle control performed by the trained model. In addition, this problem may occur regardless of the type of vehicle. In addition, the generation of such a problem is not limited to the scene of controlling the vehicle. In terms of controlling the movement, it is the same even for mobile bodies other than vehicles. Therefore, the same problem may occur in a scenario where all moving objects other than vehicles are controlled.

[0019] In this regard, the control device involved in the first technical solution of the present disclosure includes a storage unit storing a plurality of trained control models and a control unit. The control unit is configured to perform: generating one or more paths by executing operations of one or more trained control models among the plurality of trained control models; monitoring the behavior of the target user; selecting a path from the one or more generated paths based on the results obtained by monitoring the behavior of the target user; and controlling the movement of the mobile body according to the selected path. Each trained control model acquires the ability to generate a path for controlling the movement of the mobile body that is associated with the corresponding behavior of the user through machine learning.

[0020] In the first technical solution of the present disclosure, each trained control model is prepared according to the movement mode associated with the user's behavior. That is, each trained control model is not responsible for movement performed by other modes, and therefore, in the machine learning of each trained control model, the mixed existence of data of other movement modes can be suppressed. Ideally, learning data can be collected specifically for the corresponding movement mode, and a trained model can be generated using the obtained learning data. In the first technical solution of the present disclosure, a trained control model is prepared according to the movement mode, and a path generated by the trained control model corresponding to the behavior of the target user is used for the control of the mobile body. As a result, it can be expected that the accuracy of executing the control of the mobile body suitable for the user's behavior will be improved.

[0021] In addition, as another form of the control device involved in the above technical solution, one aspect of the present disclosure may be an information processing method that realizes all or part of the above constituent elements, or a program, or a storage medium readable by a device such as a computer storing such a program. Here, a storage medium readable by a device refers to a medium that stores information such as a program through electrical, magnetic, optical, mechanical or chemical effects.

[0022] [1 Application example]

[0023] Figure 1 An example of a scenario in which the present disclosure is applied is shown in a schematic manner. The control device 1 involved in the present embodiment is one or more computers configured to control the automatic movement of the target moving body M. In the present embodiment, the control device 1 is mounted on the moving body M and holds a plurality of trained control models 30. Each trained control model 30 acquires the ability to generate a path for controlling the movement of the moving body that is associated with the corresponding behavior of the user through machine learning.

[0024] The control device 1 generates one or more paths 50 by executing the operation of one or more trained control models 35 among the plurality of trained control models 30. The control device 1 monitors the behavior 40 of the target user OU, and selects a path 55 from the generated one or more paths 50 based on the result of monitoring the behavior 40 of the target user OU. Then, the control device 1 controls the movement of the moving body M according to the selected path 55.

[0025] In this embodiment, a trained control model 30 is prepared according to a movement mode associated with the user's behavior, and a path 55 generated by the trained control model 35 corresponding to the behavior 40 of the target user OU is used to control the moving body M. Each trained control model 30 can be generated specifically for the corresponding movement mode. Therefore, according to this embodiment, it can be expected that the accuracy of executing the control of the moving body suitable for the user's behavior can be improved.

[0026] Furthermore, in the present embodiment, the behavior 40 of the target user OU may not be used as an input to each trained control model 30. That is, the behavior 40 of the target user OU may not be used as an explanatory variable for generating the path of each trained control model 30. Thus, the structure of the control model 30 can be simplified, and as a result, the generation accuracy of the path based on each trained control model 30 can be expected to be improved.

[0027] In addition, a method of performing the operation of the trained control model after selecting the trained control model corresponding to the user's behavior can be considered as another method. However, in this method, the time until the user's behavior is reflected in the control of the mobile body will be longer, corresponding to the execution of the operation processing of the trained control model after selecting the trained control model corresponding to the user's behavior. In contrast, the control device 1 involved in this embodiment can perform the operation of all trained control models 35 with the probability of use regardless of whether they are used or not, and select the path 55 to be used from the operation results (one or more paths 50) according to the behavior 40 of the object user OU. As a result, the time until the user's behavior is reflected in the control of the mobile body can be shortened.

[0028] (Moving body)

[0029] As long as it can be automatically moved by mechanical control, the type of the moving body M can be appropriately selected according to the embodiment. The moving body M can be, for example, a vehicle, an aircraft, a ship, a robot device, or other movable device. The aircraft can be at least one of an unmanned aerial vehicle (drone) or a manned aircraft. In one example, Figure 1 As shown, the mobile body M may be a vehicle. In this case, it is expected that the accuracy of executing the autonomous driving suitable for the user's behavior will be improved. In addition, the type of vehicle (number of wheels, power source, size, etc.) can be selected arbitrarily. As a typical example, the mobile body M may be a car with a level 2 or higher autonomous driving capability.

[0030] (Control the action)

[0031] In one example, the movement of the object's moving body M can be controlled by directly controlling the object's moving body M. In another example, the moving body M can be provided with a dedicated control device such as a controller. In this case, the movement of the object's moving body M controlled by the control device 1 can be controlled by providing the derived result to the dedicated control device, thereby indirectly controlling the object's moving body M. The control device 1 can be installed in any place. In one example, as Figure 1 As shown, the control device 1 can be mounted on the mobile body M. In another example, the control device 1 can also be configured away from the mobile body M to control the mobile body M remotely. The control device 1 can be configured to be able to switch from the automatic control mode of the mobile body M to the manual control mode, and from the manual control mode to the automatic control mode at any timing through any operation. In addition, when the mobile body M is a vehicle, typically, the target user OU can be the driver or a passenger other than the driver. The target user OU can be appropriately determined according to the mobile body M.

[0032] (more than one trained control model)

[0033] In the present embodiment, it is not necessary to always perform all operations of the prepared plurality of trained control models 30 to generate a path. A trained control model 35 having a probability of being used can be selected from the prepared plurality of trained control models 30 by any method. In one example, one or more trained control models 35 to be used in the generation of a path can be selected according to a moving scenario such as a position, a route, a type of passage (for example, a highway, a general road, etc. in the case of a vehicle). As a specific example, the following scenario is assumed: the moving body M is a vehicle, and the control device 1 has a first control model for generating a path for a left lane change, a second control model for generating a path for a right lane change, and a third control model for generating a path for lane maintenance as the trained control models 30. In this scenario, in the scenario of driving in the leftmost lane, the control device 1 generates a path using the second control model and the third control model, respectively. In contrast, since there is no room for selecting a path of the first control model, the generation of a path based on the first control model can be omitted. On the other hand, in a scenario where lane changes can be made in both left and right directions, the control device 1 may generate a path using each of the first to third control models.

[0034] (path)

[0035] As long as the movement of the mobile body M can be controlled, the path (output of the control model) may not be particularly limited in form and may be appropriately determined according to the implementation method. In one example, the path (50, 55) may be composed of one or more control instructions. The control instructions may be configured to represent the control amount of the mobile body M. In another example, the path (50, 55) may be configured to represent the future movement path of the mobile body M and is used to derive one or more control instructions. In this case, one or more control instructions may be determined based on the path 55 by any method. The control model (30, 35) may also be referred to as a path planner.

[0036] In addition, the control instructions are related to the action of the mobile body M. The composition of the control instructions can be appropriately selected according to the implementation method. In one example, the control instructions can be composed of acceleration, deceleration, steering, or a combination of these. Acceleration and deceleration may include shifting. The control instructions can be composed, for example, to represent the control quantity (control indication value, control output quantity) of the mobile body M such as the accelerator control quantity, the brake control quantity, the steering wheel steering angle, etc. In addition, the control instructions may also include instructions related to the operation of the mobile body M. As an example, in the case where the mobile body M is a vehicle, the control instructions may include vehicle operations such as turn indicators, hazard warning lights, horns, communication processing (such as sending data to the center, making emergency calls, etc.).

[0037] (Trained control model)

[0038] The control model 30 is composed of a machine learning model having one or more operation parameters that can be adjusted by machine learning. One or more operation parameters are used for the operation of inference as the purpose (path derivation in this application). Machine learning refers to the use of learning data to adjust (optimize) the value of the operation parameter. The structure and type of the machine learning model may not be specifically limited, and may be appropriately selected according to the implementation. The machine learning model may be composed of, for example, a neural network, a support vector machine, a regression model, a decision tree model, etc. The method of machine learning may be appropriately selected according to the machine learning model adopted (such as the error back propagation method, etc.). As an example, at least any one of the multiple control models 30 may be at least partially composed of a neural network. The structure of the neural network may be appropriately determined according to the implementation, for example, by the number of layers from the input layer to the output layer, the type of each layer, the number of nodes (neurons) contained in each layer, the connection relationship between the nodes of each layer, etc. In one example, the neural network may have a recursive structure. In addition, the neural network may include, for example, any layer such as a fully connected layer, a convolutional layer, a pooling layer, a deconvolutional layer, an unpooling layer, a normalization layer, a dropout layer, and an LSTM (Long short-term memory). The neural network may have any mechanism such as an attention mechanism. The neural network may include any model such as a GNN (Graph neural network), a diffusion model, a generative model (such as a Generative Adversarial Network, a Transformer, etc.). When a neural network is used in a control model, the weights of the connections between the nodes contained in the control model and the thresholds of each node are examples of operating parameters. In addition, when a machine learning model is used, the control model may be composed of an end-to-end model structure.

[0039] Each control model 30 is configured to derive a path based on the environment of the mobile body M. The environment is an event observed by at least one of the mobile body M itself and the surroundings. In one example, at least a portion of the environment can be observed by one or more sensors S arranged inside or outside the mobile body M. The type of sensor S is not particularly limited as long as it can observe any environment in which the mobile body M moves, and can be appropriately selected according to the implementation method. In one example, the one or more sensors S may include a camera (image sensor), a radar, a LiDAR (Light Detection And Ranging)), a sonar (ultrasonic sensor), an infrared sensor, a GNSS (Global Navigation Satellite System) / GPS (Global Positioning Satellite) module, etc.

[0040] As long as the path can be derived according to the environment of the mobile body M, the input and output methods of the control model 30 may not be particularly limited and may be appropriately selected according to the implementation method. In one example, at least any one of the multiple control models 30 may be configured to derive a path based on the observation data 120 of the sensor S at more than one time point. In another example, at least any one of the multiple control models 30 may be configured to derive a path based on the recognition result of the surrounding environment. In this case, the control device 1 may also have an analytical model that infers the recognition result of the surrounding environment based on the observation data 120 of the sensor S. Alternatively, the trained control model 30 may include the analytical model. The analytical model may be arbitrarily constructed. In one example, the analytical model may be composed of a machine learning model. In addition, other information may be arbitrarily added to the input of at least any one of the multiple control models 30. For example, at least any one of the multiple control models 30 may be configured to further accept the input of any information such as a set speed, a speed limit, a location, map information, and navigation information. In addition, in one example, the control model 30 may be configured to directly output the path. In another example, the control model 30 may be configured to indirectly output the path, and the path may be obtained by performing arbitrary information processing (interpretation processing) on ​​the output of the control model 30 .

[0041] In addition, typically, multiple trained control models 30 can be constructed separately (independently). However, the construction of multiple trained control models 30 may not be limited to such an example. In an example of the present embodiment, two or more control models among the multiple trained control models 30 may also be constructed in an integrated manner. For example, one model may have n output parts (n is a natural number greater than 2) that derive paths according to the above-mentioned analytical model configured on the input side and the output of the analytical model, respectively. The one model can be regarded as n trained control models 30. That is, in the present embodiment, when the structure of the model is integrated but the model has n output parts that are configured to output paths respectively, maintaining the model can be regarded as maintaining n trained control models 30. In addition, the conditional model configured to output in accordance with the provided conditions can also be processed in the same manner. That is, multiple trained control models 30 may include conditional models configured to output paths corresponding to the input conditions (class / category). In this case, by changing the conditions provided to the conditional model and repeatedly performing the operation of the conditional model, multiple paths corresponding to the conditions provided respectively can be derived. Therefore, in the case of constructing a conditional model corresponding to n conditions, maintaining the one conditional model can be regarded as maintaining n trained control models 30.

[0042] (Monitoring of behavior)

[0043] Monitoring behavior 40 may be constituted by monitoring at least one of no specific input and specific input. Monitoring specific input may include identifying the type (indication) of the input.

[0044] In addition, the behavior 40 of the object user OU can be monitored by any method. In one example, the control device 1 can monitor the behavior 40 of the object user OU through the presence or absence of input of images, sounds, operating parts, etc. In the case of monitoring the behavior 40 through images, the behavior 40 may include gestures, the state of the object user OU (for example, the object user OU falls into an abnormal state, etc.). The operating parts may include all kinds of devices related to the operation of the moving body M. In one example, when the moving body M is a vehicle, the operating parts may include, for example, operating parts provided on the vehicle, such as turn indicators, hazard warning lights, and horns.

[0045] In one example of this embodiment, a monitoring device MD may be used to monitor the behavior 40. The monitoring device MD may include, for example, a camera (image sensor), a microphone, an operating element, a sensor for measuring the amount of operation of the operating element, etc. The behavior 40 of the target user OU may be identified by any method based on the observation data 125 obtained in real time by the monitoring device MD. The behavior 40 of the target user OU may be identified based on the observation data 125 by, for example, well-known analysis techniques such as image analysis, sound analysis, and analysis of the operation content of the operating element.

[0046] (Correspondence between behavior and control model)

[0047] The correspondence between the prepared trained control model 30 and the user's behavior can be appropriately determined according to the implementation. That is, for what kind of user's behavior the trained control model 30 is prepared, it can be appropriately determined according to the implementation. In addition, for what kind of movement path the prepared control model 30 corresponding to the behavior obtains, it can also be appropriately determined according to the implementation.

[0048] Figure 2 An example of a control model structure involved in this embodiment is shown in a schematic manner. Figure 2 In an example, the plurality of trained control models 30 include a first trained control model 31 that acquires the ability to generate a path without a user's instruction. Accordingly, the one or more trained control models 35 may include the first trained control model 31. Selecting a path 55 may include selecting a path generated by the first trained control model 31 without an instruction from the target user OU. Thus, in a scenario where there is no arbitrary instruction as the behavior 40 of the target user OU, automatic control of the mobile body M can be implemented. Furthermore, in an example, the absence of an instruction may be constituted by a specific instruction to the mobile body M without an input based on an image, a sound, an operating member, or the like.

[0049] In addition, Figure 2In the example, the plurality of trained control models 30 also include one or more second trained control models 32, each of which has the ability to generate a path under a condition with a corresponding instruction from a user. Accordingly, the one or more trained control models 35 may include one or more second trained control models 32. Selecting a path 55 may include selecting a path generated by any one of the one or more second trained control models 32 according to an instruction from the object user OU. Thus, in the case where there is an arbitrary instruction as the behavior 40 of the object user OU, automatic control of the moving body M corresponding to the instruction can be implemented. In addition, in one example, the instruction may be constituted by a specific instruction having an input based on an image, sound, an operating member, etc. to the moving body M. The instruction content and the ability of the second trained control model 32 may be appropriately determined according to the implementation method.

[0050] Figure 3 An example of the structure of the second trained control model 32 according to the present embodiment is shown schematically. Figure 3 As shown, the moving body M may be a vehicle, and the one or more second trained control models 32 may include a trained lane change model 33, which has acquired the ability to generate a path for controlling the movement of the moving body M (vehicle) under a condition with an indication of lane change. Accordingly, selecting a path 55 may include selecting a path generated by the trained lane change model 33 according to the indication of lane change of the target user OU.

[0051] The lane change indication method (behavior 40) can be appropriately determined according to the implementation method. In one example, the lane change indication can be performed by any input such as gesture (image) input, voice input, operation of a specific operating element (such as right turn signal operation or left turn signal operation). In addition, the trained lane change model 33 can be configured to generate a path for executing at least one of a change to the right lane and a change to the left lane. In the case of being configured to be able to execute both a right lane change and a left lane change, separate lane change models can be prepared for the right lane change and the left lane change, respectively. Alternatively, a lane change model that can correspond to lane changes on both the left and right sides can also be prepared.

[0052] In addition, the type of the second trained control model 32 may not be limited to the example of the trained lane change model 33, and may be appropriately determined according to the implementation method. As another example, when the moving body M is a vehicle, the second trained control model 32 may include a lane keeping model, an emergency stop model, and the like. The lane keeping model may be configured to generate a path for maintaining the lane in which the vehicle is traveling. The lane keeping model may also be included in the first trained control model 31 or the trained lane change model 33. The emergency stop model may be configured to generate a path for retreating the vehicle to the shoulder of the road based on an abnormal state of the user, such as the driver losing consciousness. In addition, the correspondence between the user's behavior and the control model 30 may not be limited to the above-mentioned example. In another example, at least one of the above-mentioned first trained control model 31 and the second trained control model 32 may be omitted.

[0053] (Path selection)

[0054] exist Figure 3 In the scenario SC1, there are other lanes on both sides of the lane in which the vehicle is traveling, so lane keeping and lane changes to the left and right can be performed. On the other hand, in the scenario SC2, the vehicle is traveling in the leftmost lane, so lane keeping and right lane changes can be performed, but left lane changes cannot be performed. Therefore, in the scenario SC2, the path of the preferred left lane change is excluded. The possibility that movement that cannot be safely performed according to such a movement scenario of the moving body M is not limited to the case where the moving body M is a vehicle. In the case where the moving body M is not a vehicle, movement that cannot be safely performed may also occur according to the movement scenario of the moving body M.

[0055] To cope with this, in one example, the control device 1 may exclude the trained control model 30 that generates an unexecutable path according to the movement scenario when selecting one or more trained control models 35 from the plurality of trained control models 30 as described above. That is, the control device 1 may exclude the unexecutable path by not selecting the trained control model 30 that generates the unexecutable path as the trained control model 35 according to the movement scenario.

[0056] In another example, on the basis of this method or instead of this method, the control device 1 can evaluate the safety of each of the one or more generated paths 50. Selecting a path 55 can be constituted by selecting a path 55 from the one or more generated paths 50 after excluding the path evaluated as having low safety. Whether the safety is low can be determined by any method. Typically, it can be determined whether the safety is low based on the comparison of the calculated safety with a threshold. The threshold can be appropriately determined according to the implementation method. In this way, it is possible to exclude paths that cannot be executed.

[0057] In addition, the method for evaluating the safety degree can be appropriately selected according to the implementation method. In one example, the trained control model (30, 35) can be configured to be able to derive a path and a safety degree by assigning a true value of the safety degree to the training data during machine learning (i.e., providing a teacher signal of the safety degree). In this case, the safety degree can be evaluated by executing the operation of the trained control model 35 to calculate the safety degree of the derived path 50. In another example, the safety degree of the generated path 50 can be evaluated based on a simulation. For example, the movement based on the generated path 50 can be simulated in real time by the control device 1 or an external computer (such as a server device), so as to evaluate the safety degree of the generated path 50. In addition, for example, by collecting the results of the simulation, multiple data sets (data sets) each consisting of a combination of the generated path (training data) and the evaluated safety degree (teacher signal / label) can be obtained. The training data may also include the environment of the mobile body M when the corresponding path is generated or the recognition result of the environment. A trained machine learning model that has acquired the ability to calculate the safety degree based on the generated path can be generated by machine learning using the obtained multiple data sets. The control device 1 can calculate the safety level of each of the one or more paths 50 by providing the generated one or more paths 50 to the trained machine learning model respectively.

[0058] In another example, the safety level can be evaluated based on whether the input data (e.g., the observation data 120 of the sensor S, the recognition result of the environment, etc.) provided to the trained control model 35 when generating the path 50 is within the scope of machine learning. For example, an autoencoder can be generated together with the trained control model (30, 35) or separately from the trained control model (30, 35) through machine learning. If the input data is within the scope of machine learning, the reconstruction error of the autoencoder is small, otherwise, the reconstruction error of the autoencoder becomes larger. That is, the larger the reconstruction error, the lower the safety level can be evaluated. Therefore, the control device 1 can perform the operation of the autoencoder together with the operation of one or more trained control models 35, calculate the difference between the input data and the output data, and thus calculate the reconstruction error. The control device 1 can evaluate the safety level of the generated path 50 based on the calculated reconstruction error. In short, the reconstruction error can be directly used as the safety level.

[0059] (Select a path)

[0060] Selecting a path 55 from one or more paths 50 may include selecting a path from the generated multiple paths and directly using the generated one path. That is, the control device 1 according to the present embodiment may have a first mode of generating multiple paths and selecting a path from the generated multiple paths, and may also have a second mode of at least temporarily generating only one path 50 from a trained control model 35 and directly using the generated one path 50 as the path 55 for controlling the moving body M. The second mode may also be omitted.

[0061] [2 Configuration example]

[0062] Figure 4 An example of a hardware configuration of the control device 1 according to the present embodiment is schematically shown. The control device 1 according to the present embodiment is a computer in which a control unit 11, a storage unit 12, an external interface 13, an input device 14, an output device 15, and a driver 16 are electrically connected.

[0063] The control unit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), etc., and is configured to perform arbitrary information processing based on programs and various data. The control unit 11 (CPU) is an example of a processor resource. The storage unit 12 can be composed of any storage device such as a hard disk drive, a solid state hard disk drive, etc. The storage unit 12 (as well as RAM, ROM) is an example of a storage unit of the present disclosure. In addition, the storage unit 12 (as well as RAM, ROM) is an example of a memory resource. In the present embodiment, the storage unit 12 stores various information such as the control program 81 and the learning result data 300.

[0064] The control program 81 is used to cause the control device 1 to execute information processing related to the control of the moving body M (described later). Figure 5) program. The control program 81 includes a series of commands for the information processing. The learning result data 300 is configured to represent information related to the trained control model 30. In one example, the learning result data 300 can be provided for each trained control model 30. In another example, one piece of learning result data 300 can be configured to represent two or more trained control models 30. In addition, as long as the trained control model 30 can be reproduced, the data structure of the learning result data 300 can be appropriately determined according to the implementation method. In one example, the learning result data 300 may include information representing the values ​​of the operation parameters of the control model 30 adjusted by machine learning. The learning result data 300 may also include information representing the structure of the machine learning model (for example, the structure of a neural network, etc.).

[0065] The external interface 13 may be, for example, a USB (Universal Serial Bus) port, a dedicated port, a wireless communication port, etc., and is configured to be connected to an external device in a wired or wireless manner. In this embodiment, the control device 1 may be connected to the sensor S and the monitoring device MD via the external interface 13. The input device 14 may be, for example, a device for inputting, such as a mouse, a keyboard, or an operating member. The output device 15 may be, for example, a device for outputting, such as a display, a speaker, etc. The input device 14 and the output device 15 may also be integrally configured, for example, by a touch panel display, etc.

[0066] The drive 16 is a device for reading various information such as programs stored in the storage medium 91. At least one of the control program 81 and the learning result data 300 may be stored in the storage medium 91 instead of being stored in the storage unit 12, or in the storage unit 12 and the storage medium 91. The storage medium 91 is configured to store various information (stored programs, etc.) through electrical, magnetic, optical, mechanical or chemical effects so that a computer or other device can read the information. The control device 1 can obtain at least one of the control program 81 and the learning result data 300 from the storage medium 91. In addition, the storage medium 91 may be a disk-type storage medium such as a CD, a DVD, or a storage medium other than a disk-type such as a semiconductor memory (e.g., a flash memory). The type of the drive 16 may be appropriately selected according to the type of the storage medium 91.

[0067] In addition, regarding the specific hardware structure of the control device 1, the components can be appropriately omitted, replaced, and added according to the implementation method. For example, the control unit 11 may also include multiple hardware processors. The hardware processor may be composed of a microprocessor, an FPGA (field-programmable gate array), a DSP (digital signal processor), an ECU (Electronic Control Unit), a GPU (Graphics Processing Unit), etc. At least any one of the external interface 13, the input device 14, the output device 15, and the driver 16 may also be omitted. In addition to a computer designed specifically for the service provided, the control device 1 may be a general-purpose computer, a terminal device, etc. In the case where the mobile body M is a vehicle, the control device 1 may be a vehicle-mounted device.

[0068] [3 Action Examples]

[0069] Figure 5 An example of a processing step related to the control of the mobile body M by the control device 1 involved in this embodiment is shown. The control unit 11 of the control device 1 executes the command included in the control program 81 through the CPU. Thereby, the control device 1 acts as a computer capable of performing the following information processing. The following processing steps are an example of a control method executed by a computer. However, the following processing steps are only an example, and each step can be changed as long as it is feasible. In addition, with respect to the following processing steps, the steps can be appropriately omitted, replaced, and added according to the embodiment.

[0070] In step S101 , the control unit 11 obtains the observation data 120 of the sensor S. The control unit 11 may obtain the observation data 120 from the sensor S directly or indirectly.

[0071] In step S102, the control unit 11 generates one or more paths 50 by executing the operation of one or more trained control models 35 among the plurality of trained control models 30. In one example, the control unit 11 may derive the path 50 by providing at least a portion of the acquired observation data 120 to the trained control model 35 and performing the operation processing of the trained control model 35. Before providing the observation data 120 to the trained control model 35, any pre-processing may be applied to the observation data 120. In another example, the control unit 11 may identify the environment of the moving body M from the observation data 120 by any method, provide the recognition result of the environment to the trained control model 35, perform the operation processing of the trained control model 35, and thereby derive the path 50.

[0072] In an example of this embodiment, the one or more trained control models 35 may include a first trained control model 31, which has the ability to generate a path for controlling the movement of the mobile body M without a specific instruction from the user. In addition, the one or more trained control models 35 may include one or more second trained control models 32, which have the ability to generate a path for controlling the movement of the mobile body M with a corresponding specific instruction from the user. When the mobile body M is a vehicle, the one or more second trained control models 32 may include a trained lane change model 33.

[0073] In addition, before executing step S102 , the control unit 11 may select one or more trained control models 35 having a probability of use from a plurality of trained control models 30 at any timing according to the movement scenario of the mobile body M. In step S102 , the control unit 11 may derive the path 50 using the selected trained control model 35 .

[0074] In step S103, the control unit 11 monitors the behavior 40 of the target user OU. The behavior 40 of the target user OU can be monitored by any method. In one example, the control unit 11 can monitor the behavior 40 of the target user OU by directly or indirectly acquiring the observation data 125 from the monitoring device MD. In this case, the control unit 11 can identify the behavior 40 of the target user OU from the observation data 125 by any method.

[0075] In step S104, the control unit 11 evaluates the safety of each of the one or more generated paths 50. In one example of this embodiment, the control unit 11 can evaluate the safety of the path 50 by any of the above methods.

[0076] In step S105, the control unit 11 selects a path 55 from the one or more generated paths 50 based on the result obtained by monitoring the behavior 40 of the target user OU. In an example of the present embodiment, selecting a path 55 may include selecting a path generated by the first trained control model 31 in the absence of a specific instruction from the target user OU. In the case where there is no trained control model 35 corresponding to the behavior 40 of the target user OU, the control unit 11 may select a path generated by the first trained control model 31, similarly to the case where there is no specific instruction. In addition, in an example of the present embodiment, selecting a path 55 may include monitoring the instruction (behavior 40) of the target user OU, and selecting a path generated by any one of the one or more second trained control models 32 based on the result of the monitoring. In the case where the moving body M is a vehicle, selecting a path 55 may include selecting a path generated by the trained lane change model 33 based on the instruction of the target user OU to change lanes.

[0077] In addition, in an example of the present embodiment, between step S104 and step S105, the control unit 11 may exclude the path evaluated as having low safety among the one or more generated paths 50 based on the result of evaluating the safety. Accordingly, selecting a path 55 may be constituted by selecting a path 55 from the one or more generated paths 50 after excluding the path evaluated as having low safety. Being evaluated as having low safety may mean that the safety does not meet the benchmark. In an example, whether the safety is low may be determined by comparing a threshold value. In this case, the benchmark of the safety is defined by the threshold value.

[0078] In addition, in one example, when the path of the trained control model corresponding to the behavior 40 of the target user OU is excluded based on the evaluation result of the safety, the control unit 11 can feedback to the target user OU that the corresponding path has been excluded by any method such as sound, image, etc. In addition, when the path corresponding to the behavior 40 of the target user OU has been excluded, the control unit 11 can select the path with the highest safety. When the path of the second trained control model 32 corresponding to the specific instruction of the target user OU has been excluded, the control unit 11 can also select the path generated by the first trained control model 31. As a result of the evaluation of the safety, when all the paths 50 are excluded, the control unit 11 can switch from the automatic control mode to the manual control mode, and output a notification to urge any user (typically the target user OU) to manually operate the moving body M.

[0079] In step S106, the control unit 11 controls the movement of the moving object M according to the selected path 55. When the control of the target moving object M is completed, the control unit 11 ends the processing procedure of the control device 1 according to this operation example.

[0080] The control unit 11 can repeatedly perform a series of information processing from step S101 to step S106 at any timing. In one example, the control unit 11 can repeatedly perform a series of information processing from step S101 to step S106 during a predetermined period (for example, during the period when the power source of the mobile body M is activated or the automatic control mode is selected). In this way, the control device 1 can continuously perform automatic control of the mobile body M.

[0081] In addition, the processing order of each step may not be limited to Figure 5 The example can be appropriately changed according to the implementation. For example, the process of step S103 can be performed at any timing before the process of step S105 is performed. The process of step S104 can be performed at any timing after the process of step S102 is performed and before the process of step S105 is performed. The process of step S104 can also be omitted.

[0082] [feature]

[0083] In this embodiment, a plurality of trained control models 30 are prepared according to each movement mode associated with the user's behavior. And, through the processing of steps S102 to S106 (except step S104), the path 55 generated by the trained control model 35 corresponding to the behavior 40 of the target user OU is used for the control of the moving body M. As a result, it can be expected that the accuracy of executing the control of the moving body M suitable for the user's behavior will be improved.

[0084] [4 Modifications]

[0085] The embodiments of the present disclosure have been described in detail above, but the descriptions so far are merely illustrative of the present disclosure in all aspects. It goes without saying that various improvements or modifications can be made without departing from the scope of the present disclosure. As long as no technical contradictions are generated, the processes and means described in the present disclosure can be freely combined for implementation.

Claims

1. A control device comprising a storage unit storing a plurality of trained control models and a control unit, The control unit is composed of: generating one or more paths by executing operations of one or more trained control models among the plurality of trained control models; Monitor the behavior of the target user; selecting a path from the one or more generated paths according to a result obtained by monitoring the behavior of the target user; as well as Control the movement of the moving object according to the selected path. Each of the trained control models acquires the ability to generate a path for controlling the movement of the moving object that is associated with the corresponding behavior of the user through machine learning.

2. The control device according to claim 1, The one or more trained control models include a first trained control model, wherein the first trained control model has acquired the ability to generate a path without the user's instruction, Selecting the one path includes selecting a path generated by the first trained control model without an instruction from the subject user.

3. The control device according to claim 1, The one or more trained control models include one or more second trained control models, each of which has the ability to generate a path under the condition of the corresponding instruction of the user, Selecting the one path includes selecting a path generated by any one of the one or more second trained control models according to an instruction of the target user.

4. The control device according to claim 3, The mobile body is a vehicle, The one or more second trained control models include a trained lane change model, Selecting the one path includes selecting a path generated by the trained lane change model according to the lane change indication of the subject user.

5. The control device according to any one of claims 1 to 4, The control unit is further configured to evaluate the safety of each of the one or more generated paths. The one path is selected by excluding a path evaluated as having a low safety level and then selecting one path from the one or more generated paths.

Images