DE102019118281A1 - Generation device, component control device, generation method, component control method, computer program - Google Patents

Abstract

A data generation device, a component control device, a generation method, a component control method, and a computer program are provided that are capable of achieving automatic control of components that are less likely to cause the driver to be himself in various situations and driving environments feel uncomfortable.
A generation device includes a detection part that detects input information (s) related to driving a human-powered vehicle; and a generation part that generates various learning models, each generating output information (s) regarding the control of a component of the human-powered vehicle based on input information (s) acquired by the detection part become.

Description

CROSS REFERENCE TO OTHER APPLICATIONS

This application claims priority from Japanese patent applications JP 2018-130244 . JP 2018-130245 and JP 2018-130246 , filed on July 9, 2018, and Japanese patent applications JP 2018-236760 . JP 2018-236761 . JP 2018 - 236762 . JP 2018-236763 . JP2018-236764 and JP 2018-236765 , filed on December 18, 2018. The entire disclosure of Japanese Patent Application No. JP 2018-130244 . JP 2018-130245 . JP 2018-130246 . JP 2018-236760 . JP 2018 - 236761 . JP 2018-236762 . JP 2018-236763 . JP 2018-236764 and JP 2018-236765 are incorporated herein by reference.

The present invention relates to a generation device for generating data relating to the control of components of a human-powered vehicle, a component control device, a generation method, a component control method, and a computer program.

Human powered vehicles that at least partially use human power are known, including a bicycle, a assisted bicycle, an electric bicycle called an e-bike, and so on. The human powered vehicle is provided with a multi-speed transmission device which shifts gears after a gear shift by a driver. An automatic gear shift control system has been conventionally proposed for automatically controlling a gear change using sensors such as a speed sensor, a cadence sensor, a chain tension sensor, etc., and for performing various calculations on an output of each of the sensors. Such a system is in the U.S. Patent No. 6047230 disclosed.

The automatic control of components, including the in U.S. Patent No. 6047230 of the disclosed transmission or the like is achieved by combinations of determinations whether or not each of the numerical values obtained from the various sensors, e.g., a speed sensor, a torque sensor, etc., mounted on the human powered vehicle is larger than a predetermined threshold ,

However, the determination using a threshold value is not sufficient to achieve the desired automatic control.

Even if the same user drives the same human-powered vehicle, the details of the component control depend on situations, whether the driving purpose is movement or racing and so on, and the driving environment.

An object of the present invention is to provide a data generation device, a component control device, a generation method, a component control method, and a computer program that achieve automatic control of components that are less likely to make the driver uncomfortable even in various situations or driving environments feels

(1) The generation device according to a first aspect of the present invention includes a detection part that detects input information (s) regarding the driving of a human-powered vehicle; and a generation part that generates various learning models each generating output information (s) related to the control of a component of the human-powered vehicle based on input information (s) acquired by the detection part become.

Thus, automatic control of the components where the driver is less likely to feel uncomfortable can be achieved through different learning models to suit different situations or driving environments.

(2) In a generating device according to a second aspect of the present invention, the output information (s) in the generating device according to the first aspect described above contains at least one of a gear stage and a gear ratio for a transmission ).

Thus, automatic control of the gear shift can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(3) In a generation device according to the third aspect of the present invention, the input information (s) in the generation device according to the above-described first or second aspect includes at least one of a traveling speed of the human-powered vehicle and a cadence of a crank of a drive mechanism ,

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(4) In a generation device according to a fourth aspect of the present invention, the input information (s) in the generation device according to one of the first to third aspects described above includes detection data of an attitude of the human-powered vehicle.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(5) In a generation device according to a fifth aspect of the present invention, the input information (s) in the generation device according to one of the first to fourth aspects described above includes detection data of a position of a user who drives the human-powered vehicle.

In this way, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(6) In a generation device according to a sixth aspect of the present invention, the input information (s) in the generation device according to one of the first to fifth aspects described above includes detection data of a driving environment of the human-powered vehicle.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(7) In a generation device according to a seventh aspect of the present invention, the generation part (s) inputs input information (s) that, in combination of a combination of information for another one of the learning models, from the input information acquired by the detection part (s) differs for each of the different learning models.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(8) In a generation device according to an eighth aspect of the present invention, in the generation device according to one of the first to seventh aspects described above, the setting of the input information different in combination is accepted for each of the different learning models.

An automatic control can thus be achieved which is adapted to the preferences of the user in order to be suitable for different situations or driving environments.

(9) In a generation device according to a ninth aspect of the present invention, the generation part in the generation device according to one of the first to eighth aspects described above arranges a plurality of data elements contained in the input information (s) acquired by the detection part, for each of the different learning models, weightings that differ from those of another learning model and enter the multiple data elements that were weighted.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(10) In a generation device according to a tenth aspect of the present invention, in the generation device according to one of the first to ninth aspects described above, the setting of the different weights for each of the different learning models is accepted.

An automatic control can thus be achieved which is adapted to the driver's preferences in order to be suitable for different situations or driving environments.

(11) In a generation device according to an eleventh aspect of the present invention, the detection part in the generation device according to one of the first to tenth aspects described above acquires input information at different times, and the generation part gives the input information (s) ) at different times, preferably in each of the learning models.

Thus, the learning model that outputs output information (s) related to the control based on the input information obtained at different times is generated to be suitable for the various situations and driving environments, which enables automatic control to be achieved for which it it is unlikely to cause discomfort.

(12) In a generation device according to a twelfth aspect of the present invention, in the generation device according to the eleventh aspect described above, an interval between the times at which the input information is acquired differs between the different learning models.

Thus, the learning model that outputs output information (s) related to the control based on the input information obtained at different times is generated to be suitable for the various situations and driving environments, which enables automatic control to be achieved for which it it is unlikely to cause discomfort.

(13) In a generation device according to a thirteenth aspect of the present invention, the detection part in the generation device according to one of the first to tenth aspects described above acquires input information at different times and the generation part inputs a variation between the input information into the learning model ,

Thus, the learning model that outputs output information (s) regarding the control based on the variation between the input information (s) obtained over multiple time points is generated to be suitable for the various situations and driving environments is what enables automatic control to be achieved that is unlikely to cause discomfort.

(14) In a generating device according to a fourteenth aspect of the present invention, in the generating device according to the thirteenth aspect described above, a time period corresponding to the variation between the input information (s) differs between the different learning models.

Therefore, the learning model that outputs output information (s) related to the control based on the input information (s) obtained over multiple time points is generated to be suitable for the various situations and driving environments is what enables automatic control to be achieved that is unlikely to cause discomfort.

(15) A generating device according to a fifteenth aspect of the present invention further includes, in the generating device according to any one of the first to fourteenth aspects described above, a selection part that selects a target learning model from among the various learning models, which Generation part generates a target learning model selected by the selection part.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(16) The generating device according to a sixteenth aspect of the present invention further includes, in the generating device according to the fifteenth aspect described above, an evaluation part that outputs the output information (s) output from a target learning model selected by the selection part (s) ), the generation part (s) updating the target learning model based on an evaluation by the evaluation part (s).

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(17) The generating device according to a seventeenth aspect of the present invention further includes, in the generating device according to the sixteenth aspect described above, an operating part that accepts a determination operation regarding the output information (s), the evaluating part performing an evaluation, by checking the output information (s) output by the target learning model in response to an input of the input information (s) acquired by the detection part into the target learning model against a determination operation accepted by the operating part.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(18) If, in a generating device according to an eighteenth aspect of the present invention, the output information (s) output from the target learning model in the generating device according to the seventeenth aspect described above and a detail of the determining operation do not match, that provides Assessment part of the output information (s) with a low assessment regarding the target learning model.

This makes it possible to achieve automatic control that is adapted to the operation of the driver.

(19) In a generating device according to a nineteenth aspect of the present invention, the generating part in FIG The generation device according to the seventeenth aspect described above, when the output information (s) output from the target learning model and a detail of the determination operation do not match, by assigning a weight based on training data including the details of the determination operation.

This makes it possible to achieve an automatic control that is adapted to the operation of the driver.

(20) In a generation device according to a twentieth aspect of the present invention, the detection part in the generation device according to the seventeenth aspect described above includes a storage part that temporarily stores multiple input information in time series, and when a determination operation of that If the operating part is accepted, the generating part sets a plurality of input information, input before and after a point in time at which the determination operation is carried out, and updates the target learning model using the input data and a detail of the operation by the operating part.

Thus, an automatic control that is less likely to make the driver feel uncomfortable can make changes accordingly to suit different situations or driving environments.

(21) In a generation device according to the twenty-first aspect of the present invention, the generation part in the generation device according to the twentieth aspect described above updates the target learning model by increasing the number of detections of input information (s) and input information (s) ) when a determination operation is accepted by the control panel.

Therefore, if the output of the learning model and the operation by the driver do not coincide with each other, intensive learning is performed to thereby achieve automatic control that is unlikely to cause discomfort.

(22) In a generation device according to the twenty-second aspect of the present invention, the generation part in the generation device according to the twentieth aspect described above updates the target learning model by increasing a frequency of acquisition of input information (s) when a determination operation is accepted by the operating part becomes.

Therefore, if the output of the learning model and the operation by the driver do not coincide with each other, intensive learning is performed to thereby achieve automatic control that is unlikely to cause discomfort.

(23) In a generation device according to the twenty-third aspect of the present invention, the evaluation part in the generation device according to the sixteenth aspect described above includes an evaluation acceptance part that accepts a rating that is accepted by the user for the action of the component on the device Is provided based on output information (s) output from each of the various learning models, and the target learning model is updated based on training data, including a detail of the rating accepted by the evaluation acceptance part, output information (s) provided by the learning model is / are output at a time when a rating is accepted and input information (s) are updated in connection with the output information (s).

This makes it possible to achieve automatic control that is adapted to the driver's preferences.

(24) A generating device according to the twenty-fourth aspect of the present invention further includes, in the generating device according to the twenty-third aspect described above, a control panel that accepts a determination operation regarding the output information (s), and the evaluation acceptance part is on the control panel or in the vicinity of the control panel to accept an evaluation operation by the user.

This makes it possible to achieve automatic control that is adapted to the driver's preferences.

(25) In a generation device according to the twenty-fifth aspect of the present invention, the evaluation acceptance part in the generation device according to the twenty-third aspect described above includes a specification part that specifies facial expressions from a photographed image obtained by photographing a face of a user obtained driving the human powered vehicle and accepting a user-made evaluation based on the facial expressions specified in the specified part.

This makes it possible to achieve automatic control that is adapted to the driver's preferences.

(26) In a generation device according to the twenty-sixth aspect of the present invention, the evaluation acceptance part in the generation device according to the twenty-third aspect described above includes a voice recognition part and accepts a rating by recognizing a user's voice.

This makes it possible to achieve automatic control that is adapted to the driver's preferences.

(27) A generating device according to the twenty-seventh aspect of the present invention further includes, in the generating device according to the first to twenty-sixth aspects described above, a model selection part that accepts the selection of one of the various generated learning models and a transmission part, that transmits the learning model selected by the model selection part.

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(28) A component control device according to the twenty-eighth aspect of the present invention includes various learning models each through a learning algorithm for outputting output information (s) related to controlling a component of a human-powered vehicle based on input information (s) driving the human powered vehicle; a detection part that detects the input information (s); a model selection part that accepts the selection of one of the different learning models; and a control part that controls the component based on output information (s) output from a selected one of the various learning models by input information (s) detected by the detection part in that a selected one of the learning models is / are entered.

Thus, component control depending on the situations in which the determination using a threshold is insufficient based on multiple input information including the measured values related to driving can be achieved to be suitable for different situations or driving environments.

(29) In a component control device according to the twenty-ninth aspect of the present invention, the detection part in the component control device according to the twenty-eighth aspect described above acquires input information that is matched with the selected one of the learning models selected by the model selection part / and enters the input information (s).

Thus, automatic control can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(30) In a component control device according to the thirtieth aspect of the present invention, the output information (s) in the component control device according to the twenty-ninth aspect described above includes at least one of a gear stage and a gear ratio for a transmission.

In this way, automatic control of the gear shift can be achieved in which the driver is less likely to feel uncomfortable to be suitable for different situations or driving environments.

(31) A generation method according to the thirty-first aspect of the present invention includes acquiring input information (s) related to driving a human powered vehicle; and generating, by a learning algorithm, various learning models, each of which generates output information (s) relating to the control of a component of the human-powered vehicle based on input information (s) acquired.

Thus, component control depending on the situations in which the determination using a threshold is insufficient based on multiple input information including the measured values related to driving can be achieved to be suitable for different situations or driving environments.

(32) A component control method according to the thirty-second aspect of the present invention includes acquiring input information (s) related to driving a human powered vehicle; Accepting the selection of any one of various learning models, each generated by a learning algorithm, such that output information (s) pertain to control of a component of the human powered vehicle based on the input information (s) is / are output; Specifying output information (s) to be output from a selected one of the learning models by inputting the acquired input information (s) into the selected one of the learning models; and controlling the component based on the specified output information (s).

Thus, component control depending on the situations in which the determination using a threshold is insufficient based on multiple input information including the measured values related to driving can be achieved to be suitable for different situations or driving environments.

(33) A computer program according to the thirty-third aspect of the present invention causes a computer to process processing of acquiring input information (s) related to driving a human-powered vehicle; and generating various learning models by a learning algorithm, each of which generates output information (s) regarding the control of a component of the human-powered vehicle based on input information (s) acquired.

Thus, component control depending on the situations in which the determination using a threshold is insufficient based on multiple input information including the measured values related to driving can be achieved to be suitable for different situations or driving environments.

According to a generation device for generating data related to the control of the components of a human powered vehicle, a component control device, a generation method, a component control method, and a computer program, the automatic control of the components that are less likely to be caused by the driver even in the various situations or driving environment feels uncomfortable.

list of figures

• 1 10 is a side view of a bicycle to which generation devices or component control devices according to embodiments 1 to 4 are applied, respectively.
• 2 Fig. 10 is a block diagram showing the internal structure of a control unit.
• 3 Fig. 12 is a flowchart showing an example of a processing procedure in a learning mode performed by a control part.
• 4 represents an example of a learning model to be created.
• 5 represents an example of a learning model to be created.
• 6 Fig. 12 is a flowchart showing an example of a processing procedure in an automatic control mode performed by the control part.
• 7 12 is a block diagram illustrating the internal structure of a control unit according to Embodiment 2.
• 8th 11 is a flowchart showing an example of a processing procedure in a learning mode performed by a control part according to Embodiment 2.
• 9 is an example of the content of input information (s) stored in a storage part and an overview of the training.
• 10 FIG. 10 is an example of a learning model to be created in Embodiment 2.
• 11 FIG. 10 is an example of a learning model to be created in Embodiment 2.
• 12 FIG. 12 is a flowchart showing an example of a processing procedure performed by a control section according to embodiment 3.
• 13 11 is a flowchart showing an example of a processing procedure performed by the control section according to Embodiment 3.
• 14 Fig. 12 is a block diagram illustrating the configuration of a control system.
• 15 is an example of a screen displayed based on an application program.
• 16 is an example of a screen displayed based on an application program.
• 17 is an example of a screen displayed based on an application program.
• 18 is an example of a screen displayed based on an application program.
• 19 is an example of a screen displayed based on an application program.
• 20 FIG. 12 is a flowchart showing an example of a processing procedure performed by a control part according to Embodiment 4.
• 21 FIG. 12 is a flowchart showing an example of the processing procedure performed by the control section according to Embodiment 4.

DETAILED DESCRIPTION OF THE EMBODIMENTS

The following descriptions of the embodiments are examples of shapes that a generating device and a component control device according to the present invention can take, although it is not intended to limit the shapes. The generation device, the component control device, a generation method, a component control method, a computer program, and a learning model according to the present invention may take forms that are different from the embodiments, for example, forms of modification of each of the embodiments and a combination of at least two modifications that are not disagree. The term "at least one of" as used in this disclosure means "one or more" of a desired choice. For example, the term "at least one of" used in this disclosure means "only one choice" or "both of two choices" when the number of its choices is two. On the other hand, the term “at least one of” as used in this disclosure means “only a single choice” or “any combination of at least two choices” if the number of their choices is three or more.

In the following description of each of the embodiments, the terms indicating directions such as e.g. B. front, rear, forward, backward, left, right, side, top, bottom and so on, with reference to the directions used when the user is seated in the saddle of a human powered vehicle.

(Embodiment 1)

1 is a side view of a human powered vehicle A to which a generating device 1 according to Embodiment 1 is applied. The human powered vehicle A is a road bike that has a support mechanism C contains the drive of the human powered vehicle A supported using electrical energy. The structure of the human-powered vehicle A can be changed as desired. In the first example, the human powered vehicle contains A not the support mechanism C , In the second example is the type of human powered vehicle A a commercial bike, a mountain bike or a hybrid bike. In the third example, the human powered vehicle contains A the features of the first example and the second example.

The human powered vehicle A is with a main body A1 , a handlebar A2 , a front wheel A3 , a rear wheel A4 , a front fork A5 , a saddle A6 and a derailleur hanger A7 Mistake. The human powered vehicle A contains a drive mechanism B , a support mechanism C , an operating device D , a gear e , an electric seat post F , an electric suspension G , a battery unit H and a control unit 100 , The human powered vehicle A contains a speed sensor S1 , a cadence sensor S2 , a torque sensor S3 , a gyro sensor S4 , an image sensor S5 and position sensors S61 . S62 and S63 , The main body A1 is with a frame A12 Mistake.

The drive mechanism B transmits a human-powered drive force to the rear wheel through chain drive, belt drive or shaft drive A4 ,

1 illustrates the drive mechanism B of the chain drive.

The drive mechanism B contains a crank B1 , a first sprocket assembly B2 , a second sprocket assembly B3 , a chain B4 and a pair of pedals B5 ,

The crank B1 contains a crankshaft B11 , a right crank B12 and a left crank B13 , The crankshaft B11 is rotatable on the on the frame A12 mounted support mechanism C supported. The right crank B12 and the left crank B13 are each with the crankshaft B11 coupled. One of the pair of pedals B5 is rotatable on the right crank B12 supported. The other of the pair of pedals B5 is rotatable on the left crank B13 supported.

The first sprocket assembly B2 has a first center axis of rotation and is with the crankshaft B11 coupled to enable a uniform rotation. The first sprocket assembly B2 contains one or more sprockets B22 , The crankshaft B11 and the first sprocket assembly B2 are coaxial to each other. As an example, the first sprocket assembly includes B2 several sprockets B22 with different outside diameters. As an example are the multiple sprockets B22 Larger in outside diameter and larger in number of gear stages when taken from the main body A1 are removed to the outside.

The second sprocket assembly B3 has a second center axis of rotation and is rotatable on a (not shown) hub of the rear wheel A4 supported. The second sprocket assembly B3 contains one or more sprockets B31 , As an example, the second sprocket assembly includes B3 several sprockets B31 with different outside diameters. As an example are the multiple sprockets B31 made smaller in outer diameter and larger in number of gears, since they are from the rear wheel A4 are removed to the outside.

The chain B4 is around one of the sprockets B22 the first sprocket assembly B2 and one of the sprockets B31 the second sprocket assembly B3 carried. If the crank B1 by a human-powered driving force on the pair of pedals B5 is exercised, rotates forward, the first sprocket assembly rotates B2 together with the crank B1 forward which is the rotation of the first sprocket assembly B2 on the second sprocket assembly B3 over the chain B4 transmits to thereby the rear wheel A4 to turn forward.

The support mechanism C contains an electric actuator C1 ,

The support mechanism C supports the propulsion of the human-powered vehicle A , As an example, the support mechanism transfers C torque to the first sprocket assembly B2 to thereby propel the human-powered vehicle A to support. The electric actuator C1 contains, for example, an electric motor. The electric actuator C1 may include a reduction gear. The electric actuator C1 contains the chain B4 that are used to transmit a driving force to the rear wheel A4 of the human powered vehicle A running. The support mechanism C is part of the components that are signaled to support the movement of the chain B4 can be controlled.

The control device D contains a control unit D1 to be operated by the user. An example of the control panel D1 are one or more buttons. Another example of the control panel D1 is a brake lever. Each time the brake levers provided on the left and right handles are tilted laterally, a gear stage or a gear ratio for the rear derailleur or transmission can be used e (English transmission) can be changed. The operator also accepts D a determination operation related to the control of various components, for example the mode switching of the support mechanism C (Energy saving mode, high performance mode, etc.), switching the operation of the electric seat post F , switching the operation of the electric suspension G and so on. The control device D is connected to communicate with each of the components to provide a signal corresponding to one on the control panel D1 performed operation on the transmission e or to transfer the other components. In the first example, the control device D connected to the gearbox e to communicate via a communication line or an electrical line that enables power line communication (PLC). In the second example is the operating device D connected to the gearbox e and communicate the other components through a wireless communication unit that enables wireless communication. In the case where the control unit (s) D1 is actuated, a control signal for switching the gear stage of the transmission e to the transmission e transmitted, and in response to the signal, the transmission shifts e the gear step in the first example. The control signal contains an INWARD signal, for example a shift to the inner sprocket B31 indicates, and an OUT signal, for example a shift to the outer sprocket B31 displays. Each of the signals can indicate the gear stage of the sprocket to be switched B31 contain. It may also be possible to switch two stages or more at the same time.

The gear e can take various forms. In the first example is the gearbox e an external transmission for switching a coupled state between the second sprocket assembly B3 and the chain B4 , In particular, the transmission changes e by switching the one with the chain B4 sprocket to be coupled B31 a ratio of the speed of the rear wheel A4 to the speed of the crank B1 , that is Gear ratio of the human powered vehicle A ,

The gear e activates the electrical actuator E1 to the chain B4 to move according to the gear ratio selected, thereby changing the gear ratio. In particular, the transmission e in the first example on the rear derailleur hanger A7 of the human powered vehicle A appropriate. In the second example is the transmission e an external transmission for shifting a coupled state between the first sprocket assembly B2 and the chain B4 , In particular, the transmission changes e by switching the one with the chain B4 sprocket to be coupled B22 a ratio of the speed of the rear wheel A4 to the speed of the crank B1 , that is, the gear ratio of the human powered vehicle A , In the third example is the transmission e an internal gearbox or a shift hub. In the third example, the moving part contains the transmission e at least one of a sleeve and a claw of the inner gear. In the fourth example is the transmission e a continuously variable transmission. In the fourth example, the moving part includes the gearbox e a spherical planet of the continuously variable transmission. The gear e is part of the components that can be controlled by a signal for switching a gear stage.

The electric seat post F is on the frame A12 appropriate.

The electric seat post F contains an electric actuator F1 , The electric actuator F1 causes the saddle A6 relative to the frame A12 increases or decreases from it. The electric actuator F1 is for example an electric motor. The electric seat post F is part of the components that can be controlled by a supported position of the saddle A6 relative to the frame A12 is set as an operating parameter. The saddle A6 can be supported at one or more support positions.

The electric suspension G can take various forms. In embodiment 1 is the electric suspension G that on the front fork A5 is provided a front suspension for damping the on the front wheel A3 applied shock. The electric suspension G contains an electric actuator G1 , The electric actuator G1 is for example an electric motor. The electric suspension G is part of the components that can be controlled by setting a damping factor, a stroke amount and a lock state as operating parameters. The electric suspension G can adjust the operating parameters by driving the electric actuator G1 to change. The electric suspension G can be a rear suspension around which on the rear wheel A4 dampen the impact.

The battery unit H contains a battery H1 and a battery holder H2 , The battery H1 is a rechargeable battery that contains one or more battery cells. The battery holder H2 is on the frame A12 of the human powered vehicle A attached. The battery H1 can on the battery holder H2 attached and detached from this. When on the battery holder H2 is attached is the battery H1 electrically with at least the electrical actuator E1 of the transmission e , the electric actuator C1 of the support mechanism C and the control unit 100 connected. The battery H1 can also be electrical with the electrical actuator F1 the electric seat post F and the electric actuator G1 the electric suspension G be connected.

The speed sensor S1 is on the frame A12 attached. The speed sensor S1 is a sensor for outputting a signal indicating the driving speed of the human powered vehicle A displays. The speed sensor S1 one on the front wheel A3 provided magnets and one on the front fork A5 contains provided main body for detecting the magnet, for example, measures a rotational speed.

The cadence sensor S2 is intended to be a cadence of each of the right crank B12 and the left crank B13 to eat. The cadence sensor S2 outputs a signal that shows the measured cadence. The torque sensor S3 is provided to measure torques that are applied to the right crank B12 and the left crank B13 be created. The torque sensor S3 outputs a signal that the measured torque is at least one from the right crank B12 and the left crank B13 displays.

The gyro sensor S4 is on the frame A12 attached. The gyro sensor S4 is a sensor for outputting signals, the yaw, roll and pitch of the human powered vehicle A Show. The gyro sensor S4 can output a signal indicating at least one of the three axes, but is not limited to all three axes.

The image sensor S5 is on the frame A12 provided so that it faces the front. In the first example is the image sensor S5 along with a light like that on the front fork A5 attached that it faces forward. In the second example he is on the handlebar A2 assembled.

The image sensor S5 outputs a video that corresponds to the user's field of vision using a camera module. The image sensor S5 outputs a video signal obtained by photographing an object in the direction of travel. The image sensor S5 may be a module that is integrally equipped with an image recognition unit for performing recognition processing for separately recognizing a road, a building, and another moving vehicle from the video and that outputs a recognition result.

The position sensors S61 . S62 and S63 are, for example, piezoelectric sensors. The position sensors S61 . S62 and S63 are on each of the sections of the human powered vehicle A provided to which the weight of the user is applied. The position sensors S61 are provided on both handles, for example. The position sensors S62 are in one or more positions of the saddle A6 provided along its surface. The position sensors S63 are on both pedals B5 the crank B 1 provided. The position sensors S61 . S62 and S63 each give a signal that responds to the weight applied. Instead of the position sensors S61 . S62 and S63 or in addition, a gyro sensor can be attached to the helmet to sense the user's position.

2 is a block diagram showing the internal structure of the control unit 100 represents. The control unit 100 contains a control part 10 , a storage part 12 and an input-output part 14 ,

The control unit 100 is at any position on the frame A12 intended.

In the first example, as in 1 is shown is the control unit 100 between the first sprocket assembly B2 and the frame A12 intended.

In the second example, the control unit 100 in the battery holder H2 intended.

The control part 10 is a processor that uses a central processing unit (CPU) or a graphics processing unit (GPU) and performs the processing by controlling a learning algorithm described later and that in the human-powered vehicle A provided components using a memory, such as a built-in read-only memory (ROM), a random access memory (RAM) and so on.

The control part 10 captures time information (s) at any time using an internal clock.

The storage part 12 contains non-volatile memory, such as flash memory. The storage part 12 saves a tutorial 1P and a control program 2P , The tutorial 1P can in the control program 2P be included. The tutorial 1P can be obtained by using on a recording medium 18 saved tutorial 8P read out and in the / the memory part 12 is copied. The control program 2P can be obtained by using on a recording medium 19 saved memory program 9P read out and in the / the memory part 12 is copied.

The storage part 12 stores several different learning models 1ma . 1Mb , ... by the from the control section 10 performed processing are generated. The storage part 12 saves correspondence between the different learning models 1ma . 1Mb , ... and various modes that will be described later. The storage part 12 saves for each of the different learning models 1ma . 1Mb , ... or for each of the modes (a) setting information (s) that contains / contain settings with which (a) input information (s) from the human powered vehicle A attached sensor groups S1 - S5 and S61 - S63 to be combined and set as (one) input information (s) and which information (s) is / are to be weighted. The setting contained in the setting information (s) can be used for the different learning models 1ma . 1Mb , ... or be the same or different for the different modes. The control part 10 accepted for each of the different learning models 1ma . 1Mb , ... setting a different combination of input information (s) in setting information (s) and storing the setting information (s) based on the accepted setting. The control part 10 similarly accepted for each of the different learning models 1ma . 1Mb , ... settings of weights to be assigned to multiple data items included in each item of input information (s) in the setting information (s), and stores the setting information (s) based on the accepted setting. The setting can be accepted in advance or at any time.

The input-output part 14 is with at least the sensor groups S1 -S5 and S61-S63, the operating device D and the electrical actuator to be controlled E1 connected to the human powered vehicle A are mounted. The input-output part 14 is also with the electrical actuators to be controlled F1 and G1 connected. The control part 10 receives via the input-output part 14 an input of a signal indicating a speed or cadence from one of the speed sensor S1 or the cadence sensor S2 displays. The control part 10 receives an input of a signal representing a position of the human powered vehicle A indicates, in particular a signal that yaw, roll or pitch from the gyro sensor S4 displays. The control part 10 receives input from the position sensors of a signal indicating a position of the user, in particular a signal indicating a weight distribution S61 . S62 and S63 , The control part 10 performs processing using the information (s) from these sensor groups S1 - S5 and S61 - S63 is received as input information (s). The control part 10 receives via the input / output part 14 a signal from the control device D , In the in 2 The block diagram shown is the input / output part 14 with the electric actuator F1 and the electric actuator G1 connected even though it is not connected to the electrical actuator F1 and the electric actuator G1 must be connected if these actuators F1 and GI should not be controlled.

The control unit 100 works in a learning mode and an automatic control mode. The control unit 100 works in other different modes, both in learning mode and in automatic control mode. In the first example, the different modes include a "racing mode", a "city mode" and a "mountain mode". In the second example, the different modes include a "power mode", a "normal mode" and an "eco mode". In the third example, the different modes are merely classified into a “first mode” and a “second mode” without purpose. The control device D is with a mode selection button D2 to accept a mode selection for each type of human powered vehicle A Mistake. The control device D stores information indicating which mode is selected by the mode selection key D2 is selected in the integrated memory. The control part 10 can refer to the information (s) relating to the selected mode of the mode selection button D2 through the control device D displays from the memory. The mode selection button D2 corresponds to a “selection part” of a “generation device” or a “model selection part” of a “component control device”.

When working in the learning mode, the control part (s) generates 10 different learning models 1ma . 1Mb , ... that correspond to different modes based on the tutorial 1P through a learning algorithm of in-depth learning. The control unit corresponds in the learning mode 100 the "generating device". When operating in the automatic control mode, the control part (s) functions 10 as a model selection part, which is the selection of one of the different learning models 1ma . 1Mb accepted by one of the different modes based on the control program 2P is selected. The control part 10 acts as a control part, the components including the transmission e based on output information (s) controlled by a selected one of the learning models 1ma . 1Mb , ... is / are output by that of the input-output part 14 recorded input information into the selected one of the learning models 1ma . 1Mb , .... is / are entered. In the automatic control mode, the control unit corresponds 100 the "component control device".

First, the operation of the control unit 100 described in learning mode. 3 Fig. 11 is a flowchart showing an example of a processing procedure in the learning mode performed by the control part 10 is carried out. The learning algorithm in the present embodiment is supervised deep learning using a neural network (hereinafter referred to as NN). The learning algorithm can be an unguarded learning algorithm or a recurring neural network. The learning algorithm can serve for reinforcing learning. The control part 10 executes the processing procedure in the in 3 shown flowchart repeatedly during the learning mode for each predetermined sampling period (for example 30 milliseconds) for each component to be controlled.

The control part 10 determines the mode used by the control device D is selected from the different modes (step S101 ). In step S101 can the control part 10 determine which mode via the control device D as described above. The control part 10 chooses a target model of the learning models 1ma . 1Mb , ... according to the specific mode from the different learning models 1ma . 1Mb , ... off (step S102 ).

The control part 10 captures input information related to the driving of the human powered vehicle A through the input / output part 14 (Step S103 ). In step S103 corresponds to the control part 10 a "registration part". In step S103 relates to the control part 10 specifically to each of the signal levels from the sensor groups S1 - S5 and S61 - S63 by the input / output part 14 input every sampling period depending on the sensors, and temporarily stores the signal level in the internal memory of the control section 10 or in one in the input-output part 14 integrated memory. The sampling period for each sensor may be the same as or different from the predetermined sampling period.

In step S103 the control part 10 do not enter all the information (s) from the input-output part 14 can be detected as described later. The control part 10 can only record a driving speed and a cadence. The control part 10 can perform preprocessing, such as discrimination, extraction, calculation, etc., on the input information (s) that can be captured.

The control part 10 refers to actual control data of an object to be controlled in connection with the one or more in step S103 input information (s) recorded (step S105 ). The control data represents details of the operation performed by the control panel D1 is accepted. The control part 10 refers to step for example S105 on a Gear step or a gear ratio for the transmission e , In this case, the control part (s) 10 the gear stage based on a signal from the operating device D can recognize or recognize the current gear stage with reference to that from the transmission e Recognize feedback (s). The control data in connection with the entered information (s) on which in step S105 Referring to, a state may be a predetermined time (e.g., 10 milliseconds, 50 milliseconds, 100 milliseconds, 1 second) after the input information (s) has been entered.

The control part 10 reads the setting that the in step S101 corresponds to certain mode from those in the memory section 12 stored setting information (s) from (step S107 ) and selects the input information (s) to be input from the input information (s) acquired based on the setting (step S109 ). By processing in step S109 be for one of the learning models 1ma . 1Mb , ... that correspond to the particular mode from which in step S103 captured input information (s) Parts of input information (s) that are in combination with the other learning models 1ma . 1Mb distinguish, ... that correspond to the other modes, selected and entered.

The control part 10 assigns the selected input information (s) to the weights contained in the setting (step S111 ). By processing in step S111 become one of the learning models 1ma . 1Mb , ... that correspond to the specific mode, several data elements that are in the / in step S103 recorded input information (s) are included, assigned weights that differ from those that the other learning models 1ma . 1Mb , ... that correspond to the other modes.

The control part 10 identifies the input information, each in step S109 and step S111 extracted and weighted with the tax data on which in step S105 Reference was made to thereby generate training data (step S113 ). The control part 10 outputs the marked input information (s) to the NN according to the specific mode 13 a step S115 ). The control part 10 carries out training processing for training parameters in the intermediate layer of the NN 13 through (step S117 ). The control part 10 ends the training based on the entered information (s) for a sampling time after the execution of the training processing.

The procedure for creating the different learning models 1ma . 1Mb , ... is not limited to the supervised learning method in which information (s) are tagged with control data related to the actual control as described above. The learning models 1ma . 1Mb , ... can be based on the NN 13 are generated from which output information (s) relating to the control are / are output as numerical values. The control part 10 can do any of the learning models 1ma . 1Mb , ... using parameters in the intermediate layer 133 train using an error propagation method by calculating an error based on actual control data. Using the NN 13 who, for example, the gear stage for the gear e outputs as a numerical value, an error is calculated between the output gear stage and the actual gear stage and at least one of the weight and the preload in the intermediate layer 133 can be trained while the calculated error is spreading. Furthermore, the control part (s) 10 using the NN 13 , which was generated beforehand in such a way that input information (s) marked with a gear stage or a gear ratio is accumulated and specified as the initial state, carry out a training in such a way that the error between that of the NN 13 output gear stage or gear ratio and the actual gear stage or gear ratio is minimized by the error propagation method. As will be described later, the control part (s) can 10 perform a training by reinforcing learning that advances the training by providing an evaluation based on a comparison between the output information (s) regarding the output control and the actual control data. The learning algorithms can be used between different learning models 1ma . 1Mb ,... be different.

By repeating the steps in the flowchart of 3 Training processing shown for each of the selected modes becomes the NN 13 in each of the learning models 1ma . 1Mb , ... changed the output information (s) regarding the control of the components of the human powered vehicle A based on the input information (s) related to the human powered vehicle A output. The NN 13 is used as each of the learning models 1ma . 1Mb , ... generated according to the selected mode. 4 and 5 illustrate examples of the learning models to be generated 1ma . 1Mb , As in 4 and 5 shown, contain the learning models 1ma . 1Mb , .... one input layer each 131 , in which several of the sensor groups S1 - S5 and S61 - S63 captured input information can be entered, and an output layer 132 , from which tax data are output. Each of the learning models 1ma . 1Mb , ... contains an intermediate layer 133 between the input layer 131 and the output layer 132 is arranged and contains a group of nodes formed by one or more layers. The one with the output layer 132 bonded intermediate layer 133 is a connected layer for aggregating the multiple nodes to the number of nodes in the output layer 132 , In the examples in 4 and 5 Without limitation, there are three nodes in the output layer 132 shown. The nodes of the intermediate layer 133 each have a parameter that contains at least one of the weighting and bias in connection with the node in the previous layer. The control part 10 generates training data by providing input information (s) with actual control data associated with the input information (s) based on the tutorial 1P features. The control part 10 puts the generated training data for each selected mode in the input layer 131 one so the parameters in the intermediate layer 133 be trained.

With reference to 4 and 5 become the learning models 1ma . 1Mb , ... specifically described. In the concrete example, each of the learning models 1ma . 1Mb , ... described, from which (an) output information (s) relating to the control of a gear stage or a gear ratio for the transmission e will be issued. As in 4 and 5 is shown by the speed sensor S1 received driving speed in the input layer 131 entered. The driving speed is, for example, the speed per hour. The driving speed can be the number of revolutions of the front wheel A3 or the rear wheel A4 per unit of time. One from the cadence sensor S2 Cadence obtained is in the input layer 131 entered. At least one of the driving speed of the human powered vehicle A and the cadence of the crank B1 of the drive mechanism B is in the input layer 131 entered. As in the learning model 1Mb of 5 shown, a power obtained by calculation using the torque and the cadence can be input to the input layer 131 can be entered. A torque from the torque sensor S3 can be obtained in the input layer 131 can be entered. In the in 4 shown learning model 1ma the cadence and torque are entered while they are given much larger weights than the other input information (s) based on the setting of the corresponding mode. In the in 5 shown learning model 1Mb the vehicle speed is input while a much larger weight than the other input information (s) is assigned based on the setting of the corresponding mode.

Detection data of the attitude of the human powered vehicle A by the gyro sensor S4 are obtained are in the input layer 131 entered.

The detection data is information that the inclination of the human powered vehicle A specify. The pitch is determined by a yaw component having an axis in the vertical direction and a roll component having an axis in the front-back direction of the human powered vehicle A and a pitch component that has an axis in the right-left direction of the human-powered vehicle A has shown. In the in 5 shown learning model 1Mb the roll and yaw tendencies are entered while much larger weights than the other input information (s) are assigned based on the setting of the appropriate mode.

One from the image sensor S5 received video signal is in the input layer 131 entered. That from the image sensor S5 video signal obtained is a video signal corresponding to the user's field of view, that is, data obtained by detecting a driving environment. In the first example, the input video signal is each of the successive multi-frame pictures. In the second example, the input video signal consists of several pieces of data obtained by performing various filter processing on a frame picture. In the third example, the input video signal is information (s) indicating the kind of an object present in the traveling direction, which is from the image recognition part based on the video from the image sensor S5 is recognized. The distance from the object in the direction of travel can be included. The distance may be a distance obtained according to the image processing by the image recognition part (s) or by a human-powered vehicle A mounted radar.

In the in 4 shown learning model 1ma and the in 5 shown learning model 1Mb a video signal is not selected as input information (s) or input while being weighted much less than the other information (s).

Another example of the detection data for a driving environment is time data, weather data, luminance intensity data or moisture data. The time data is, for example, a time that is set by an internal timer of the control part 10 is measured. The weather data are, for example, at least one of localized amounts of precipitation, air humidity, wind speed and wind direction at the location where the vehicle driven by human power A drives, which can be obtained from an external weather data processing server. The moisture data can be obtained from a moisture sensor attached to the main body A1 of the human powered vehicle A is appropriate. The luminance intensity data is obtained by providing a luminance intensity sensor at an arbitrary position of the main body A1 of the human powered vehicle A , for example on the handlebars A2 , receive.

User position detection data from each of the position sensors S61 . S62 and S63 are obtained are in the input layer 131 entered. The detection data are, for example, weight distribution data. The position sensors S61 . S62 and S63 in the present embodiment are piezoelectric sensors. In the first example, signals are from the position sensors S61 . S62 and S63 are output in the input layer 131 entered. In the second example, the control part (s) determines 10 whether the driver is in a basic posture, a forward inclined position or on pedals, and gives the result of the distinction in the input layer 131 on. In the in 4 shown learning model 1ma the weight distribution indicating the user's position is not selected as input information (s) or is input while a much smaller weight than the other information (s) is assigned.

In the input layer 131 Information (s) to be entered can be found under the different learning models 1ma . 1Mb , ..., as described above, may be different. All input information to be selected does not have to be in the input layer 131 can be entered. As by dotted lines in 4 specified, the input information (s) to be selected can be divided into groups, and the input information (s) can be set for a different NN for each group 13 can be entered. In this case, the output layer 132 Output control information for each group.

The output layer 132 provides information that includes at least one of a gear stage and a gear ratio for the transmission e is out. In particular, there is the output layer 132 Ratios that correspond to the nodes of the gear stage and the gear ratio.

This enables the control section 10 to select the gear level with the highest probability.

As in 4 and 5 shown, the different learning models differ 1ma . 1Mb , ... in the details of the input information (s) or in the distribution of the weights of the input information (s). The learning models 1ma . 1Mb , ... with the same input information (s) and their weight distribution can generate different output information. Through such training processing, the learning models 1ma . 1Mb , ..., which differ from one mode to another, which output information which is / are adapted to a driving style of the user depending on the various input information.

Second, the operation of the control unit 100 described in automatic control mode. 6 Fig. 14 is a flowchart showing an example of a processing procedure in the automatic control mode performed by the control part 10 is carried out. The control part 10 does that in the flowchart in 6 shown processing procedure repeatedly during the automatic control mode. The processing procedure can be carried out for any predetermined control period (e.g. 30 milliseconds).

The control part 10 determines a mode by the operating device D is selected from the different modes (step S201 ). The control part 10 accepted in step S201 the selection of one of the different learning models 1ma . 1Mb , ... through the mode selection via the control unit (s) D1 the operating device D , The control part 10 selects one of the generated learning models 1ma . 1Mb , ... according to the specific mode and reads it from the memory section 12 off (step S203 ).

The control part 10 captures input information related to the driving of the human powered vehicle A through the input / output part 14 (Step S205 ). The control part 10 corresponds to a “detection part” of the “component control device” in step S205 , In step S205 relates to the control part 10 on the signal levels from the sensor groups S1 - S5 and S61 - S63 over the input / output part 14 entered in each control period and temporarily stores them in the internal memory of the control section 10 or that in the input-output part 14 integrated memory.

The control part 10 reads out a setting that is in step S201 specific mode or that in step S203 selected learning model 1ma . 1Mb , ... corresponds to from which in the memory part 12 stored setting information (step) S207 ) out. The control part 10 captures (an) input information (s) associated with the selected one of the learning models 1ma . 1Mb , ... is matched from the input information (s) based on the setting in step S205 has been recorded (step S209 ). The control part 10 assigns the weighted input information (s) contained in the setting (step S211 ).

The control part 10 gives that in step S209 selected and in step S211 weighted input information (s) in the in step S203 selected learning model 1ma . 1Mb , ... a step S213 ).

The control part 10 specifies output information (s) regarding the control of the component to be controlled, which is to be output by the acquired input information (s) in the selected one of the learning models 1ma . 1Mb , ... is entered (step S215 ). The control part 10 specified for example in step S215 a result of a distinction in one of the gear stage and the gear ratio for the transmission e as output information (s).

The control part 10 refers to the state of the component to be controlled based on the specified output information (s) (step S217 ).

The control part 10 determines whether or not an output of a control signal is required based on the relationship between the details of the control indicated by the specified output information (s) and the state referred to (step S219 ).

If in step S219 it is determined that an output of a control signal is required ( S219 : YES), gives the control part 10 a control signal for controlling a component based on the specified output information (s) to the component to be controlled (step S221 ) and ends the tax processing performed in one tax period. If in step S219 it is determined that an output of a control signal is not required ( S219 : NO), ends the control part 10 the processing without outputting a control signal based on the specified output information (s) related to the control to the object to be controlled. The reference processing in step S217 and the determination processing in step S219 are not necessarily carried out. In this case, the control part (s) 10 the result of a distinction between a gear stage and a gear ratio for the transmission e for example, as output information (s) in step S215 and can output a control signal to the component to be controlled based on the specified output information (s) related to the controller without referring to the state of the component to be controlled based on the specified output information (s) ( S221 ).

The control part 10 relates, for example, to a gear stage or a gear ratio for the transmission e in step S217 , In this case, the control part (s) compare 10 in step S219 the in step S215 specified gear or the gear ratio with that in step S217 specified gear ratio or gear ratio and determines whether there is a difference or not. If it is determined that there is a difference, the control part (s) determines 10 in step S219 that an output of a control signal is necessary.

The learning mode in Embodiment 1 can be executed from the time when the control unit 100 to an initial state in the human powered vehicle A until the point at which the error between that of each of the learning models 1ma . 1Mb , ... outputted output information (s) and the control data related to the control are reduced to a predetermined error or less based on the actual operation. After the reduction, the control part (s) can 10 the control unit 100 be operated in automatic control mode. The learning mode can also be executed until the error similarly decreases after a reset operation.

In embodiment 1, the control unit accepts 100 the mode selection by the mode selection button D2 the operating device D to help you choose one of the different learning models 1ma . 1Mb , ... to accept. The control unit 100 can autonomously select a mode based on the driving environment without accepting the mode selection by the operation. The control unit 100 can choose another mode based on the predetermined schedule.

(embodiment 2 )

7 Fig. 3 is a block diagram showing the internal structure of a control unit 100 according to embodiment 2. The human powered vehicle A and the control unit 100 in Embodiment 2 are similar in structure to those in Embodiment 1 except for a memory part and the details of processing to be described later, and the common components are denoted by the same reference numerals and their detailed Description is not repeated. In embodiment 2, the storage part (s) stores 12 the control unit 100 Input information (s) provided by the input-output part 14 from the sensor groups S1 - S5 and S61 - S63 be entered on the human powered vehicle A are assembled, together with time information (s) over several sampling periods. The storage part 12 stores the input information (s) cyclically using memory areas of the multiple sampling periods, so that the old input information (s) is / are automatically overwritten.

8th Fig. 12 is a flowchart showing an example of a processing procedure in a learning mode by a control part 10 is carried out according to embodiment 2. The control part 10 executes the processing procedure of the in 8th shown flowchart repeats every predetermined sampling period (e.g. 30 milliseconds) during the learning mode for each component to be controlled.

The control part 10 determines a mode by the operating device D is selected from different modes (step S121 ). The step S121 corresponds to a "selection part". In step S121 can the control part 10 , as described above, determine which mode is via the control device D is selected. The control part 10 chooses a target model of the learning models 1ma . 1Mb , ... according to the specific mode from the different learning models 1ma . 1Mb , ... off (step S122 ).

The control part 10 captures input information related to the driving of the human powered vehicle A through the input / output part 14 (Step S123 ) and refers to actual control data of a component that is to be controlled at essentially the same time (step S125 )). In step S125 relates to the control part 10 for example to a gear or a gear ratio for the transmission e , The control part 10 specifies the time at which the input information (s) in step S123 is detected (step S127 ). The control part 10 can use time information (s) captured by the internal timer as the time of acquisition or can convert the time information (s) into time information (s) and use the time information (s).

The control part 10 reads out the setting in step S121 corresponds to certain mode from which in the memory part 12 stored setting information (s) from (step S129 ) and selects information (s) to be input from the input information (s) acquired based on the setting (step S131 ). The control part 10 assigns the selected input information (s) to the weights contained in the setting (step S133 ).

The control part 10 stores the time information (s) that the in step S127 specified time indicates the input information (s) in step S131 respectively. S133 extracted and weighted (s), and the actual tax data referenced in step S125 Reference was made in the memory section 12 while they are connected (step S135 ).

The control part 10 captures a group of input information (s) captured at various times and in the storage part 12 have been saved (step S136 ). The control part 10 identifies the captured group of input information (s) with the control data referenced in step S125 Reference was made to generate training data (step S137 ). The control part 10 gives the marked group of input information recorded at different times in the NN 13 one that matches the selected one of the learning models 1ma . 1Mb , ... corresponds to (step S139 ). The control part 10 carries out training processing for training parameters in the intermediate layer of the NN 13 off (step S141 ) and ends the learning processing at a sampling time.

9 illustrates an example of the contents of those in the storage part 12 stored input information (s) and the outline of the training. As in 9 Shown in embodiment 2 is the training processing for each of the learning models 1ma . 1Mb , ... using the input information (s) acquired at various times and the control data which are a result of the change of the input information (s) contained in the storage part 12 is saved as training data. The actual component control is based on the change in the state of the human powered vehicle A depending on its context, and is not performed independently for each sampling period. It is assumed that each of the learning models 1ma . 1Mb , ..., which output control data based on the information (s) entered at several points in time, is generated for each mode and used in practice to simulate the control of the components.

In embodiment 2, the control part (s) 10 multiple input information to one of the learning models 1ma . 1Mb , ... that correspond to the selected mode and controls the components based on the one of the learning models 1ma . 1Mb , ... tax data output also in Automatic control mode. The sampling period of the input information (s), ie the interval in which the input information (s) is / are acquired, can be carried out among the various learning models 1ma . 1Mb , ... can be set differently. For every learning model 1ma . 1Mb , ... a different sampling period can be set. A suitable sampling period for each of the learning models 1ma . 1Mb , ... can be evaluated using the learning algorithm.

In embodiment 2, instead of inputting the multiple input information in step S139 Variations can be entered in the multiple input information. The control part 10 can acquire input information at different times before and after each sampling period, calculate the variation of the input information and the calculated variation of the input information into a selected one of the learning models 1ma . 1Mb , ... enter. The control part 10 There may be a trend of variation (increase, decrease, maintain) in each of the learning models 1ma . 1Mb , ... or can calculate a waveform according to a time series distribution as a variation and the calculation result in the selected learning model 1ma . 1Mb ,.... enter. The variation value can contain one or more time difference values. 10 and 11 are examples of learning models 1ma and 1Mb to be generated in Embodiment 2. Compared to the in 4 and 5 of embodiment 1 are examples 10 and 11 different in that each input is a waveform that represents its variation. The learning model 1ma in 10 and the learning model 1Mb in 11 differ from each other in the details of the selection of the input information (s), the weights and the range of the variations. Therefore, the amount of time that corresponds to the variation of the entered information (s) can vary among the different learning models 1ma . 1Mb , ... be different.

(Embodiment 3)

In embodiment 3, the control unit switches 100 based on one on the control panel D1 the operating device D performed operation between the learning mode and the automatic control mode.

The control part 10 updates a learning model in learning mode 1M based on the operation performed on the keypad D1 ,

12 and 13 are each a flowchart showing an example of a processing procedure performed by the control part 10 is carried out in embodiment 3. The control part 10 repeatedly executes the following processing procedure performed in the automatic control mode. Some of the processing procedures described in the 12 and 13 shown flowcharts are shown, which with those in the 6 The flowchart shown in Embodiment 1 are denoted by the same step numbers, and their detailed description is not repeated.

The control part 10 selects input information (s) in connection with a selected one of the learning models 1ma . 1Mb , ... according to the selected mode, assigns weights to the selected input information (s) and enters the weighted input information (s) ( S213 ) and specifies output information (s) related to control from the selected one of the learning models 1ma . 1Mb , ... should be output (s) ( S215 ). The control part then determines 10 whether an operation from the control panel D1 is accepted or not (step S231 ). The control part 10 determines whether a shift between the gear stage and the gear ratio on the control panel D1 is carried out in embodiment 3 or not. In the case of controlling the components other than the transmission e can perform an operation related to the control of each of the components from the control panel D1 be accepted. If in step S231 it is determined that no surgery will be accepted ( S231 : NO), sets the control part 10 continue the operation in automatic control mode ( S217 - S221 ).

The control part 10 can be set to determine that in step S231 an operation is performed because the need for training arises when the gear shift is in the opposite direction to that in the crotch S215 Output information (s) output is performed and set so as not to determine that the operation will be performed if it is performed in the same direction.

If in step S231 it is determined that the operation from the control panel D1 is accepted ( S231 : YES), the control part begins 10 , the learning model 1M update in learning mode (step S233 ).

The control part 10 begins to measure an elapsed time when the operation in the learning mode with respect to the operation by the control panel (s) D1 is started as a trigger (step S151 ).

The control part 10 refers to the control data generated by the operation of the control panel D1 can be determined, ie a gear stage or a gear ratio in embodiment 3 in the learning mode (step S153 ). The control part 10 evaluates the output information (s) from the selected one of the learning models 1ma . 1Mb , ... is / are to be output based on the Tax data on which in step S153 Reference is made to being determined as the goal (step S155 ). In step S155 corresponds to the control part 10 an "assessment part".

In step S155 leads the control part 10 an assessment by giving the output information (s) from one of the learning models 1ma . 1Mb , ... that in step S215 specified selected mode, based on the control data checked based on the determination operation, in step S153 Reference was made. For example, the control part (s) determines 10 whether the one of the learning models 1ma . 1Mb , ... output information (s) output and the details of the determination operation match or not within a predetermined error range. If it is determined that a match occurs, the control part (s) delivers 10 a high rating. If the output information (s) is from a targeted one of the learning models 1ma . 1Mb , ... are output, and the details of the determination operation do not match, the control section delivers 10 a low rating of the output information (s) for the targeted learning models 1ma , 1Mb, .... If it is determined that no match occurs, the control part (s) delivers 10 a different rating depending on the degree of agreement between those in step S215 specified output information (s) and the control data referenced in step S153 Reference was made. If the difference between those in step S215 output information (s) specified and the information in step S153 specified control data is larger, delivers the control part 10 a lower rating. When a gear step is shifted by the control panel D1 in the opposite direction to that in step S215 output control data is carried out, supplies the / the control part 10 a much lower rating. For example, the rating can be represented by a numerical value and positive values are accepted for the high rating, while negative values are accepted for the low rating. The grade is accepted as larger if the absolute value is larger.

The control part 10 ranks based on the rating in step S155 weightings to the training data by identifying the in step S211 captured input information (s) are generated with the output information (s) by the selected one of the learning models 1ma . 1Mb , ... is / are output in order to be targeted (step S157 ). The control part 10 updates the selected one of the learning models 1ma . 1Mb , ... which is targeted as a target based on the training data to which the weights are assigned based on the evaluation (step S159 ). In step S 157, the control part / 10 Weights that indicate a higher level of importance for the training data when the rating is lower. Processing in steps S157 and S159 can be executed in the case where the output information (s) output from the selected one of the learning models targeted and the details of the determination operation in the evaluation in step S155 do not match each other.

The control part 10 determines whether the elapsed time, its measurement in step S151 started, is equal to or longer than a predetermined time or not (step S161 ).

When it is determined that the elapsed time is less than the predetermined time ( S161 : NO), shortens the control part 10 the sampling period, ie increases the frequency of acquisition of input information (step) S163 ). The control part 10 detects input information (s) in the changed sampling period and selects information (s) based on the setting corresponding to the selected mode (step S165 ). The control part 10 refers to actual tax data (step S167 ) and generates training data by marking the in step S165 input information (s) recorded with the control data to which reference is made (step S169 ). The control part 10 updates one of the learning models 1ma . 1Mb , ... which are selected by bringing them into accordance with the mode selected based on the training data of the marked input information (s) (step S171 ), and processing returns to step S161 back. In step S 169, the control part (s) can 10 Assign weights that indicate the level of importance in step S155 Specify the marked input information (s) to be entered as a function of the degree of assessment, and the weighted training data to the corresponding one of the learning models 1ma . 1Mb ,....to deliver.

If in step S161 it is determined that the elapsed time is equal to or greater than the predetermined time ( S161 : YES), the control part ends 10 the measurement of the elapsed time (step S173 ), ends the operation in the learning mode, switches it to the automatic control mode (step S175 ) and ends the processing.

In embodiment 3, the control unit 100 at times several input information in time series in the memory part 12 in connection with actual control data based on the determination operation similar to Embodiment 2. The control part 10 can select and save input information based on the setting corresponding to the selected mode. The control part 10 may perform the following processing instead of updating the control output information (s) with a low rating in step S157 are provided. If through the control panel D1 executed determination operation is accepted, the control part sets 10 the input information acquired before and after the time when the determination operation is performed as input data and updates the selected one of the learning models 1ma . 1Mb , ..., which is targeted as the target, by the input data and the operation details by the control unit (s) D1 , In the first example, the determination detail is by the control unit (s) D1 at least one of a gear stage and a gear ratio for the transmission e , In the second example, the determination detail is a supported position for the electric seat post F , In the third example, the determination detail is setting information (s) for a damping factor, a lift amount, and a lock state of the electric suspension G , The following description is made with reference to FIG 9 in embodiment 2. The actual control data is changed from "2" to "3" at the time when the control part (s) 10 the input information (s) associated with the time information (s) “1645” is recorded. Therefore, it is determined that the determination operation here from the control panel D1 is accepted. The control part 10 updates one of the learning models 1ma . 1Mb , ... which is selected on the basis of the training data which are identified with the control data “3” of the operation details, without the training data that are identified by identifying the time information “1334” before the time information “1645” with the Output information (s) “2” acquired input information (s) can be obtained, assigning the weighting to a low rating. The control part 10 can similarly label the input information (s) acquired for the time information (s) “1028” with the control data rewritten from “2” to “3”, and the labeled input information (s) into the corresponding learning model 1ma . 1Mb , ... enter and update the same.

In step S163 can the control part 10 one of the learning models 1ma . 1Mb , ... update by increasing the number of acquisitions instead of or in addition to increasing the acquisition frequency.

In embodiment 3, the control part ends 10 depending on whether the elapsed time is equal to or greater than the specified time in step S161 is the operation in the learning mode related to the operation by the operating unit D1 is started as a trigger, and switches it to automatic control mode. Instead of step S161 can the control part 10 induce one of the learning models 1ma . 1Mb , ... that in step 171 have been updated to output information (s) related to the controller, to check the output information (s) related to the controller based on the control data when a determination operation by the operating part D1 to accept the output information (s) from one of the learning models 1ma . 1Mb , ... reevaluate based on the result of the check and end the operation in the learning mode when a high score occurs. The control part 10 delivers a high rating if that of any of the learning models 1ma . 1Mb , ... output information (s) output match the control data determined by the operation within a predetermined error range.

(Embodiment 4)

In embodiment 4, the learning model generation processing and the component control processing are performed by a user's terminal instead of the control unit 100 executed. 14 Fig. 3 is a block diagram showing the construction of a control system 200 represents. The tax system 200 contains a terminal 2 and the control unit 100 , The control unit 100 in embodiment 4 includes a control part 11 , a storage part 12 , an input / output part 14 and a communication part 16 , Some of the components of the control unit 100 in embodiment 4, that of the control unit 100 in Embodiments 1 and 2 are common with the same reference numerals, and the detailed description thereof is not repeated.

The control part 11 the control unit 100 in embodiment 3, which is a processor using a CPU, performs processing by controlling each of the components using a memory such as a built-in ROM, RAM, etc. The control part 11 does not carry out that of the control part 10 training processing performed by the control unit in Embodiment 1. The control part 11 receives an input from the sensor groups S1 - S5 and S61 - S63 working on the human powered vehicle A are mounted, via the input / output part 14 and transmits the input via the communication part (s) 16 to the terminal 2. The control part 11 relates to a control state of the operating device D and one from the control device D issued operation signal and transmits it via the / the communication part 16 to the terminal 2. The control part 11 puts on the basis of the operating device D output operation signal or an instruction issued by the terminal 2, a control signal to the electric actuator to be controlled E1 on.

The communication part 16 is a communication port, and the control part 11 transmits and receives information to and from the terminal 2 via the communication part 16 , The communication part 16 is a USB (Universal Serial Bus) port in the first example. The communication part 16 in the second example is a wireless short-range communication module, for example Bluetooth (registered trademark).

The terminal 2 is a portable compact communication terminal that is to be used by the user. The terminal 2 is a smartphone in the first example. In the second example, the terminal 2 is a portable device, for example a so-called smart watch or the like. For the smartphone, a smartphone holder on the handlebar A2 of the human powered vehicle A attached and the smartphone can be used while it is placed on the holder (see 15 ).

The terminal 2 contains a control part 20 , a storage part 21 , a display part (s) 23 , a control unit 24 , an imaging part 25 , a first / first communication part 26 and a second / second communication part 27 ,

The control part 20 contains a processor such as a CPU, a GPU and memory and so on. The control part 20 can be constructed as individual hardware (SoC: System On a Chip), which in the processor, the memory, the memory part 21 , the first part (s) of communication 26 and the first part (s) of communication 27 is integrated.

The control part 20 conducts training of the output information (s) related to the control of the human powered vehicle A and component control based on training according to an application program 20P that in the memory part 21 is saved.

The storage part 21 contains non-volatile memory, such as flash memory. The storage part 21 saves a tutorial 10P and the application program 20P , The tutorial 10P can in the application program 20P be included. The storage part 21 saves several learning models 2Ma . 2Mb , ... by the from the control section 20 processing performed. The storage part 21 stores data on which the control part (s) relate 20 refers. The tutorial 10P can be obtained by using on a recording medium 28 saved tutorial 28P read out and in the / the memory part 21 is copied. The application program 20P can be obtained by putting one on a storage device 29 saved application program 29P read out and in the / the memory part 21 is copied.

The display part 23 includes a display device such as a liquid crystal panel, an organic electroluminescent display (EL display), or the like. The control unit 24 is an interface for accepting an operation by the user and includes a physical button and a touch pad device integrated into the display. The control unit 24 can be one on the screen by the display part 23 is displayed, accept the operation performed using the physical button or the touch panel.

The voice-input-output part 22 contains a speaker, a microphone and so on. The voice-input-output part 22 is with a speech recognition part 221 provided and can accept the operation by recognizing the details of the operation from the voice input through the microphone.

The imaging part 25 outputs a video signal obtained using an image sensor. The control part 20 can obtain an image at any time by an image sensor of the imaging part 25 is photographed.

The first part of communication 26 is a communication module that the communication part 16 the control unit 100 equivalent. The first part of communication 26 is a USB communication port in the first example. The first part of communication 26 is a short-range wireless communication module in the second example.

The second part of communication 27 is a wireless communication module for achieving transmission and reception of information (s) to and from another communication device (not shown) via a public communication network or according to a predetermined mobile communication standard. The second part of communication 27 uses a network card, a wireless communication device or a bearer communication module.

In embodiment 4, the control unit detects 100 continuously through the input-output part 14 (a) input information (s) from the sensor groups S1 - S5 and S61 - S63 will be obtained on the human powered vehicle A are mounted, and transmits the input information (s) via the communication part 16 to the terminal 2.

The control part 20 of the terminal 2 operates in the learning mode and in the automatic control mode. The control part 20 works in other different modes, both in learning mode and in automatic control mode. The examples of the details of the various modes are the same as those described in Embodiment 1. The control part 20 of the terminal 2 functions as a generation part, which the different learning models 2Ma . 2Mb , ... that correspond to the different modes through a learning algorithm of in-depth learning based on the learning program 10P generated when it is operated in learning mode. In the learning mode, the terminal 2 corresponds to a “generating device”. The control part 20 of the terminal 2 functions as a control part that includes the components including the transmission e using one of the learning models 2Ma . 2Mb automatically controls ... that correspond to a selected one of the various modes available on the application program 20P are based when they are operated in automatic control mode. In the automatic control mode, the terminal 2 corresponds to a “component control device”.

The operation of the terminal 2 in the learning mode is similar to that of the control part 10 in the control unit 100 shown in Embodiment 1 or Embodiment 2. The operation of the terminal 2 in the automatic control mode is also similar to that of the control part 10 in the control unit 100 shown in Embodiment 1 or Embodiment 2.

In embodiment 4, the terminal 2 corresponding to the “generation device” or the “component control device” accepts a rating by the user based on the information (s) provided by the operating part 24 or the imaging part (s) 25 be obtained and updated one of the learning model 2Ma . 2Mb , according to a selected mode depending on the details of the rating.

15 and 19 each illustrate an example of a screen based on the application program 20P is shown. In the in 15 illustrated example, the terminal 2 is a smartphone and on the handlebar A2 attached so that the user the display part (s) 23 able to see. 15 illustrates a main screen 230 based on the application program 20P on the display part 23 is shown. The control part 20 causes the display part (s) 23 , the main screen 230 display when running the application program 20P is selected. The control part 20 establishes communication with the control unit 100 forth when executing the application program 20P is started. The control part 20 can display a message indicating a communication status on the main screen 230 show as in 15 is shown. On the main screen 230 are selection buttons 232 and 234 for learning mode and automatic control mode included. The control part 20 starts operation in learning mode when the selection button 232 is selected, whereas the control part (s) 20 operation in automatic control mode starts when the selection button 234 is selected.

16 shows an example of the details of a learning mode screen 236 that appears when the selection button 232 is selected for the learning mode. On the learning mode screen 236 is a mode selection screen 238 contain. The mode selection screen 238 contains a button 240 , a button 242 and a button 244 that correspond to the modes to be selected. The control part 20 accepts the mode selection with the buttons 240 - 244 , The control part 20 accepts the mode selection of the mode of one of the different learning models 2Ma . 2Mb , ... corresponds with the buttons 240 - 244 ,

17 shows a screen example when one of the different modes is selected in learning mode. 17 shows the learning mode screen 236 when the button 242 is selected. After a mode has been selected, the control section shows 20 the learning mode screen 236 and executes the training processing. On the screen example in 17 a message is displayed indicating that a workout is in progress. On the learning mode screen 236 is a set button 246 contain. 18 shows a screen example when the setting button 246 is selected. When the setting button 246 is selected, the control part (s) shows 20 on the learning mode screen 236 , as in 18 shown, a setting screen 248 to accept a combination of input information that is one of the learning models 2Ma . 2Mb , ... which correspond to the selected mode and the setting of the weights which are to be assigned to the respective parts of the input information (s). If a setting button 250 on the setting screen 248 who in 18 is selected, the control part (s) stores 20 by the control panel (s) 24 in input fields within the setting screen 248 details entered or selected as setting information (s) in the storage part 12 ,

19 is an example of the content of a screen 252 for the automatic control mode to be displayed when the selection button 234 is selected for the automatic control mode. If for the automatic control mode screen 252 it is determined that an output of a control signal to the transmission e is required, and a control signal is output, shows the control part (s) 20 a message stating that a Gear level or a gear ratio has been changed, or outputs the message by voice from the voice input / output part 22 out. The control part 20 accepts a rating of this message. The control part 20 accepts the user rating for the action of each of the components based on that of each of the learning models 2Ma . 2Mb , ... output information that differs for each mode by an example described below.

The first example of the “rating acceptance part” contains a high rating button 254 and a low rating button 256 to accept a rating in the in 19 Automatic control mode screen shown 252 are included. The control part 20 recognizes via the touch field of the control panel 24 on the rating button 254 or 256 performed selection operation and accepts the evaluation. The high rating button 254 in the in 19 The example shown is selected if the automatic control of the gear stage or the gear ratio is comfortable. The low rating button 256 is selected if the automatic control is not comfortable. The control part 20 recognizes which of the buttons 254 and 256 is selected and recognizes the accepted rating details.

As a modification of the first example of the "rating acceptance part", only the low rating button can be used 256 are displayed. As a modification of the first example of the “evaluation acceptance part”, a key for executing such an evaluation that the shifted gear ratio or gear ratio is too heavy (too far FORWARD) and a key for executing such an evaluation that it / it is too easy (too INWARD).

The second example of the "evaluation acceptance part" can be a physical button on the control unit D1 is provided. A specific evaluation acceptance key can be on the control panel D1 be provided. An evaluation acceptance key can be placed separately near the operating device D be provided. The control part 20 can recognize the details of the rating based on whether or not the specific rating acceptance button is pressed, and the manner in which a multi-purpose button is pressed and a combination with a lever operation. The third example of the "evaluation acceptance part" is a speech recognition part 221 of the voice input-output part 22 ,

The control part 20 accepts a rating by recognizing the user's voice by the speech recognition part 221 , The control part 20 Determines whether the automatic control is comfortable or not based on the recognized voice. The fourth example of the "evaluation acceptance part" is the imaging part 25 , The control part 20 Fig. 12 is a specification part for specifying facial expressions from an image obtained by photographing the face of the user who is the human-powered vehicle A drives and accepts the rating based on the specified facial expressions. The rating is obtained by determining whether the automatic control is comfortable or not, and a high rating is accepted if it is comfortable.

20 and 21 are each a flowchart showing an example of the processing procedure performed by the control part 20 is carried out in embodiment 4. The control part 20 repeatedly executes the following processing procedure in the automatic control mode. It can be executed at a predetermined control period (e.g. 30 milliseconds).

The control part 20 determines one of the different modes on the mode selection screen 238 in 16 selected mode (step S301 ). The control part 20 accepts the selection of one of the different learning models 2Ma . 2Mb , ... through the mode selection on the mode selection screen 238 in step S301 , The control part 20 selects one of the generated learning models 2Ma . 2Mb , ... according to the specific mode and reads it from the memory section 21 off (step S303 ).

The control part 20 captures input information related to the driving of the human powered vehicle A by the control unit 100 through the first part of the communication 26 will be obtained (step S305 ). In step S305 is the control part 20 a “detection part” of the “component control device”. In step S305 detects the control part 20 Signal level from the control section 10 the control unit 100 related to the signal levels from the sensor groups S1 - S5 and S61 - S63 for each tax period and temporarily stores them in the internal memory of the control section 10 or in the input-output section 14 integrated memory.

The control part 20 reads the setting that the in step S301 specific mode, or the setting corresponding to that in step S303 selected learning model 2Ma . 2Mb , ... corresponds to from the / in the memory part 21 stored setting information (s) from (step S307 ). The control part 20 captures input information related to the selected one of the learning models 2Ma . 2Mb , ... are assigned based on the setting from the step (s) S305 input information (s) recorded (step S309 ). The control part 20 points out the input information (s) recorded in the setting for (step S311 ).

The control part 20 gives that in step S309 selected and in step S311 with the weighted input information (s) in the in step S303 selected learning model 2Ma . 2Mb , ... a step S313 ).

The control part 20 specifies output information (s) regarding the control of the component to be controlled that is to be output by the captured input information (s) in the selected one of the learning models 2Ma . 2Mb , ... is entered (step S315 ). The control part 20 specified in step S315 for example, a result of the distinction between a gear stage and a gear ratio for the transmission e as output information (s).

The control part 20 refers to the state of the component to be controlled based on the specified output information (s) (step S317 ).

The control part 20 determines whether or not an output of a control signal is required based on the relationship between the details of the control indicated by the specified output information (s) and the designated state (step S319 ).

If in step S319 it is determined that an output of a control signal is required ( S319 : YES), gives the control part 20 a control instruction for controlling the component based on the specified output information (s) to the control unit 100 off (step S321 ). If in step S319 it is determined that an output of a control signal is not required ( S319 : NO), the control part moves 20 with processing to step S323 continued.

The control part 20 accepts a rating within a predetermined period of time after processing in the step S321 (Step S323 ).

The control part 20 determines whether or not the accepted rating is a high rating (step S325 ). If it is determined that the rating is high ( S325 : YES), the control part ends 20 the control processing for a control cycle.

If in step S325 it is determined that the rating is low ( S325 : NO), the control part begins 20 work in learning mode, ie the selected one of the learning models 2Ma . 2Mb , ... to update (step S327 ). The control part 20 can display a message indicating that the learning mode is on the display part 23 started.

In step S325 can the control part 20 start the selected learning model 2Ma . 2Mb , ... update when a rating is accepted regardless of whether the accepted rating is a high rating or not.

The control part 20 begins to measure the elapsed time from the start of the learning mode (step S401 ). The control part 20 generates training data including the details of the step S323 accepted assessment by the selected one of the learning models 2Ma . 2Mb , ... output information (s) output in step S315 were selected according to the time the user accepts the rating and the input information (s) associated with that in step S305 captured output information (step) S402 ). The control part 20 arranges the training data weights based on the details of the step S323 accepted rating for (step S403 ). The control part 20 updates the selected one of the learning models 2Ma . 2Mb , ... based on the weighted training data (step S405 ). In step S405 instructs the control section 20 Weights that indicate a higher level of importance for the training data when the rating is lower.

The control part 20 determines whether the elapsed time since the start of the measurement in step S401 is equal to or longer than a predetermined time or not (step S407 ).

When it is determined that the elapsed time is shorter than the predetermined time ( S407 : NO), detects the control part (s) 20 (a) input information (s) in learning mode (step S409 ) and refers to actual control data based on the control panel D1 (Step S411 ).

The control part 20 identifies those in step S409 acquired input information (s) with the control data referred to, thereby generating training data (step S413 ), and then updates the learning models 2Ma . 2Mb , ... based on the training data of the marked input information (step) S415 ) and directs processing to step S407 back.

If in step S407 it is determined that the elapsed time is equal to or longer than the predetermined time ( S407 : YES), the control part ends 20 the measurement of the elapsed time (step S417 ) and transmits the selected one of the models 2Ma . 2Mb , ..., which has been updated in connection with the identification information (s), via the second communication part 27 to another Server device (step S419 ). The control part 20 switches operation to automatic control mode (step S421 ) and ends the processing.

The control part 20 can do any of the different learning models 2Ma . 2Mb , ... from the second communication part 27 transmitted to another server at a predetermined frequency, such as once a day or once a week, not at the end of the learning mode, while the terminal is powered by the human powered vehicle A is separated.

In embodiment 4, the terminal 2 can temporarily enter information that has been entered successively in time series in the memory section 21 in connection with actual control data similar to that in Embodiment 2. In this case, the control part (s) 20 perform the following processing instead of updating using the training data that in step S405 Weights are assigned based on the details of the rating. The control part 20 sets multiple input information as input before and after the time at which the evaluation is accepted by the evaluation acceptance part, and updates a selected one of the learning models 2Ma . 2Mb , ... using the training data including that from the selected one of the learning models 2Ma . 2Mb , ... output information (s) output and the details of the rating accepted by the rating accepting part.

In embodiment 4, the terminal 2 transmits via the second communication part 27 the different learning models generated 2Ma . 2Mb , ... that are addressed to another server. The control unit 100 in embodiment 1, a communication part such as the second communication part can 27 contain and can the generated learning model 1M transmitted, which is addressed to another server.

In embodiments 1 to 4, output information (s) provided by the learning models 1ma . 1Mb , ... or the learning models 2Ma . 2Mb , ... are output, mainly control data relating to the control of the transmission e described. The component of the human powered vehicle to be controlled A can the electric seat post F or the electric suspension G instead of the gearbox e his. The components of the human-powered vehicle to be controlled A can the gear e who have favourited Electric Seat Post F and the electric suspension G or the electric seat post F and the electric suspension G his.

LIST OF REFERENCE NUMBERS

100

Control unit (generating device and component control device)

10, 11

control part

12

memory part

1Ma, 1Mb

learning model

1P

tutorial

2P

control program

2

terminal

20

control part

21

memory part

2Ma, 2Mb

learning model

10P

tutorial

20p

application program

A4

rear wheel

B1

crank

B3

second sprocket assembly

D

operating device

D1

control panel

e

transmission

E1

electrical actuator

F

electric saddle post

G

electric suspension

S1

speed sensor

S2

Cadence Sensor

S3

torque sensor

S4

gyro sensor

S5

image sensor

S61, S62, S63

position sensor

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• JP 2018130244 [0001]
• JP 2018130245 [0001]
• JP 2018130246 [0001]
• JP 2018236760 [0001]
• JP 2018236761 [0001]
• JP 2018 [0001]
• JP 236762 [0001]
• JP 2018236763 [0001]
• JP 2018236764 [0001]
• JP 2018236765 [0001]
• JP 236761 [0001]
• JP 2018236762 [0001]
• US 6047230 [0003,0004]

Claims (27)

A generating device comprising: a detection part that detects input information (s) regarding the driving of a human-powered vehicle; and a generation part that generates various learning models, each of which generates output information (s) regarding the control of a component of the human-powered vehicle based on input information (s) acquired by the detection part , Generation device after Claim 1 wherein the output information (s) includes at least one of a gear stage and a gear ratio for a transmission. Generation device after Claim 1 or 2 wherein the input information (s) comprises at least one of a driving speed of the human powered vehicle and a cadence of a crank of a drive mechanism. The generating device according to one of claims 1 to 3, wherein the input information (s) acquisition data of one or more of an attitude of the human-powered vehicle; a position of a user driving the human powered vehicle; and includes a driving environment of the human powered vehicle. 5. The generation device according to claim 1, wherein the generation part (s) inputs input information (s) that, in combination of a combination of information for another of the learning models, from the input information acquired by the detection part for each of the different ones Learning models differentiate / differentiate. Generation device according to one of claims 1 to 5, wherein the setting of the combination of different input information for each of the different learning models is accepted. Generation device according to one of claims 1 to 6, wherein the generation part (s) assigns several data elements contained in the input information (s) acquired by the acquisition part (s) for each of the different learning models to different weights from those of another learning model and enters the multiple weighted data items. A generating device according to any one of claims 1 to 7, wherein the setting of the different weights is accepted for each of the different learning models. Generation device according to one of claims 1 to 8, wherein the acquisition part (s) acquires input information at different times, and the generating part preferably enters the entered information (s) at the various times in each of the learning models, preferably An interval between the times at which the entered information is recorded differs between the different learning models. Generation device according to one of claims 1 to 9, wherein the acquisition part (s) acquires input information at different times and the generation part inputs a variation between the input information into the learning model, preferably a time period that corresponds to the variation between the input information (s) differs , between the different learning models. The generating device according to any one of claims 1 to 10, further comprising a selection part that selects a target learning model from the various learning models, the generating part generating a target learning model selected by the selection part. Generation device after Claim 11 , further comprising an assessment part that evaluates the output information (s) output from a target learning model selected by the selection part, the generation part generating the target learning model based on an evaluation by the updated the assessment part. Generation device after Claim 12 , further comprising a control part that accepts a determination operation regarding the output information (s), the evaluation part performing evaluation by using the output information (s) output from the target learning model In response to an input of the input information (s) acquired by the acquisition part into the target learning model against a determination operation accepted by the operating part. Generation device after Claim 13 , wherein if the output information (s) output from the target learning model and a detail of the determination operation do not match, either the evaluation part (s) provides the output information (s) with a low rating with respect to the target learning model or the generation part the target learning model by assigning a weighting based on training data, including the details of the determination operation. Generation device after Claim 13 or 14 , wherein the detection part includes a storage part that temporarily stores a plurality of input information in succession in time series, and when a determination operation is accepted by the operating part, the generation part adds a plurality of input information acquired before and after a time which the determination operation is performed as an input and updates the target learning model using the input data and a detail of the operation by the operating unit (s). Generation device after Claim 15 when a determination operation is accepted by the operating part, wherein the generation part (s) updates the target learning model by increasing the number of acquisitions of input information (s) and / or increasing the acquisition frequency of input information (s), if one Determination operation is accepted by the control unit. Generating device according to one of claims 12 to 16, wherein the assessment part includes an assessment acceptance part that accepts an assessment provided by the user for the action of the component based on output information (s) output from each of the different learning models, and the target learning model is updated based on training data including a detail of the rating accepted by the rating acceptance part, output information (s) output from the learning model at a time when a rating is accepted, and input information (s) is updated in connection with the output information (s). Generation device after Claim 17 , further comprising an operation part (s) that accepts a determination operation regarding the output information (s), wherein the evaluation acceptance part is provided on the operation part or in the vicinity of the operation part to accept an evaluation operation by the user. Generation device after Claim 17 or 18 wherein the evaluation acceptance part includes a specification part that specifies facial expressions from a photographed image obtained by photographing a face of a user driving the human-powered vehicle and a user-made evaluation based on the facial expressions specified in the specified part. The generating device according to any one of claims 17 to 19, wherein the evaluation acceptance part includes a voice recognition part and accepts a rating by recognizing a user's voice. The generating device according to any one of claims 1 to 20, further comprising: a model selection part that accepts the selection of one of the various learning models generated; and a transmission part that transmits the learning model selected by the model selection part. A component control device comprising: various learning models each generated by a learning algorithm for outputting output information (s) regarding the control of a component of a human powered vehicle based on input information (s) related to the driving of the human powered vehicle; a detection part that detects the input information (s); a model selection part that accepts the selection of one of the different learning models; and a control part that controls the component based on output information (s) output from a selected one of the various learning models by inputting the input information (s) detected by the detection part into the a selected one of the learning models is entered. Component control device according to Claim 22 , wherein the acquisition part (s) acquires input information (s) that are brought into conformity with the one of the learning models selected by the model selection part, and inputs the input information (s). Component control device according to Claim 22 or 23 wherein the output information (s) includes at least one of a gear stage and a gear ratio for a transmission. Production process comprising: Acquiring input information (s) related to driving a human powered vehicle; and Generated by a learning algorithm of different learning models, each of which generates (one) output information (s) relating to the control of a component of the human-powered vehicle based on detected input information (s). A component control method comprising: acquiring input information (s) related to driving a human powered vehicle; Accepting the selection of any one of various learning models, each generated by a learning algorithm, such that output information (s) regarding the control of a component of the human powered vehicle is / are output based on the input information (s) ; Specifying output information (s) to be output from a selected one of the learning models by inputting the acquired input information (s) into the selected one of the learning models; and controlling the component based on the specified output information (s). Computer program that causes a computer to process the following: Acquiring input information (s) related to driving a human powered vehicle; and Generating different learning models by means of a learning algorithm, each of which generates (one) output information (s) relating to the control of a component of the human-powered vehicle based on (one) acquired input information (s).

Images