JP2020142566A - Evaluation device, control system, notification system, generation method for learning model, computer program and storage medium - Google Patents

Abstract

To provide an evaluation device, a control system, a notification system, a generation method for learning model, computer program and storage medium, used for a bicycle including an electrically-driven component.SOLUTION: The evaluation device evaluates the favorability of control for components of a human-power drive vehicle using a learning model according to travel information concerning travel of the human-power drive vehicle. The evaluation device highly evaluates such control that the traction of the human-power drive vehicle becomes better and evaluates low such control that traction loss is estimated.SELECTED DRAWING: Figure 4

Description

The present invention relates to an evaluation device, a control system, a notification system, a learning model generation method, a computer program, and a storage medium.

Bicycle control systems, including electric components, are known. For example, Patent Document 1 discloses a control system in which an operation device, a control device, and an electric component are wirelessly connected.

Japanese Unexamined Patent Publication No. 2016-16591

Currently, it is required to evaluate the control in the control system of the bicycle.

An object of the present invention is to provide an evaluation device, a control system, a notification system, a learning model generation method, a computer program, and a storage medium for evaluating the preference of control for a component of a human-powered vehicle.

The evaluation device according to the first aspect of the present invention evaluates the preference of control for the components of the human-powered vehicle by using a learning model according to the traveling information regarding the traveling of the human-powered vehicle.

According to this evaluation device, by using the learning model, it is possible to evaluate the preference of control for the component according to the driving information of the human-powered vehicle. Here, the preference of control is information suggesting whether the state of the human-powered vehicle is good or bad when the component is controlled. The preference of control is evaluated, for example, by the quality of traction of a human-powered vehicle. Therefore, the evaluation device highly evaluates the control that improves the traction of the human-powered vehicle, and evaluates the control that is presumed to lose traction.

The evaluation device of the second aspect according to the first aspect of the present invention retrains the learning model according to the evaluation result of the preference and the control result of the component according to the evaluation result.

According to this evaluation device, the learning model is retrained according to the evaluation result of the evaluation device and the control result of the component, so that a more preferable control state can be realized as the learning progresses.

In the evaluation device of the third aspect according to the first or second aspect of the present invention, the traveling information includes speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device, and the operation input. And information about at least one of the history of control results of the component, regeneration amount, slope, road surface condition, position, weather, and vehicle information.

According to this evaluation device, speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device, history of operation input and component control result, regeneration amount, inclination, road surface condition, position, The preference for control over the component can be assessed according to the weather and at least one of the vehicle information.

In the evaluation device of the fourth aspect according to the first or second aspect of the present invention, the traveling information of the manpower-driven vehicle includes information regarding the traction of the man-powered vehicle.

According to this evaluation device, it is possible to evaluate the preference of control for the component according to the information regarding the traction of the man-powered vehicle.

The control system according to the fifth aspect of the present invention includes an evaluation device according to any one of the first to fourth aspects, and a control device that controls the component based on the evaluation result of the evaluation device.

According to this control system, since the control device controls the components based on the evaluation result of the evaluation device, a more preferable control state can be realized without depending on the operation of the rider.

In the control system of the sixth aspect according to the fifth aspect of the present invention, the component includes an assist device including an electric drive unit that outputs an auxiliary driving force.

According to this control system, since the control device controls the assist device that outputs the auxiliary driving force, a more preferable control state can be realized with respect to the auxiliary drive control.

In the control system of the seventh aspect according to the fifth or sixth aspect of the present invention, the component includes a braking device including an electric drive unit.

According to this control system, since the control device controls the brake device, a more preferable control state can be realized with respect to the brake control.

In the control system of the eighth aspect according to any one of the fifth to seventh aspects of the present invention, the component includes an antilock braking device including an electric drive unit.

According to this control system, the control device controls the antilock brake device, so that a more preferable control state can be realized with respect to the antilock brake control.

In the control system of the ninth aspect according to any one of the fifth to eighth aspects of the present invention, the component includes a suspension device including an electric drive unit.

According to this control system, since the control device controls the suspension device, a more preferable control state can be realized with respect to the suspension control.

The notification system according to the tenth aspect of the present invention includes an evaluation device according to any one of the first to fourth aspects and a notification device for notifying information on the evaluation result of the evaluation device.

According to this notification system, since a notification device for notifying information on the evaluation result is provided, the rider can grasp the quality of the control state of the human-powered vehicle.

In the notification system of the eleventh aspect according to the tenth aspect of the present invention, the information regarding the evaluation result includes the information regarding the control state of the component.

According to this notification system, the information regarding the evaluation result includes the information regarding the control state of the component, so that the rider can grasp the quality of the control state of the component.

In the method of generating the learning model according to the twelfth aspect of the present invention, the preference of control for the components of the human-powered vehicle is evaluated by using a function according to the driving information regarding the running of the human-powered vehicle using a computer. , The learning model is generated by updating the function according to the evaluation result of the preference and the control result of the component according to the evaluation result.

According to this learning model generation method, it is possible to generate a learning model for evaluating the preference for controlling the components by collecting the driving information of the human-powered vehicle.

The computer program according to the thirteenth aspect of the present invention causes a computer to execute a process of evaluating the preference for control of the components of the human-powered vehicle by using a learning model according to the traveling information regarding the running of the human-powered vehicle. Is a computer program for.

According to this computer program, an execution environment for evaluating the control ambition for the components according to the driving information of the human-powered vehicle can be realized by the computer.

The storage medium according to the 14th aspect of the present invention stores the computer program according to the 13th aspect.

According to this storage medium, by installing a computer program stored in the storage medium in the computer, it is possible to realize an execution environment in which the control ambition for the component is evaluated according to the driving information of the human-powered vehicle.

According to the present application, it is possible to evaluate the preference for control over the components of a human-powered vehicle.

It is a side view of the human-powered vehicle to which the evaluation device of 1st Embodiment is applied. It is a block diagram which shows the internal structure of the evaluation apparatus which concerns on 1st Embodiment. It is a schematic diagram which shows the implementation example of the learning model. It is a flowchart explaining the learning procedure of a learning model. It is a block diagram explaining the control system which concerns on 2nd Embodiment. It is a flowchart explaining the procedure of the process executed by a control device. It is a flowchart explaining the generation procedure of a learning model. It is a flowchart explaining the re-learning procedure of a learning model. It is a block diagram explaining the notification system which concerns on 5th Embodiment.

Hereinafter, the present invention will be specifically described with reference to the drawings showing the embodiments thereof.
(First Embodiment)
FIG. 1 is a side view of the human-powered vehicle 1 to which the evaluation device 100 of the first embodiment is applied. The human-powered vehicle 1 is a vehicle that uses human power at least partially with respect to the driving force for traveling. A vehicle that uses only an internal combustion engine or an electric motor as a driving force is excluded from the human-powered vehicle 1 of the present embodiment. The human-powered vehicle 1 is, for example, a bicycle including a mountain bike, a road bike, a cross bike, a city cycle, and the like.

The human-powered vehicle 1 includes a vehicle body 10, front wheels 12, rear wheels 14, handlebars 16, seats 18, and a drive mechanism 20. In the following description, the terms representing the front-rear, left-right, and up-down directions are used with reference to the direction in which the rider is seated on the seat 18 of the human-powered vehicle 1.

The vehicle body 10 includes a frame 10A and a front fork 10B. The front wheel 12 is rotatably supported by the front fork 10B. The rear wheel 14 is rotatably supported by the frame 10A. The handlebar 16 is supported by the frame 10A so that the traveling direction of the front wheels 12 can be changed.

The drive mechanism 20 transmits a human-powered driving force to the rear wheels 14. The drive mechanism 20 includes a crank 22, a front sprocket assembly 24A, a rear sprocket assembly 24B, a chain 26, and a pair of pedals 28, 28. The drive mechanism 20 may further include a chain device for suppressing the chain 26 from coming off.

The crank 22 includes a right crank 22A, a left crank 22B, and a crankshaft 22C. The crankshaft 22C is rotatably supported by the frame 10A. The right crank 22A and the left crank 22B are respectively connected to the crankshaft 22C. One of the pair of pedals 28, 28 is rotatably supported by the right crank 22A and the other is rotatably supported by the left crank 22B.

The front sprocket assembly 24A is connected to the crankshaft 22C and rotates integrally with the crankshaft 22C. The front sprocket assembly 24A includes, for example, a plurality of front sprockets having different outer diameters. The outer diameters of the plurality of front sprockets increase outward from the central surface of the vehicle body 10 in a direction parallel to the rotation axis of the crankshaft 22.

The rear sprocket assembly 24B is rotatably supported by a hub (not shown) of the rear wheel 14. The rear sprocket assembly 24B includes, for example, a plurality of rear sprockets having different outer diameters. The outer diameters of the plurality of rear sprockets decrease from the central surface of the vehicle body 10 toward the outside in a direction parallel to the rotation axis of the hub assembly.

The chain 26 is wound around the front sprocket assembly 24A and the rear sprocket assembly 24B. When the crank 22 rolls forward due to the human-powered driving force applied to the pedals 28 and 28, the front sprocket assembly 24A rolls forward together with the crank 22. The rotation of the front sprocket assembly 24A is transmitted to the rear sprocket assembly 24B via the chain 26 to rotate the rear wheels 14. A belt or shaft may be used instead of the chain 26.

The human-powered vehicle 1 further includes an operating device 30, transmissions 32 and 33, an assisting device 34, a braking device 36, an antilock braking device 38, a suspension device 40, and a battery unit 42.

The operating device 30 is provided on, for example, the handlebar 16. The operation switch 30 includes operation switches 30A and 30B operated by a rider's finger. The operation switches 30A and 30B are switches for controlling at least one of the transmission 32, the assist device 34, the brake device 36, the antilock brake device 38, and the suspension device 40.

The number of operating devices 30 mounted on the human-powered vehicle 1 does not have to be one, and may be plural. In the example shown in FIG. 1, the operation device 30 is configured to include two operation switches 30A and 30B, but may be configured to include one or three or more operation switches. The operation device 30 is not limited to the configuration including the switch, and may be configured to include an operation lever, an operation dial, and the like.

The operation device 30 is connected to each component so that a control signal corresponding to the operation of the operation switch 30A or the operation switch 30B can be transmitted to the component to be controlled. In one example, the operating device 30 is connected to a controlled component by a communication line or an electric wire capable of PLC (Power Line Communication). The component to be controlled is at least one of the transmission 32, the assist device 34, the brake device 36, the antilock brake device 38, and the suspension device 40. In another example, the operating device 30 is connected to a controlled component by a wireless communication unit capable of wireless communication. The operation device 30 may further output information regarding the operation input to the evaluation device 100.

The transmission 32 is one of the components of the human-powered vehicle 1 whose operation is controlled according to the control signal transmitted from the operation device 30. The transmission 32 includes an electric actuator 32A and a chain guide 32B. An example of the electric actuator 32A is an electric motor. The transmission 32 drives the electric actuator 32A and operates the chain guide 32 to change the rear sprocket around which the chain 26 is wound and switch the gear ratio of the human-powered vehicle 1.

An example of the transmission 32 is an external transmission. When the transmission 32 is an external transmission, the rear sprocket assembly 24B includes a plurality of rear sprockets having different outer diameters. When the transmission 32 receives a control signal instructing upshifting from the operation device 30, the transmission 32 is electrically operated so as to change the rear sprocket around which the chain 26 is wound to a rear sprocket having an outer diameter smaller than that of the current rear sprocket. The chain guide 32B is operated by the actuator 32A. When the transmission 32 receives a control signal instructing downshifting from the operation device 30, the transmission 32 is electrically operated so as to change the rear sprocket around which the chain 26 is wound to a rear sprocket having an outer diameter larger than that of the current rear sprocket. The chain guide 32B is operated by the actuator 32A.

The transmission 33 is one of the components of the human-powered vehicle 1 whose operation is controlled according to the control signal transmitted from the operation device 30. The transmission 33 includes an electric actuator 33A and a chain guide 33B. An example of the electric actuator 33A is an electric motor. The transmission 33 operates the chain guide 33B by driving the electric actuator 33A to change the front sprocket around which the chain 26 is wound and switches the gear ratio of the human-powered vehicle 1.

An example of the transmission 33 is an external transmission. When the transmission 33 is an external transmission, the front sprocket assembly 24A includes a plurality of front sprockets having different outer diameters. In the present embodiment, when the transmission 33 receives the control signal instructing the shift up from the operation device 30, the front sprocket around which the chain 26 is wound is changed to a sprocket having an outer diameter larger than that of the current sprocket. The chain guide 33B is operated by the electric actuator 33A. When the transmission 33 receives a control signal instructing downshifting from the operation device 30, for example, the front sprocket around which the chain 26 is wound is changed to a front sprocket having an outer diameter smaller than that of the current front sprocket. , The chain guide 33B is operated by the electric actuator 33A.

The transmissions 32 and 33 may be internal transmissions. For example, when the transmission 32 is an internal transmission, the transmission 32 is provided, for example, on the hub of the rear wheel 14, and the rotation input to the rear sprocket assembly 24B is changed and transmitted to the rear wheel 14. At this time, the rear sprocket assembly 24B contains a single rear sprocket. At least one of the transmissions 32 and 33 may be a continuously variable transmission.

The assist device 34 is provided for the drive mechanism 20. The assist device 34 has an electric drive unit 34A. The electric drive unit 34A includes a drive circuit, a memory, an electric motor, and the like, and the drive circuit operates the electric motor to output an auxiliary driving force to assist the traveling of the human-powered vehicle 1. In the present embodiment, the assist device 34 is one of the components automatically controlled by the control device 200 described later. The control device 200 is shown in FIG. The assist device 34 may be controlled according to a control signal output from the operation device 30. The control state of the assist device 34 includes a plurality of control states having different auxiliary driving forces. The control state in the assist device 34 is also referred to as an assist level. The assist device 34 changes the assist level according to, for example, a human-powered driving force acting on the crank 22. The assist level includes, for example, a strong assist level, a medium assist level, a weak assist level, and a zero assist level when four levels are provided. At the zero assist level, the motor unit 34 is not driven and assist is not performed.

The assist device 34 may also be configured to function as a regenerative braking device for braking the human-powered vehicle 1. The control result of the assist device 34 including the regeneration amount may be output to the evaluation device 100 and stored as a history in the storage unit 106 of the evaluation device 100. The storage unit 106 of the evaluation device 100 is shown in FIG.

The braking device 36 is provided for the front wheels 12 and the rear wheels 14. The brake device 36 provided on the front wheel 12 is a rim brake including an electric drive unit 36A and a caliper 36B including a brake pad that can contact the wheels of the front wheel 12. The electric drive unit 36A of the brake device 36 includes a drive circuit and an electric motor. The brake device 36 is one of the components automatically controlled by the control device 200. The brake device 36 may be connected to the brake operating device 36C provided on the handlebar 16. In response to the control signal from the control device 200 or the operation of the brake operating device 36C, the drive circuit of the electric drive unit 36A operates the caliper 36B by the electric motor to bring the brake pads into contact with the wheels of the front wheels 12. , The rotational force of the front wheel 12 is attenuated. The brake device 36 can operate in a plurality of control states having different braking forces. The control result of the brake device 36 may be output to the evaluation device 100 and stored as a history in the storage unit 106 of the evaluation device 100. The brake device 36 may output information regarding an operation input to the brake operation device 36C to the evaluation device 100. The information regarding the operation input includes, for example, a method of operating the brake, an operation amount, and information indicating which of the left and right brake devices 36 is the operation.

The brake device 36 provided on the rear wheel 14 is a rim brake including an electric drive unit 36A and a caliper 36B including a brake pad that can contact the wheels of the rear wheel 14. The electric drive unit 36A of the brake device 36 includes a drive circuit and an electric motor. The brake device 36 is one of the components automatically controlled by the control device 200. The brake device 36 may be connected to the brake operating device 36C provided on the handlebar 16. In response to the control signal from the control device 200 or the operation of the brake operating device 36C, the drive circuit of the electric drive unit 36A operates the caliper 36B by the electric motor to bring the brake pads into contact with the wheels of the rear wheels 14. Attenuates the rotational force of the rear wheels 14. The brake device 36 can operate in a plurality of control states having different braking forces. The control result of the brake device 36 may be output to the evaluation device 100 and stored as a history in the storage unit 106 of the evaluation device 100. The brake device 36 may output information regarding an operation input to the brake operation device 36C to the evaluation device 100.

The braking device 36 may be a disc brake including a brake pad that can come into contact with a rotating body attached to the hub of the front wheel 12 and the hub of the rear wheel 14. In a disc brake, for example, a fluid is controlled by an electric motor to move a brake pad, and the brake pad is pressed against a rotating body (disc brake rotor) to attenuate the rotational force of the front wheels 12 or the rear wheels 14. Alternatively, the disc brake directly moves the brake pads by the electric motor and presses the brake pads against the rotating body to attenuate the rotational force of the front wheels 12 or the rear wheels 14.

The anti-lock braking device 38 is provided for the braking device 36. The anti-lock braking device 38 includes an electric drive unit 38A including a drive circuit, a reservoir for storing hydraulic oil, a pump for supplying hydraulic oil to the caliper 36B, and a valve for adjusting the hydraulic pressure given to the caliper 36B of the braking device 36. Etc. are provided. The electric drive unit 38A controls the opening and closing of the valve and adjusts the hydraulic pressure applied to the caliper 36B to control the braking force applied from the caliper 36B to the front wheels 12 or the rear wheels 14. The antilock brake device 38 is one of the components automatically controlled by the control device 200. The antilock brake device 38 can operate in a plurality of control states in which the magnitude of the hydraulic pressure applied to the caliper 36B is different. The control result of the antilock brake device 38 is output to the evaluation device 100 and stored as a history in the storage unit 106 of the evaluation device 100.

The suspension device 40 is, for example, a front suspension provided on the front fork 10B to attenuate the impact applied to the front wheels 12. In another example, the suspension device 40 is a rear suspension that damps the impact applied to the rear wheels 14. The suspension device 40 includes an electric drive unit 40A. The electric drive unit 40A includes a drive circuit and an electric motor. The suspension device 40 is controlled by setting, for example, a damping factor, a repulsive force, a stroke amount, and a lockout state as operating parameters. The suspension device 40 is one of the components automatically controlled by the control device 200. The suspension device 40 can operate in a plurality of control states having different cushioning properties. The control result of the suspension device 40 is output to the evaluation device 100 and stored as a history in the storage unit 106 of the evaluation device 100.

The battery unit 42 includes a battery 42A and a battery holder 42B. Battery 42A is a storage battery that includes one or more battery cells. The battery holder 42B is fixed to the frame 10A of the human-powered vehicle 1. The battery 42A is detachably attached to the battery holder 42B. The battery 42A is electrically connected to the transmission 32, the assist device 34, the brake device 36, the antilock brake device 38, the suspension device 40, and the like.

The human-powered vehicle 1 further includes a vehicle speed sensor S1, an acceleration sensor S2, an attitude sensor S3, a load sensor S4, a torque sensor S5, a cadence sensor S6, an inclination sensor S7, an image sensor S8, a position sensor S9, and a weather sensor S10. At least one may be provided. In the following description, when it is not necessary to explain each of them individually, the sensors S1 to S10 are also described.

The vehicle speed sensor S1 is a sensor that outputs a signal corresponding to the speed of the human-powered vehicle 1. The vehicle speed sensor S1 is attached to the frame 10A or the front fork 10B. The vehicle speed sensor S1 includes, for example, a Hall element, and measures the rotational speed of the front wheels 12 or the rear wheels 14 by detecting magnets (not shown) provided on the front wheels 12 or the rear wheels 14. In one example, the vehicle speed sensor S1 outputs a signal indicating the rotational speed of the front wheels 12 or the rear wheels 14. In another example, the vehicle speed sensor S1 calculates the speed of the human-powered vehicle 1 based on the measured rotation speeds of the front wheels 12 or the rear wheels 14, and outputs a signal indicating the calculated speed.

The acceleration sensor S2 is a sensor that outputs a signal corresponding to the acceleration in the front-rear direction of the human-powered vehicle 1. The acceleration sensor S2 is provided on the frame 10A, for example, and measures the acceleration by detecting the displacement of the weight when the acceleration in the front-rear direction is applied as a change in capacitance or resistance value. The acceleration sensor S2 may be a motion sensor that measures the acceleration in the yaw direction, the roll direction, and the pitch direction of the human-powered vehicle 1.

The posture sensor S3 is a sensor that outputs a signal according to the posture of the rider. An example of the posture sensor S3 is a piezoelectric sensor, which is provided at a plurality of locations of the human-powered vehicle 1 on which the weight of the rider is applied. For example, the posture sensor S3 is provided at one or a plurality of locations such as the grip provided on the handlebar 16, the surface of the seat 18, and the pedal 28. The posture sensor S3 may be a motion sensor built in a wearable terminal mounted on the rider.

The load sensor S4 is a sensor that outputs a signal corresponding to the load applied to the human-powered vehicle 1. An example of the load sensor S4 is a strain gauge type load cell. The load sensor S4 is provided at one location or several locations of the human-powered vehicle 1 such as the axle of the front wheel 12 or the rear wheel 14, the handlebar 16, the seat 18, and the pedal 28. The load sensor S4 may be a sensor that measures the ground contact load of the front wheels 12 and the rear wheels 14, or may be a sensor that measures the load balance of the human-powered vehicle 1. At least a part of the attitude sensor S3 and the load sensor S4 may be shared.

The torque sensor S5 is a sensor that outputs a signal corresponding to the torque applied to the crank 22. The torque sensor S5 is provided in either the right crank 22A, the left crank 22B, the crankshaft 22C, the front sprocket assembly 24A, or the drive path from the crankshaft 22C to the front sprocket assembly 24A. The torque sensor S5 includes a strain sensor, a magnetostriction sensor, an optical sensor, a pressure sensor, and the like, and detects the torque applied to the crank 22.

The cadence sensor S6 is a sensor that outputs a signal indicating cadence. The cadence sensor S6 is attached to the frame 10A. The cadence sensor S6 includes, for example, a Hall element, and measures the number of rotations of the crankshaft 22C per unit time by detecting a magnet (not shown) provided on the right crank 22A or the left crank 22B.

The evaluation device 100 applies power to the human-powered vehicle 100 based on the torque applied to the crank 22 detected by the torque sensor S5 and the rotation speed of the crankshaft 22C detected by the cadence sensor S6 per unit time. Can be calculated.

The tilt sensor S7 is a sensor that outputs a signal according to the tilt angle of the human-powered vehicle 1. The tilt angle detected by the tilt sensor S7 is, for example, a rotation angle around the pitch axis along the left-right direction of the human-powered vehicle 1. As an example, the tilt sensor S7 includes a sensor that detects the angular velocity of the pitch angle, and calculates a value obtained by integrating the angular velocity around the pitch axis as the pitch angle. The tilt sensor S7 may be configured to measure the rotation angle around the roll axis along the front-rear direction of the human-powered vehicle 1 and the rotation speed around the yaw axis along the vertical direction of the human-powered vehicle 1.

The image sensor S8 is a sensor that outputs a video signal obtained by photographing the surroundings of the human-powered vehicle 1. The image sensor S8 is provided at an appropriate position on the frame 10A or the handlebar 16 so that the road surface is included in the image pickup range, for example. The image sensor S8 may be a wearable camera mounted on the rider.

The position sensor S9 is a sensor that outputs a signal indicating the current position of the human-powered vehicle 1. The position sensor S9 is attached to the frame 10A or the handlebar 16. The position sensor S9 includes, for example, a GPS (Global Positioning System) communication device, and positions the current position of the human-powered vehicle 1 by receiving radio waves from GPS satellites. The position sensor S9 may be provided with a memory for sequentially recording the positioned position information, and may be configured to output information on the travel path on which the human-powered vehicle 1 has traveled based on the position information recorded in the memory.

The weather sensor S10 is a sensor that measures the weather around the human-powered vehicle 1 and outputs a signal according to the measurement result. The weather sensor S10 is attached to the frame 10A and measures at least one of the wind direction, the wind speed, the temperature of the outside air, and the humidity of the outside air. The meteorological sensor S10 communicates with an external meteorological server (not shown) and acquires at least one of the wind direction, wind speed, outside air temperature, and outside air humidity around the human-powered vehicle 1 from the meteorological server. It may be configured to be used.

The evaluation device 100 is a kind of computer. In one example, the evaluation device 100 is a dedicated terminal such as a cycle computer provided in the human-powered vehicle 1. In another example, the evaluation device 100 is a general-purpose terminal such as a smartphone, a tablet terminal, or a wearable terminal possessed by the rider of the human-powered vehicle 1.

The evaluation device 100 acquires travel information regarding the travel of the human-powered vehicle 1. The traveling information acquired by the evaluation device 100 includes speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device 30, history of operation input and component control results, regeneration amount, inclination, and road surface. Includes information about the situation, location, weather, and at least one of the vehicle information. The traveling information may include information on the traction of the human-powered vehicle 1. The evaluation device 100 can acquire travel information from sensors S1 to S10 or components mounted on the human-powered vehicle 1. The traveling information acquired by the evaluation device 100 may include not only the information directly obtained from the sensors S1 to S10 or the components but also the information derived by calculation, and the information stored in the storage unit 106 of the evaluation device 100. May include. For example, the traveling information stored in the storage unit 106 may include information on the braking distance with respect to the operation input to the brake operating device 36C, and may include vehicle information such as the frame size, the weight of the human-powered vehicle 1, and the tire diameter. Good.

The evaluation device 100 evaluates the preference of control for the components of the human-powered vehicle 1 by using the learning model 120 according to the travel information regarding the traveling of the human-powered vehicle 1. Here, the preference of control is information suggesting whether the state of the human-powered vehicle 1 is good or bad when the control for the component is executed. The evaluation device 100 can evaluate the preference of control based on, for example, the quality of traction of the human-powered vehicle 1. The evaluation device 100 may highly evaluate the control that improves the traction of the human-powered vehicle 1, and may underestimate the control that is estimated to reduce the traction. The learning model 120 used by the evaluation device 100 to evaluate the preference for control over the components will be described in detail with reference to FIG.

FIG. 2 is a block diagram showing an internal configuration of the evaluation device 100 according to the first embodiment. The evaluation device 100 includes an input unit 102, an arithmetic processing unit 104, a storage unit 106, and an output unit 108.

The input unit 102 is at least one of an operating device 30, a transmission 32, an assist device 34, a braking device 36, an anti-lock braking system (ABS) device 38, a suspension device 40, and sensors S1 to S10. It has an interface to connect to. The interface included in the input unit 102 is, for example, a wired interface, and the operation device 30 and the like are connected via a communication cable. Information based on the signal input through the input unit 102 is temporarily stored in the storage unit 106. The interface included in the input unit 102 may be a wireless interface. As a wireless interface, a communication interface conforming to a communication standard including Bluetooth (registered trademark), WiFi (registered trademark), ZigBee (registered trademark), LTE (Long Term Evolution), and other wireless LAN (Local Area Network). Can be used.

The arithmetic processing unit 104 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. The ROM included in the arithmetic processing unit 104 stores a computer program or the like for controlling the operation of each hardware unit included in the evaluation device 100. The CPU in the arithmetic processing unit 104 executes a computer program stored in the ROM or the storage unit 106 to control the operation of each hardware unit, and thereby according to the traveling information regarding the traveling of the human-powered vehicle 1, the human-powered vehicle 1 A process of evaluating the preference of control for the components of the above using the learning model 120 is realized. Data used during execution of the calculation is temporarily stored in the RAM included in the calculation processing unit 104.

The arithmetic processing unit 104 is configured to include a CPU, ROM, and RAM, but is a GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), quantum processor, volatile or non-volatile memory. It may be one or a plurality of arithmetic circuits including the above.

The storage unit 106 includes a memory such as an EEPROM (Electronically Erasable Programmable Read Only Memory) and a SRAM (Static Random Access Memory). The storage unit 106 stores various computer programs including the evaluation processing program 110 executed by the arithmetic processing unit 104, a learning model 120 and the like, which will be described later.

The evaluation processing program 110 is a computer program for causing a computer to execute a process of evaluating the preference of control for the components of the human-powered vehicle 1 by using the learning model 120 according to the traveling information regarding the traveling of the human-powered vehicle 1. Is. By executing the evaluation processing program 110, the arithmetic processing unit 104 evaluates the preference for control of the components of the human-powered vehicle 1 according to the traveling information of the human-powered vehicle 1 by using the learning model 120.

Various computer programs including the evaluation processing program 110 may be provided by the storage medium M in which the computer program is stored. The storage medium M is, for example, a portable memory such as a CD-ROM, a USB memory, an SD (Secure Digital) card, a micro SD card, and a compact flash (registered trademark). In the present embodiment, the storage medium M is a non-temporary storage medium that readablely stores various computer programs including the evaluation processing program 110. The arithmetic processing unit 104 reads various computer programs from the storage medium M using a reading device, and installs the read various computer programs in the storage unit 106.

The learning model 120 stored in the storage unit 106 is described by the definition information thereof. The definition information of the learning model 120 includes structural information of the learning model 120, various parameters such as weights and biases between nodes used in the learning model 120, and the like. The learning model 120 in the present embodiment is used to evaluate the preference of control for the components of the human-powered vehicle 1 according to the travel information regarding the traveling of the human-powered vehicle 1. The learning model 120 is, for example, a learning model constructed by using DQN (Deep Q-Network). The learning model 120 is not limited to DQN, and may be a learning model constructed using DDQN (Double DQN), CDQN (Continuous DQN), or the like.

The output unit 108 includes an output interface that outputs an evaluation result by the arithmetic processing unit 104. The output format of the evaluation result is arbitrary. In one example, the output unit 108 includes a display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display. In this case, the output unit 108 may display the evaluation result regarding the preference of control on the display device as character information or image information. In another example, the output unit 108 includes an audio output device such as a speaker. In this case, the output unit 108 may output the evaluation result regarding the preference of control as voice from the voice output device. In yet another example, the output unit 108 includes a communication interface. In this case, the output unit 108 may output the evaluation result regarding the preference of control to an external device connected to the communication interface.

FIG. 3 is a schematic diagram showing an implementation example of the learning model 120. The learning model 120 is, for example, a learning model constructed by DQN. In DQN, a multi-layer neural network is used in which the state s is input and the value of the action value function Q (s, a) corresponding to the selectable action a is output. In the following description, the value of the action value function Q (s, a) is also described as the Q value. The multi-layer neural network constituting the learning model 120 includes, for example, an input layer 122, an intermediate layer 124, and an output layer 126.

The input layer 122 includes a plurality of nodes. The number of nodes of the input layer 122 is prepared as many as the number of states to be input. In the present embodiment, the state s is travel information regarding the travel of the human-powered vehicle 1. The driving information includes speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device 30, history of operation input and component control result, regeneration amount, inclination, road surface condition, position, weather, And, it is a state distinguished by at least one of the vehicle information.

The intermediate layer 124 is composed of a plurality of layers. Each layer of the intermediate layer 124 includes a plurality of nodes. The nodes of each layer in the intermediate layer 124 are connected to the nodes existing in the preceding and following layers by a desired weight and bias in one direction. The weights and biases that characterize the connection state between nodes are updated by learning. The intermediate layer 124 calculates using each node and outputs the calculation result to the output layer 126.

The output layer 126 includes a plurality of nodes. The number of nodes of the output layer 126 is prepared as many as the number of actions a to be distinguished. In this embodiment, the action a is distinguished by the control state of one or more components included in the human-powered vehicle 1. For example, the action a is distinguished according to whether or not the brake device 36 of the human-powered vehicle 1 is operating. Further, the action a is distinguished according to whether the left or right brake device 36 is operated, the force or speed at which the brake operation device 36C is operated, or the like. The action a is not limited to the control state of the brake device 36, and may be distinguished by the control state of the assist device 34, the antilock brake device 38, or the suspension device 40.

Each node of the output layer 126 outputs a Q value for each action a. The Q value is an expected value of profits obtained in the future when the action a is taken in the state indicated by the state s. The learning model 120 is learned so that the Q value becomes high in the control state where the traction of the human-powered vehicle 1 is good, and the Q value becomes low in the control state where the traction of the human-powered vehicle 1 is lost. ..

Next, the evaluation procedure using the learning model 120 will be described.
FIG. 4 is a flowchart illustrating a learning procedure of the learning model 120. In step S101, the arithmetic processing unit 104 of the evaluation device 100 acquires the traveling information. Specifically, the arithmetic processing unit 104, through the input unit 102, speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device 30, history of operation input and component control result, Acquire driving information including at least one of regeneration amount, inclination, road surface condition, position, weather, and vehicle information. The traveling information acquired by the arithmetic processing unit 104 may be information related to the traction of the human-powered vehicle 1. The traveling information may be not only the information obtained by the various sensors S1 to S10 but also the information calculated by the arithmetic processing unit 104 or the information stored in the storage unit 106.

In step S102, the arithmetic processing unit 104 executes the arithmetic by the learning model 120 by inputting the traveling information acquired from the input unit 102 or the like into the input layer 122 of the learning model 120. The travel information input to the input layer 122 represents the state s in the learning model 120 constructed by DQN. The traveling information input to the input layer 122 is given to the intermediate layer 124. In the intermediate layer 124, each node performs an operation using weight multiplication, bias addition, and activation function on the input from the immediately preceding layer, and the operation result is output to the output layer 126. The output layer 126 outputs a Q value through each node.

In step S103, the arithmetic processing unit 104 acquires the Q value obtained as the arithmetic result of the learning model 120 from the output layer 126.

In step S104, the arithmetic processing unit 104 evaluates the preference for control of the components of the human-powered vehicle 1. The learning model 120 outputs a Q value for each of the actions a corresponding to the control state of the component from each node included in the output layer 126. The arithmetic processing unit 104 evaluates the preference of control for the component by referring to the Q value output from each node of the output layer 126. For example, the arithmetic processing unit 104 can evaluate the control state corresponding to the action a showing the highest value among the Q values output from each node of the output layer 126 as the most preferable control.

As described above, the evaluation device 100 in the first embodiment can evaluate the preference of control for the components of the human-powered vehicle 1 according to the travel information regarding the traveling of the human-powered vehicle 1 by using the learning model 120.

(Second Embodiment)
In the second embodiment, the control system that controls the components of the human-powered vehicle 1 will be described based on the evaluation result by the evaluation device 100.

FIG. 5 is a block diagram illustrating a control system according to the second embodiment. The control system according to the second embodiment includes an evaluation device 100 and a control device 200 that controls the components of the human-powered vehicle 1 based on the evaluation result by the evaluation device 100. The control device 200 may be provided as an independent device, or may be provided within a component to be controlled. In the example shown in FIG. 5, the evaluation device 100 and the control device 200 are described as separate bodies, but the evaluation device 100 and the control device 200 may be integrated.

The control device 200 includes a control unit 202, a storage unit 204, an input unit 206, and an output unit 208. The control unit 202 includes, for example, a CPU, a ROM, a RAM, and the like. The ROM included in the control unit 202 stores a control program or the like for controlling the operation of each hardware unit included in the control device 200. The CPU in the control unit 202 executes the control program stored in the ROM and various programs stored in the storage unit 204 to control the operation of each hardware unit, thereby realizing the control device of the present application. Specifically, the control unit 202 generates a control signal for controlling the components of the human-powered vehicle 1 based on the determination result from the evaluation device 100 input through the input unit 206, and outputs the generated control signal to the output unit 208. Output from.

The control unit 202 is not limited to the above configuration. The control unit 202 may be one or more control circuits including a single-core CPU, a multi-core CPU, an FPGA, a volatile or non-volatile memory, and the like. Further, the control unit 202 may have functions such as a clock for outputting date and time information, a timer for measuring the elapsed time from giving the measurement start instruction to giving the measurement end instruction, and a counter for counting the number.

The storage unit 204 includes memories such as EEPROM and SRAM. The storage unit 204 stores a computer program executed by the control unit 202, data used by the computer program, and the like.

The input unit 206 includes an interface for connecting the evaluation device 100 via a cable. The evaluation result output from the evaluation device 100 is input to the input unit 206. The evaluation result input to the input unit 206 represents the preference for control over the components of the human-powered vehicle 1. The input unit 206 outputs the input evaluation result to the control unit 202. In the present embodiment, the evaluation device 100 is connected via a cable, but data may be transmitted and received between the evaluation device 100 and the control device 200 via a wireless communication interface. .. In wireless communication, a wireless communication method conforming to communication standards such as Bluetooth, WiFi, ZigBee, LTE, and other wireless LANs can be used.

The output unit 208 includes an interface for connecting the components to be controlled via a cable. The component may include an assist device 34 having an electric drive unit 34A that outputs an auxiliary drive force. The component may also include a braking device 36 having an electric drive unit 36A. The component may also include an antilock brake device 38 having an electric drive unit 38A. The component may also include a suspension device 40 having an electric drive unit 40A. In the present embodiment, the control device 200 is connected to the assist device 34, the brake device 36, the antilock brake device 38, and the suspension device 40. The output unit 208 outputs the control signal output from the control unit 202 to the component to be controlled. In the present embodiment, the components of the human-powered vehicle 1 are connected via a cable, but data can be transmitted and received between the control device 200 and the components of the human-powered vehicle 1 via a wireless communication interface. It may be configured as. In wireless communication, a wireless communication method compliant with the above-mentioned various communication standards can be used.

Hereinafter, a control procedure in which the control device 200 controls the components of the human-powered vehicle 1 based on the evaluation result of the evaluation device 100 will be described.

FIG. 6 is a flowchart illustrating a procedure of processing executed by the control device 200. In step S201, the control unit 202 of the control device 200 acquires the evaluation result by the evaluation device 100. The evaluation result includes information on the behavior having the highest Q value.

In step S202, the control unit 202 generates a random number. The control unit 202 can generate a random number by using a known method. The random number generated by the control unit 202 is a random number of 0 or more and 1 or less.

In step S203, the control unit 202 compares the random number generated in step S202 with ε, and determines whether or not the random number is larger than ε. Here, ε is a parameter for determining whether the Q value should be used or the action should be randomly selected. ε is a value of 0 or more and 1 or less, may be a fixed value set in advance, or may be a variable value that decreases from 1 to 0 according to the elapsed time.

When the control unit 202 determines that the random number is larger than ε, the control unit 202 selects the action having the highest Q value in step S204.

When the control unit 202 determines that the random number is ε or less, the control unit 202 randomly selects an action in step S205.

In step S206, the control unit 202 controls the components of the human-powered vehicle 1 according to the action selected in step S204 or step S205. The component to be controlled may be selected according to the evaluation result of the evaluation device 100.

When the component to be controlled is the assist device 34, the control unit 202 outputs a control signal to the assist device 34. For example, when the assist level before control is the weak assist level and the assist level corresponding to the action selected in step S204 or step S205 is the strong assist level, the control unit 202 sets the assist level from the weak assist level to the strong assist level. The control signal for transitioning to the level is output to the assist device 34. When the assist level before control and the assist level corresponding to the action selected in step S204 or step S205 are the same, the control unit 202 does not control the assist device 34.

When the component to be controlled is the brake device 36, the control unit 202 outputs a control signal to the brake device 36. For example, when the control state before control corresponds to the first braking force and the action selected in step S204 or step S205 corresponds to the second braking force, the control unit 202 first controls the braking force of the braking device 36. A control signal for transitioning from power to the second braking force is output. When the braking force before control and the braking force corresponding to the action selected in step S204 or step S205 are the same, the control unit 202 does not control the braking device 36.

When the component to be controlled is the antilock brake device 38, the control unit 202 outputs a control signal to the antilock brake device 38. For example, when the control state before control corresponds to the first oil pressure and the action selected in step S204 or step S205 corresponds to the second oil pressure, the control unit 202 sets the oil pressure of the anti-lock brake device 38. A control signal for transitioning from the 1st hydraulic pressure to the 2nd hydraulic pressure is output. When the oil pressure before control and the oil pressure corresponding to the action selected in step S204 or step S205 are the same, the control unit 202 does not control the antilock braking device 38.

When the component to be controlled is the suspension device 40, the control unit 202 outputs a control signal to the suspension device 40. For example, when the control state before control shows the first buffering property and the action selected in step S204 or step S205 shows the second cushioning property, the control unit 202 sets the cushioning property of the suspension device 40 to the first. A control signal for transitioning from the cushioning property to the second buffering property is output. When the cushioning property before control and the buffering property corresponding to the action selected in step S204 or step S205 are the same, the control unit 202 does not control the suspension device 40.

In the present embodiment, the action is selected by the ε-greedy method that determines whether the Q value should be used or the action should be randomly selected using the parameter ε, but this is different from the ε-greedy method. Actions may be selected by an algorithm.

(Third Embodiment)
In the third embodiment, a method of generating the learning model 120 will be described.

The evaluation device 100 uses a computer to evaluate the preference of control for the components of the human-powered vehicle 1 according to the traveling information regarding the running of the human-powered vehicle 1, using a function, and evaluates the preference and the evaluation result of the preference. The learning model 120 is generated by updating the control result of the component according to the evaluation result and the function according to the control result. More specifically, the evaluation device 100 generates the learning model 120 by updating the behavioral value function Q (s, a) based on the following equation and improving the parameters that characterize the learning model 120.
_{Q (s t, a t)} ← Q (s t, a t) + α (r t + 1 + γmax a 'Q (s t + 1, a') - Q (s t, a t)) ... formula (1 )

The action value function Q (s, a) is the expected value of the profit obtained in the future when the action a is taken in the state s. r is the reward. The subscript t of the state s, the action a, and the reward r is a number indicating one step in the thinking process that repeats in time series. The subscript t is also called a trial number. If the state changes after the action is decided, the trial number is incremented. Thus, reward r _{t + 1} in the formula (1) may act a _t in state s _t is selected, a reward condition is obtained when it becomes s _{t + 1.} α is the learning rate and γ is the discount rate. The action a'is an action that maximizes the action value function Q (s _{t + 1} , at _{+ 1} ) among the actions a _{t + 1 that} can be taken in the state _{st + 1} . _{max a 'Q (s t +} 1, a') is a maximized action value function by the action a 'is selected.

FIG. 7 is a flowchart illustrating a procedure for generating the learning model 120. In step S301, the arithmetic processing unit 104 of the evaluation device 100 initializes the definition information of the learning model 120. The definition information of the learning model 120 includes structural information of the learning model 120, various parameters such as weights and biases between nodes used in the learning model 120, and the like. In the following, various parameters included in the definition information of the learning model 120 will be described as parameter θ for convenience.

In step S302, the arithmetic processing unit 104 of the evaluation device 100 acquires the traveling information. The traveling information is observed by sensors S1 to S10 and the like. The arithmetic processing unit 104 uses speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device 30, history of operation input and component control results, regeneration amount, inclination, and road surface as travel information. Get at least one of the status, location, weather, and vehicle information. The traveling information acquired by the arithmetic processing unit 104 may be information related to the traction of the human-powered vehicle 1. The traveling information may be not only the information observed by the various sensors S1 to S10 but also the information calculated by the arithmetic processing unit 104 or the information stored in advance in the storage unit 106.

In step S303, the calculation processing unit 104 inputs the traveling information acquired from the input unit 102 or the like to the input layer 122 of the learning model 120, and executes the calculation by the learning model 120 to calculate the Q value. The traveling information input to the input layer 122 is given to the intermediate layer 124. In the intermediate layer 124, each node performs an operation using weight multiplication, bias addition, and activation function on the input from the immediately preceding layer, and the operation result is output to the output layer 126. The output layer 126 outputs a Q value through each node. The arithmetic processing unit 104 outputs the calculated Q value of each action to the control device 200.

In step S304, the control unit 202 of the control device 200 selects an action to be taken by the human-powered vehicle 1. The control unit 202 uses, for example, the ε-greedy method to determine whether an action should be selected using the Q value or a random action.

In step S305, the control unit 202 controls the components of the human-powered vehicle 1 so that the human-powered vehicle 1 takes the selected action.

In step S306, the arithmetic processing unit 104 of the evaluation device 100 acquires the traveling information after the components are controlled. The travel information acquired by the arithmetic processing unit 104 is the same as the travel information acquired in step S302.

In step S307, the arithmetic processing unit 104 evaluates the reward. In the present embodiment, the reward is determined based on the running state of the human-powered vehicle 1. For example, if the traction of the human-powered vehicle 1 can be driven without being reduced, a positive reward is given, and if the traction of the human-powered vehicle 1 is reduced, or if the human-powered vehicle 1 falls, a negative reward is given. Is given. If the current trial number is t, the reward given in step S307 is r _{t + 1} in equation (1).

In step S308, the arithmetic processing unit 104 calculates the value of the objective function. In order to properly update the behavioral value function Q (s, a), it is necessary to optimize the learning model 120 that outputs the behavioral value function Q (s, a). At the stage where the learning is not completed, it is difficult to set the target teacher data because the behavioral value function Q (s, a) is not known. Therefore, in the present embodiment, the parameter θ that characterizes the learning model 120 is improved by the objective function that minimizes the TD (Temporal Difference) error shown in the second term of the equation (1).

In step S309, the arithmetic processing unit 104 determines whether or not the learning is completed. In the present embodiment, a threshold value for determining whether or not the TD error is sufficiently small is predetermined, and when the value of the objective function is equal to or less than the threshold value, the arithmetic processing unit 104 determines that learning has been completed. To do.

When the arithmetic processing unit 104 determines in step S309 that the learning has not been completed, the arithmetic processing unit 104 updates the action value function Q (s, a) in step S310. For example, the arithmetic processing unit 104 specifies the changing direction of the parameter θ that reduces the objective function based on the partial differential due to the parameter θ of the TD error, and changes the parameter θ. The specification of the change direction of the parameter θ is not limited to the partial differentiation based on the parameter θ of the TD error, and various gradient descent methods can be used. By the above processing, the parameter θ can be changed so that the action value function Q (s, a) approaches the target. After updating the action value function Q (s, a), the arithmetic processing unit 104 returns the processing to step S303.

When the arithmetic processing unit 104 determines that the learning is completed in step S309, the arithmetic processing unit 104 updates the definition information of the learning model 120 in step S311. The arithmetic processing unit 104 stores the θ obtained by learning in the storage unit 106 as definition information of the learning model 120 to be referred to when evaluating the preference of control of the human-powered vehicle 1.

As described above, in the present embodiment, the process of selecting the action a that maximizes the value of the action value function (s, a) and the process of controlling the human-powered vehicle 1 based on the selected action a are repeated. , The learning model 120 can be generated by optimizing the behavioral value function (s, a).

In the present embodiment, the evaluation device 100 of the human-powered vehicle 1 is configured to generate the learning model 120, but the external server may be configured to generate the learning model 120. In this case, a communication device (not shown) capable of communicating with the external server is connected to the evaluation device 100. Alternatively, a communication device capable of communicating with an external server is built in the evaluation device 100. The evaluation device 100 uploads the traveling information acquired from the sensors S1 to S10 and the like to an external server through the communication device. The external server selects the action a that maximizes the value of the action value function (s, a) based on the action value function Q (s, a), and notifies the human-powered vehicle 1 of the information of the selected action a. .. The control device 200 of the human-powered vehicle 1 controls the components based on the information of the action a selected by the external server. In this way, the processing of the external server that selects the action a that maximizes the value of the action value function (s, a) and the processing of the control device 200 that controls the human-powered vehicle 1 based on the selected action a are repeated. At the same time, the action value function (s, a) is optimized on the external server to generate the learning model 120.

(Fourth Embodiment)
In the fourth embodiment, a configuration for re-learning the learning model 120 will be described.

FIG. 8 is a flowchart illustrating a re-learning procedure of the learning model 120. In step S401, the arithmetic processing unit 104 of the evaluation device 100 reads out the definition information of the learning model 120 stored in the storage unit 106.

In step S402, the arithmetic processing unit 104 of the evaluation device 100 acquires the traveling information. The traveling information acquired by the arithmetic processing unit 104 is the same as that in the third embodiment.

In step S403, the arithmetic processing unit 104 inputs the traveling information acquired from the input unit 102 or the like to the input layer 122 of the learning model 120, and calculates the Q value by executing the arithmetic by the learning model 120.

In step S404, the control unit 202 of the control device 200 selects an action to be taken by the human-powered vehicle 1. The control unit 202 uses, for example, the ε-greedy method to determine whether an action should be selected using the Q value or a random action.

In step S405, the control unit 202 controls the components of the human-powered vehicle 1 so that the human-powered vehicle 1 takes the selected action.

In step S406, the arithmetic processing unit 104 of the evaluation device 100 acquires the traveling information after the components are controlled. The travel information acquired by the arithmetic processing unit 104 is the same as the travel information acquired in step S402.

In step S407, the arithmetic processing unit 104 evaluates the reward. In the present embodiment, the reward is determined based on the running state of the human-powered vehicle 1. For example, if the traction of the human-powered vehicle 1 can be driven without being reduced, a positive reward is given, and if the traction of the human-powered vehicle 1 is reduced, or if the human-powered vehicle 1 falls, a negative reward is given. Is given.

In step S408, the arithmetic processing unit 104 calculates the value of the objective function. In the present embodiment, the parameter θ that characterizes the learning model 120 is improved by the objective function that minimizes the TD error, which is shown in the second term of the equation (1).

In step S409, the arithmetic processing unit 104 determines whether or not the learning is completed. In the present embodiment, a threshold value for determining whether or not the TD error is sufficiently small is predetermined, and when the value of the objective function is equal to or less than the threshold value, the arithmetic processing unit 104 determines that learning has been completed. To do.

When the arithmetic processing unit 104 determines in step S409 that the learning has not been completed, the arithmetic processing unit 104 updates the action value function Q (s, a) in step S410. For example, the arithmetic processing unit 104 specifies the changing direction of the parameter θ that reduces the objective function based on the partial differential due to the parameter θ of the TD error, and changes the parameter θ. The specification of the change direction of the parameter θ is not limited to the partial differentiation based on the parameter θ of the TD error, and various gradient descent methods can be used. By the above processing, the parameter θ can be changed so that the action value function Q (s, a) approaches the target. After updating the action value function Q (s, a), the arithmetic processing unit 104 returns the processing to step S403.

When the arithmetic processing unit 104 determines that the learning is completed in step S409, the arithmetic processing unit 104 updates the definition information of the learning model 120 in step S411. The updated definition information of the learning model 120 is stored in the storage unit 106.

As described above, in the present embodiment, the evaluation device 100 relearns the learning model 120 according to the evaluation result of the preference and the control result of the component according to the evaluation result. That is, in the present embodiment, the action value is repeated while repeating the process of selecting the action a that maximizes the value of the action value function (s, a) and the process of controlling the human-powered vehicle 1 based on the selected action a. The learning model 120 can be retrained by optimizing the function (s, a).

(Fifth Embodiment)
In the fifth embodiment, a notification system for notifying the evaluation result of the evaluation device 100 will be described.

FIG. 9 is a block diagram illustrating a notification system according to a fifth embodiment. The notification system according to the fifth embodiment includes an evaluation device 100 and a notification device 300 that notifies the evaluation result by the evaluation device 100. The evaluation result notified by the notification device 300 may include information regarding the control state of the component. In the example shown in FIG. 9, the evaluation device 100 and the notification device 300 are described as separate bodies, but the evaluation device 100 and the notification device 300 may be integrated.

The notification device 300 is, for example, a display device such as a liquid crystal display or an organic EL display. The notification device 300 can notify the rider of information regarding the evaluation result by displaying character information or image information on the display device according to the evaluation result of the evaluation device 100.

The notification device 300 may be a light emitting element such as an LED. The notification device 300 can notify the rider of information regarding the evaluation result by turning on, blinking, or turning off the light emitting element according to the evaluation result of the evaluation device 100.

The notification device 300 may be an audio output device such as a speaker. The notification device 300 can notify the rider of information regarding the evaluation result by outputting voice from the voice output device according to the evaluation result of the evaluation device 100.

The notification device 300 may be a communication device. The notification device 300 can notify the rider of information regarding the evaluation result by transmitting character information or image information to an external communication device according to the evaluation result of the evaluation device 100. The external communication device is, for example, a smartphone, a tablet, or a wearable terminal carried by the rider.

The notification device 300 may be a vibration device such as a vibrator. The notification device 300 can notify the rider of information regarding the evaluation result by vibrating the vibrating device according to the evaluation result of the evaluation device 100.

The notification device 300 does not need to constantly notify the evaluation result of the evaluation device 100, and may notify information about the evaluation result under specific conditions. For example, the notification device 300 may notify information about the evaluation result when the deceleration of the human-powered vehicle 1 becomes lower than a predetermined value as compared with the operation input of the brake operation device 36C.

As described above, in the present embodiment, the evaluation result of the evaluation device 100 can be notified to the rider by the notification device 300.

The embodiments disclosed this time should be considered as exemplary in all respects and not restrictive. The scope of the present invention is indicated by the scope of claims, not the above-mentioned meaning, and is intended to include all modifications within the meaning and scope equivalent to the scope of claims.

1 ... Human-powered vehicle, 30 ... Operation device, 32, 33 ... Transmission device, 34 ... Assist device, 36 ... Brake device, 38 ... Anti-lock braking device, 40 ... Suspension device, 42 ... Battery unit, 100 ... Evaluation device, 102 ... Input unit, 104 ... Arithmetic processing unit, 106 ... Storage unit, 108 ... Output unit, 110 ... Evaluation processing program, 120 ... Learning model, 200 ... Control device, 202 ... Control unit, 204 ... Storage unit, 206 ... Input Unit, 208 ... Output unit, 300 ... Notification device

Claims (14)

An evaluation device that evaluates the preference of control over the components of the human-powered vehicle by using a learning model according to the traveling information regarding the running of the human-powered vehicle. The evaluation device according to claim 1, wherein the learning model is retrained according to the evaluation result of the preference and the control result of the component according to the evaluation result. The traveling information includes speed, acceleration, attitude, load, driving force, cadence, gear ratio, operation input to the operation device, history of the operation input and control result of the component, regeneration amount, inclination, road surface condition, position, The evaluation device according to claim 1 or 2, which includes information on the weather and at least one of the vehicle information. The evaluation device according to claim 1 or 2, wherein the traveling information of the human-powered vehicle includes information regarding traction of the human-powered vehicle. The evaluation device according to any one of claims 1 to 4,
A control system including a control device that controls the component based on the evaluation result of the evaluation device. The control system according to claim 5, wherein the component includes an assist device having an electric drive unit that outputs an auxiliary driving force. The control system according to claim 5 or 6, wherein the component includes a braking device having an electric drive unit. The control system according to any one of claims 5 to 7, wherein the component includes an antilock braking device including an electric drive unit. The control system according to any one of claims 5 to 8, wherein the component includes a suspension device including an electric drive unit. The evaluation device according to any one of claims 1 to 4,
A notification system including a notification device that notifies information regarding an evaluation result of the evaluation device. The notification system according to claim 10, wherein the information regarding the evaluation result includes information regarding the control state of the component. Using a computer
According to the driving information regarding the driving of the human-powered vehicle, the preference for control over the components of the human-powered vehicle is evaluated by using a function.
A method of generating a learning model, which generates a learning model by updating the function according to the evaluation result of the preference and the control result of the component according to the evaluation result. On the computer
A computer program for executing a process of evaluating the preference of control for a component of the human-powered vehicle by using a learning model according to driving information related to the running of the human-powered vehicle. A storage medium in which the computer program according to claim 13 is stored.

Images