JP4336776B2 - Skill training equipment - Google Patents

Description

  The present invention relates to a skill training apparatus for acquiring a predetermined skill.

2. Description of the Related Art Conventionally, various skill training apparatuses are known for trainers to train skills for the purpose of acquiring various skills such as sports and skills for performing surgery.
For example, while the trainee is performing a predetermined motion such as a swing of a golf club, a measurement unit that measures the exercise pattern by the trainee, an exercise pattern measured by the measurement unit, and a preset target exercise pattern There is known a skill training device that includes a stimulus applying unit that compares the above and according to the error, and applies a stimulus to the trainee for realizing a target exercise pattern. Inverted pendulum
Japanese Patent No. 2562786

  By the way, the movement pattern that the trainer performs for skill training with the conventional skill training device is set in advance because the optimal exercise pattern for the trainee varies depending on various factors such as the physique and personality of each trainer. In most cases, there is a slight difference from the target movement pattern. If this difference is large, it may be impossible for the trainee to perform an action that matches the target exercise pattern, and the trainer's body may be burdened.

  On the other hand, by registering multiple target exercise patterns of the skill training device and changing each of these target exercise patterns to correspond to differences in the trainer's physique, personality, etc., individual differences can be handled. It is possible to do so. However, there is a problem that it is difficult to determine which target exercise pattern is suitable for each trainer.

  On the other hand, as a conventional method for acquiring skills, there is known a method in which a teacher who is proficient in the skill to be acquired teaches the trainer by taking a gait and takes over the skill. Yes.

  With this method, the instructor senses the trainee's physique, personality, and skill level of the movement, and provides guidance suitable for the trainer. Can improve skills quickly. In addition, the training that is instructed by the gait by the instructor is accompanied by a force-tactile sensation, so that the effect is great when the instructor himself / herself needs to acquire independently like a skill.

  However, in the case of the method of instructing by an instructor, the number of excellent instructors is small, and the number of trainers that one instructor can instruct at a time is limited, so it is not possible to instruct many trainers. There is.

  The present invention has been made in view of these problems, and an object of the present invention is to provide a skill training apparatus capable of performing training as instructed by a leader when there is no leader.

  The skill training apparatus according to claim 1, wherein the motion detection means detects the movement of the trainee, and the calculation means detects the detection result by the motion detection means and preset calculation conditions. Based on the above, the first correction force for correcting the exerciser's movement to an appropriate movement is calculated, and the correction means generates the first correction force calculated by the calculation means and gives it to the trainee. Correct the movement of the trainee.

Further, when the instructor is directly instructing the trainee, the second corrective force detecting means detects the second corrective force actually applied to the trainee by the instructor and stops the operation of the corrective means. The storage means samples and stores the second correction force detected by the second correction force detection means and the detection result by the motion detection means. Then, when an update command is input from the outside, the second correction force is applied to the trainee by the update means based on the data stored in the storage means and the first correction force calculated by the calculation means. Update the calculation conditions so that

  As a result, according to the skill training apparatus of the present invention, the first correction close to the second correction force applied by the instructor so that the operation is suitable for the trainee by changing the calculation condition by the updating means. The force can be calculated by the calculation means and applied to the trainee by the correction means.

  Accordingly, the trainer can perform skill training with an operation suitable for the trainer by receiving the same first correction force as that being trained by the trainer even when there is no trainer. And since it becomes training accompanied by the force tactile sense of correction power, more effective training can be performed when a trainee like a skill needs to acquire the skill independently.

In addition, the skill training apparatus can reduce the time for which the instructor teaches one trainer, and can increase the number of trainers that the instructor can instruct.
In addition, the trainer can receive the guidance from the instructor again when the skill level increases to some extent, and change the calculation conditions by the update means at this time, so that the skill training according to the skill level of the trainee can be performed. .

  By the way, in the skill training apparatus according to claim 1, as the movement of the trainee detected by the motion detection means, various positions such as the position and speed of the trainee's motion part can be considered. When the trainee detects a movement of the trainee and applies a force to try to move, the first correction force is applied after the displacement due to this force is detected, resulting in a time lag and a sense of incongruity. Can be considered. For this reason, in order to bring the first correction force close to the second correction force by the instructor without a sense of incongruity, as described in claim 2, the motion detection means is one of the movements of the trainee. It is good to comprise so that the force generated when moving may be detected.

  That is, according to such a configuration, the force generated when the trainee moves is included in the detection result of the motion detection unit based on the calculation of the first correction force by the calculation unit. A first correction force corresponding to the generated force can be applied from the correction means. For this reason, there is no time lag in the first correction force applied to the trainee, and it is possible to approach the state where the instructor actually feels the operation force of the trainer and is applying the second correction force. In addition, when the trainee receives the first correction force, the trainee corrects the movement, so that the trainee can have bidirectionality between the force caused by the trainer's movement and the first correction force applied to the trainee. .

It should be noted that the force detected by the motion detection means here may be the trainee's force itself, or may be estimated from the acceleration of the trainee's movement.
However, the motion detection means does not detect or estimate the force generated when the trainee moves, but only detects the position and speed of the trainee's motion location as the trainee's motion, and the computing means detects this detection result. The first correction force may be calculated based on the above. In this way, the first correcting force can be applied by the correcting means when the trainee moves. Thereby, for example, the trainer's hand can be attached to the instructor's hand so that the trainer can trace the instructor's movement. Such training is effective when the trainee's skill level is low or when there is a habit of movement. In this way, it is preferable to be able to train by switching whether or not the operating force of the trainee is used as the basis for calculating the first correction force by the calculation means, depending on the personality and skill level of the trainee. .

  Further, the skill training apparatus according to claim 1 or claim 2 is for correcting the movement of the trainee when the trainee operates the operation target. In addition, the motion detection means may be configured to detect the motion state of the operation target in addition to the motion of the trainee.

That is, if comprised in this way, the operation state of an operation target object will be included in the input to a calculating means, and the appropriate 1st correction force according to the operation state of an operation target object can be added to a trainee.
For example, in a skill training device for training bat swing motion in baseball, when performing swing motion training when actually hitting a ball rather than simply swinging, the trajectory of the ball that is the operation target is By detecting, the first correction force can be applied so that the swing motion of the trainee becomes an appropriate motion according to the trajectory of the sphere.

  In addition, since the skill training apparatus of the present invention is intended for skill training, the operation object may not be an actual object, and the operation state of the operation object is detected from the actual operation object. In addition to the above, an operation state of a virtual operation target may be estimated.

  In addition, the skill training device according to claim 1 is a device for correcting the motion when the trainee operates the operation target, and replacing the motion detection means with the motion of the trainee. You may comprise so that the operation state of a target object may be detected.

  That is, if comprised in this way, a 1st correction force can be applied only according to a motion of an operation target object, without being influenced by a motion of a trainee. Thereby, the training which traces the movement used as the model by the leader with respect to the movement of the operation target can be performed.

  Moreover, there are various calculation methods by the calculation means in the skill training apparatus according to claims 1 to 4, for example, a calculation condition expressing a relationship between the exerciser's movement and the first correction force by a function or a conditional expression. The method to use can be considered. However, when inputting human movements, since the elements of multiple positions, speeds, accelerations, and forces are usually input items, and each of these input items is entangled with each other and affects the first correction force, The calculation conditions are often complicated and numerous, and the design of such calculation means and update means is difficult. For this reason, in order to make it easy to design the calculating means and the updating means, it is preferable to configure as in the fifth aspect.

  That is, in the skill training apparatus according to claim 5, the neural network that describes the relationship between the neuron and the connection load so that the result detected by the motion detection means is input and the first correction force is output. In accordance with the model, the calculation means calculates the first correction force, and the update means calculates the first correction force calculated by the calculation means based on the data stored in the storage means when an update command is input from the outside. The combined load is updated so that the first correction force approaches the second correction force applied to the trainee.

  As a result, according to the present invention (Claim 5), the calculation means can be easily designed. In other words, a neural network model (so-called neural network model) describes a combination of neurons in various combinations, and describes the state in which each input affects each other, the neuron to which each input is input, and other neurons. It can be easily described simply by joining them together. In addition, there are several well-known methods for updating, and it is sufficient to update based on this, and the design of the updating means can be facilitated.

  On the other hand, when performing skill training, the degree of improvement is better if there is a period during which training is carried out by yourself alone at some time rather than when the instructor is attached all the time and corrective power is added. It is known to be expensive.

  In order to implement this training method, the skill training apparatus according to any one of claims 1 to 5 can be switched according to the passage of time based on a preset schedule as described in claim 6. It is good to comprise so that operation stop and stop cancellation | release of a correction means may be switched.

  According to such a configuration, it is possible to increase the efficiency of training by setting the schedule such that a period in which the first correction force is applied by the correction means and a period in which the first correction force is not applied alternately occur.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 illustrates an overall configuration of the skill training apparatus according to the embodiment, and FIG. 2 is an explanatory diagram illustrating operations of the skill training apparatus according to the embodiment as functional blocks.

  The skill training apparatus according to the present embodiment is placed on a table that moves in one axial direction in a virtual space, and a bar-shaped virtual inverted pendulum that can move in one axial direction is placed in a direction in which the virtual inverted pendulum tends to fall. This is a device for the trainer P to train and acquire the skill that prevents the robot from falling down. In other words, it simulates the play of moving a hand so that the inverted pendulum does not fall down by placing one end of an elongated pendant inverted pendulum such as a pencil or a spear on the palm of the hand. It is a device for acquiring.

  As shown in FIG. 1, the skill training apparatus of the present embodiment is based on a screen 33 on which a virtual inverted pendulum 33a is projected, an arithmetic computer 10 that calculates the posture of the virtual inverted pendulum 33a, and the like, and the calculation result. A display computer 31 that generates an image signal, a projector 32 that projects an image based on the image signal onto a screen 33, a stator 28 disposed in front of the screen 33, and a slider 27 provided on the stator 28 , And a driver 29 for driving the slider 27.

  The slider 27 includes a handle 21 for the trainer P to operate the slider 27, a handle 22 for the instructor T to correct the operation of the trainer P, two force sensors 23 and 24, and a resolver. 25. The force sensors 23 and 24 and the resolver 25 are connected to the computing computer 10 via the A / D converter 41.

  The slider 27 is held by a stator 28 so as to be movable in the left-right direction in FIG. 1, and constitutes a direct drive linear motor with the stator 28, and moves on the stator 28 in the left-right direction according to a drive signal from a driver 29. A first correction force fc for moving to the position is generated.

  The driver 29 is configured to receive a command signal from the computing computer 10 and output a drive signal for causing the slider 27 to generate a first correction force fc corresponding to the signal.

The resolver 25 includes a sensor that detects a position in one axial direction, and detects the position x of the slider 27 with respect to the stator 28.
The force sensors 23 and 24 are composed of sensors that detect force by a force sensor made of an aluminum alloy and having a parallel plate structure. The force sensor 23 detects the force in the horizontal direction in FIG. 1 applied to the handle 21 and the force sensor 24 to the handle 22. That is, the force sensor 23 detects the operation force fl of the handle 21 by the trainee P, and the force sensor 24 detects the second correction force ft to the handle 22 by the instructor T.

  Further, the A / D converter 41 is configured to convert an analog signal into a digital signal, and outputs from each sensor provided on the slider 27 at every predetermined time and converted into digital data as a buffer memory. And output to the computing computer 10.

  Further, the display computer 31 is constituted by a general personal computer or the like, and receives the information on the posture of the virtual inverted pendulum 33a calculated by the calculation computer 10, and generates image data of the virtual inverted pendulum 33a.

The projector 32 converts the image data generated by the display computer 31 into an image and projects it on the screen 33.
The screen 33 is formed of a white plate having a predetermined transparency, and transmits the image projected from the rear by the projector 32 and displays it on the front surface.

  The computing computer 10 is composed of a general personal computer or the like, and as shown in FIG. 2, a virtual reality computing module (hereinafter referred to as Vr module) 12, a coach neural network module (hereinafter referred to as Nc module). 11), the schedule module 13, and the learning module 15 are formed by software.

  The Vr module 12 is configured to sequentially calculate a preset shape and the posture of the virtual inverted pendulum 33a having a mass, and differentiates the position x of the slider 27 input from the resolver 25 to obtain the speed of the slider 27. v is obtained, the posture of the virtual inverted pendulum 33a when the lower end of the virtual inverted pendulum 33a is changed to the position x and the speed v is calculated, and the inclination θ of the virtual inverted pendulum 33a with respect to the vertical direction and Obtain the angular velocity ω. Each state quantity obtained by the Vr module 12 is output to the Nc module 11 and the display computer 31.

  Further, as shown in FIG. 3, the Nc module 11 forms a neural network model in which the input layer includes eight units 51, the intermediate layer includes eight units 51, and the output layer includes one unit 51. The position x of the slider 27, the velocity v of the slider 27, the inclination θ of the virtual inverted pendulum 33a and the angular velocity ω calculated by the Vr module 12, and the operating force fl of the trainee P detected by the force sensor 23 The past velocity vo, the past angular velocity ωo, and the past stored in the input buffer provided in the Nc module 11 before a predetermined time (for example, 50 msec which is five sampling times of 10 msec at the A / D converter) The operating force flo is input to each unit 51 of the input layer.

  Then, the output from each of the eight units 51 of the input layer is input to each of the eight units 51 of the intermediate layer, and the output from each of the eight units 51 of the intermediate layer is input to the unit 51 of the output layer, A correction force signal fc obtained from the first correction force fc generated in the slide 27 is output as an instruction signal to the driver 29 as an output from the unit 51 of the output layer.

Further, the Nc module 11 inputs the second correction force ft of the instructor T detected by the force sensor 24 as a teaching signal.
In each unit 51, a coupling load 52 that determines a signal transmission rate is set for each input end. Each unit 51 obtains a value obtained by multiplying each input value by the value of the coupling load 52 and adds the value, and outputs a value obtained by a preset output function. A sigmoid function is used as the output function of the unit 51 in the intermediate layer, and a linear function is used as the output function of the unit 51 in the output layer.

  The past velocity vo and the past angular velocity ωo before a predetermined time (50 msec) input to the input layer are used to approximate the state in which the acceleration of the slider 27 and the angular acceleration of the virtual inverted pendulum 33a are input. Input to the input layer.

  The computing computer 10 is provided with an external storage device (not shown). And the computer 10 for a calculation can memorize | store the value set to each coupling load 52 in this external storage device, and the value of the coupling load 52 was memorize | stored for every trainer P who uses the said apparatus. You can switch to something.

  Further, the schedule module 13 is provided with a switch 14 that cuts off the transmission of the correction force signal fcs from the output layer of the Nc module 11 to the driver 29, and a time table for training by the trainee P. Then, the schedule module 13 switches between the intermittent correction force signal fcs based on this time table. In addition, the switch 14 can be switched between on and off by an operation input from the outside using a keyboard or the like of the computing computer 10.

  Further, when a learning command is input by an external operation using the keyboard of the computing computer 10 or the like, the learning module 15 performs processing (details will be described later) shown in the flowchart of FIG. The value of each combined load 52 is updated and learned so that the correction force signal fcs from the same is the same as the teaching signal.

The operation | movement at the time of performing the training which does not invert the virtual inverted pendulum 33a using such an inverted pendulum stabilization training apparatus 1 is demonstrated.
First, with the switch 14 of the computing computer 10 turned off by an external operation, the trainee P stands in front of the screen 33, operates the handle 21, and moves the slider 27. Then, based on the position x of the slider 27 detected by the resolver 25, the Vr module 12 calculates the speed v of the slider 27 and the posture of the virtual inverted pendulum 33a, and the virtual inverted pendulum displayed on the screen 33. The lower end portion of 33a is moved by the same amount as the movement amount of the slider 27.

  At this time, if no operation is performed on the handle 21, the virtual inverted pendulum 33a is tilted with the inclination θ increasing with the passage of time. When the trainee P moves the handle 21 in the tilt direction of the virtual inverted pendulum 33a, the tilt of the virtual inverted pendulum 33a decreases. Thus, by always moving the handle 21 in the tilt direction of the virtual inverted pendulum 33a, the virtual inverted pendulum 33a can be kept standing.

  However, the virtual inverted pendulum 33a tilts in the opposite direction if the amount of movement of the handle 21 is too large. Further, if the handle 21 is moved in the same direction as the inclination θ, the inclination θ increases and it tends to fall down. Thus, in order to stabilize the virtual inverted pendulum 33a, it is necessary for the trainee P to operate the handle 21 to the left and right with appropriate timing and force adjustment.

  When the switch 14 is connected by an external operation and the correction force signal fcs is input to the driver 29 for such an operation of only the trainer P, the operation force fl of the handle 21 by the trainer P and the virtual inversion The first correction force signal fcs calculated by the Nc module 11 according to the posture of the pendulum 33 a is output to the driver 29, and the first correction force fc is generated by the slider 27. Then, the slider 27 moves by the resultant force of the first correction force fc and the operation force fl by the trainee P.

  At this time, if the Nc module 11 is updated by the learning module 15 to be described later, even if the operation force fl by the trainee P is applied in an inappropriate direction, the force that corrects this is the first correction force. The state in which the virtual inverted pendulum 33a stands is maintained by the addition of fc.

  If it does in this way, the 1st correction force fc will also be transmitted to the trainer P via the handle | steering-wheel 21, and the trainer P can recognize his own operation mistake, and can know correct timing and force adjustment.

  In addition, if the trainer P performs an inappropriate operation while the trainer P is operating the handle 21 with the switch 14 turned off by an external operation, the trainer knows the timing and power of the handle operation. It is also possible for T to perform the training by applying the second correction force ft to the handle 22 and moving the slider 27 so that the virtual inverted pendulum 33a does not fall.

  Next, the update of the Nc module 11 by the learning module 15 performed when the correction force signal fcs is cut off by the switch 14 will be described with reference to the flowchart shown in FIG.

First, in S110, data to be input to the input layer of the Nc module 11 and data of the second correction force ft of the instructor T are acquired from the buffer memory of the A / D converter 41.
Note that the data acquired at this time is that the correction force signal fcs is cut off by the switch 14 and each of the virtual inverted pendulums 33a when the trainer P is trained under the guidance of the trainer T is not tilted. The sensor data may be data that is input during the training, or may be data that has been recorded in advance by a data recorder or the like and is reproduced.

  Next, in S120, the data acquired in S110 is input to the input layer of the Nc module 11. Then, this input signal is transmitted to the intermediate layer of the Nc module 11 according to the coupling load of each unit, and further the signal transmitted to the output layer is output.

Next, in S130, the difference between the output from the output layer of the Nc module 11 and the teacher signal is obtained.
Next, in S140, each of the coupling loads 52 from the output layer to the input layer is explained by the back-propagation method (for details, “Neural network computer learns from the brain and nerves” by Kazuyuki Aihara, published by Tokyo Denki University Press). Present), the difference obtained in S130 is updated to be “0”.

Next, in S150, data at another timing is input to the input layer, and the output of the output layer at this time is compared with the teacher signal.
Next, in S160, if the comparison result in S150 is smaller than a predetermined value that is considered to be sufficiently small, it is determined that it has converged, and if it is larger than this predetermined value, the process returns to S120. Repeat the process.

  In other words, by performing such processing, the condition that the second correction force ft is applied by the instructor T in various situations is learned by updating the combined load 52 using each input as a parameter. The first correction force fc generated by the output correction force signal fcs can be brought close to the second correction force ft applied by the instructor T.

  Incidentally, in this embodiment, a neural network that approximates the action of the instructor T by performing learning about 1000 times by the back-propagation method using 200 pieces of data when the virtual inverted pendulum 33a is kept inverted for 15 seconds. A model has been obtained.

  As described above, according to the skill training apparatus of this embodiment, the learning module 15 updates the combined load 52 of the Nc module 11 based on the data when the instructor T is instructing the trainer P. (In other words, the first correction force fc applied to the trainee P by the correction force signal fcs from the Nc module 11 is the same as the second correction force ft when the instructor T is instructing. it can. Thereby, even when there is no leader T, it is possible to perform the same training as in the state where the leader T is instructing.

  Since the operation force fl of the trainee P is included in the input to the Nc module 11, the magnitude of the first correction force fc changes according to the state of the operation force fl of the trainer P, and the instructor T This makes it possible to train with the feeling of being guided by a gait.

  Further, when the instructor T corrects the steering operation of the trainer P, the instructor T applies correction power in consideration of the individuality such as the physique and personality of the trainer P. The first correction force fc can be made in consideration of the individuality of the trainer P by making the learning.

  Further, the schedule module 13 cuts off the output from the Nc module 11 every predetermined period so that the first correction force fc is not applied, so that only the trainer P trains at some training time. There is a period for performing skills training, and when performing skill training, it is possible to increase the level of progress compared to the time when the leader corrects all the time.

  In addition, when the trainee P is recognized to have a certain level of skill improvement, the trainer P performs the retraining by the trainer T and updates the Nc module 11, so that the trainer P has improved the skill level. Correction force according to the skill level of P can be applied, and training suitable for the skill level of skill can be performed.

[Relationship with the present invention]
The force sensor 23, the resolver 25, and the Vr module 12 of the present embodiment correspond to the operation detection unit of the present invention, the Nc module 11 corresponds to the calculation unit, the slider 27 and the driver 29 correspond to the correction unit, and the force. The sensor 24 corresponds to second correction force detection means, the A / D converter 41 corresponds to storage means, the learning module 15 corresponds to update means, and the schedule module 13 corresponds to switching means.
[Modification]
As mentioned above, although embodiment of this invention was described, this invention is not limited to said specific embodiment, It can implement with a various form besides this.

  In the present embodiment, the apparatus for performing the training so as not to tilt the virtual inverted pendulum 33a whose force is applied only in one axial direction has been described. However, the present invention is not limited to this, and a plurality of correction forces are applied to a plurality of operation locations. The apparatus may be configured to add For example, a sensor and an actuator corresponding to the movement of each joint such as an arm and a shoulder of the trainee P and a sensor for detecting the second correction force ft by the instructor T are provided, and the sensor output of the trainer P is expressed as Nc. A skill training apparatus configured to input the input layer of the module 11, provide a plurality of units 51 in the output layer of the Nc module 11, and drive the actuator described above may be used. According to such a skill training apparatus, it is possible to perform a training that simulates a state in which a corrective force is applied by an instructor to a more complicated operation such as a baseball or golf swing training. In addition, a skill training device or the like may be used for training a surgical skill or the like by detecting the movement of the fingertip and applying a correction force to the fingertip.

  In the present embodiment, the operating force fl of the trainer P detected by the force sensor 23 is one of the inputs to the input layer of the Nc module 11, but the operating forces fl and flo are used as the input layer of the Nc module 11. It may be configured not to input to.

  In this way, the first correction force fc generated for the trainee P by the correction force signal fcs from the Nc module 11 can be generated regardless of the operating force fl of the trainer P. Accordingly, it is possible to perform training such that the trainer P traces the movement of the trainer T by adding the hand of the trainer P to the hand of the trainer T.

Such training is effective when the trainee is less skilled or has a habit.
Further, the position x of the slider 27, the speed v of the slider 27, and the past speed vo are not input as input to the input layer of the Nc module 11, and the inclination θ, the angular speed ω, and the past angular speed of the virtual inverted pendulum 33a are not input. It may be configured to input only the operation state related to the operation target such as ωo. In this way, it is possible to perform a training that does not affect the movement of the trainer P and traces the movement of the leader T operating the virtual inverted pendulum 33a as it is.

  Further, in this embodiment, because of the skill of manipulating the operation object called the virtual inverted pendulum 33a, information regarding the posture of the virtual inverted pendulum 33a is input to the input layer of the Nc module 11, but in baseball or the like The skill training apparatus in which there is no operation target object, such as training of simple swing movements or dance training, and information on the state of the operation target object is not input to the Nc module 11 may be used.

  In this embodiment, the Nc module 11 is configured to form a neural network model. However, the present invention is not limited to this. For example, the data when the instructor T is instructing the trainer P is analyzed. Then, a function and a conditional expression for obtaining the first correction force fc are set by the learning module 15 using the inclination θ, the angular velocity ω, the position x and the speed v of the handle 21 as variables, and the function and the conditional expression. Thus, the Nc module 11 may be configured to output the correction force signal fcs in the state of the handle 21 and the virtual inverted pendulum 33a.

  However, the functions and conditional expressions in this case are complicated and diverse, and the data analysis in the learning module 15 is not easy, and a large amount of labor is required for designing the Nc module 11. On the other hand, when a neural network model is formed as in the present embodiment, the neural network model basically describes various combinations of neuron connections, and states in which each input affects each other, Each input can be described easily by connecting the input neurons and other neurons. Moreover, the Nc module 11 can be easily learned by a well-known method using normal training data instructing the trainer P by the instructor T. Moreover, since the data at the time of actual training are reflected as they are, the first correction force fc close to the second correction force ft by the instructor P can be applied to the trainer P.

It is a figure showing the whole structure of the inverted pendulum stabilization training apparatus 1 of a present Example. It is a block diagram showing operation | movement of the inverted pendulum stabilization training apparatus 1 of a present Example. It is a figure showing the structure of the Nc module 11 of a present Example. It is a flowchart figure showing the process which learns the Nc module 11 of a present Example.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Inverted pendulum stabilization training apparatus, 10 ... Computer for calculation, 11 ... Neural network module for coaches, 12 ... Virtual reality calculation module, 13 ... Schedule module, 14 ... Switch, 15 ... Learning module, 21, 22 ... Handle, 23, 24 ... Force sensor, 25 ... Resolver, 27 ... Slider, 28 ... Stator, 29 ... Driver, 31 ... Display computer, 32 ... Projector, 33 ... Screen, 41 ... A / D converter, 51 ... Unit, 52 ... bond load.

Claims (6)

Motion detection means for detecting the movement of the trainee;
An arithmetic means for calculating a first correction force for correcting an exerciser's movement to an appropriate movement based on a detection result by the movement detection means and a preset calculation condition;
A correction means for correcting the movement of the trainee by generating the first correction force calculated by the calculation means and giving it to the trainer;
In a skill training device equipped with
A second corrective force detecting means for detecting a second corrective force actually applied by the instructor to the trainer when the instructor is directly instructing the trainer;
Storage means for sampling and storing the second correction force detected by the second correction force detection means and the detection result by the motion detection means;
When an update command is input from the outside, based on the data stored in the storage means, the first correction force calculated by the calculation means approaches the second correction force applied to the trainee. Updating means for updating the calculation condition;
Switch means for switching between operation stop and stop release of the correction means;
Equipped with a,
At the time of releasing the stop of the correction means by the switch means, the correction means is adapted to correct the movement of the trainee,
The storage means samples and stores the second correction force detected by the second correction force detection means and the detection result by the action detection means when the operation of the correction means by the switch means is stopped. A skill training device characterized by The motion detection means is
The skill training apparatus according to claim 1, wherein a force generated when the trainee moves is detected as one of the trainee's movements. The skill training device is for correcting the movement when the trainee operates the operation target,
The skill training apparatus according to claim 1, wherein the motion detection unit detects a motion state of an operation target in addition to a motion of a trainee. The skill training device is for correcting the movement when the trainee operates the operation target,
The skill training apparatus according to claim 1, wherein the motion detection unit detects a motion state of an operation target instead of a motion of a trainee. The calculation means takes the result detected by the motion detection means as an input, and in accordance with a neural network model describing the relationship between neurons and connection weights so that the first correction force becomes an output, Calculate the first correction force,
The update means is a second correction force applied to the trainee by the first correction force calculated by the calculation means based on the data stored in the storage means when an update command is input from the outside. Update the combined load to approach
The skill training apparatus according to any one of claims 1 to 4, wherein 6. The apparatus according to claim 1, further comprising switching means for switching operation stop and release cancellation of the correction means using the switch means based on a preset schedule. The described skill training device.

Images