JP5985870B2 - Shape calibration method - Google Patents

Description

  The present invention relates to a shape calibration method for calibrating a presentation shape of a tactile display that causes a surface undulation on CAD data to be perceived by operating a stimulator to give a stimulus to a person's palm.

Conventionally, CAD (Computer Aided Design) data is read, and the surface undulation on the CAD data is made to be perceived by a human by a tactile display that makes the surface undulation on the CAD data be simulated (see, for example, Patent Document 1). ).
As a case where such surface undulation is perceived, for example, there is a case where an operation of correcting the surface distortion of a mold is performed.

The tactile display disclosed in Patent Document 1 detects the movement of a tactile device including a plurality of stimulators that apply local stimuli to a finger joint using an image processing device. Then, in the virtual space, the movement of the haptic device is associated with the position of the virtual object model defined based on the CAD data.
The tactile display disclosed in Patent Document 1 applies a local stimulus to a finger joint of a person who moves a tactile device by operating a stimulator in accordance with the surface undulation of a virtual object model (CAD data), and virtual Make people perceive surface undulations on object models.

International Publication No. 2011/036787

Here, there is a difference between the perceived shape (or the actual shape) when a person perceives the presented shape of the tactile display and the presented shape (the distance by which the stimulator is operated according to the position of the stimulator). .
Thus, for example, if the tactile display presents a sine curve peak with height H and width L, a person may perceive that the surface relief is a sine curve peak with height H1 and width L1. (See the presentation shape W ′ and the perceptual shape W1 shown in FIG. 3).

  Therefore, when the stimulator is operated according to the surface undulation on the CAD data as in the tactile display disclosed in Patent Document 1, the surface undulation on the CAD data may not be accurately perceived. was there.

  The present invention has been made in view of the above situation, and provides a shape calibration method capable of accurately perceiving surface undulations on CAD data.

  The shape calibration method according to the present invention is a shape calibration method for calibrating a presentation shape of a tactile display that perceives surface undulation on CAD data by operating a stimulator to give a stimulus to a human palm, A tactile display allows a person to perceive a predetermined presentation shape on a trial basis, and a tactile sensation in which either one of the shape imagined by the person or the actual shape matches the perceived shape when the person perceives the presentation shape. When the obtained shape data corresponding to the presented shape and the shape data corresponding to one of the shape and real shape imaged by the person are used as learning data, learning of the neural network is performed by the tactile display, The shape data corresponding to the surface relief on the CAD data is input to the learned neural network and learned. The neural network outputs shape data corresponding to the presentation shape of the tactile display calibrated so that the surface undulation on the CAD data matches the perceived shape, and corresponds to the presentation shape of the calibrated tactile display The stimulator is operated according to the shape data to be performed.

  The shape calibration method according to the present invention uses mold CAD data as the CAD data.

  The shape calibration method according to the present invention is a shape calibration method for calibrating a presentation shape of a tactile display that perceives surface undulation on CAD data by operating a stimulator to give a stimulus to a human palm, A tactile display allows a person to perceive a predetermined presentation shape on a trial basis, and a tactile sensation in which either one of the shape imagined by the person or the actual shape matches the perceived shape when the person perceives the presentation shape. The neural network is learned by the tactile display using the temporal change of the stimulator as learning data when it is obtained, and the current output value of the stimulator and the current value of the stimulator before the current time An output value is input to the learned neural network, and from the learned neural network, surface undulation on the CAD data; To output a calibrated the output value of the current stimulator as the Satoshisa to shape matches the calibrated in accordance with the output value of the current stimulator, operating the stimulator is intended.

  The shape calibration method according to the present invention uses mold CAD data as the CAD data.

  The present invention has the effect that surface undulations on CAD data can be accurately perceived.

Explanatory drawing which shows the structure of a tactile display. Explanatory drawing which shows a stimulator. The figure which shows the difference between a perceptual shape and a presentation shape. Explanatory drawing which shows the neural network of 1st embodiment. Explanatory drawing which shows a mode that learning of a neural network is performed. The figure which shows the state which corrects a presentation shape. Explanatory drawing which shows the neural network of 2nd embodiment. Explanatory drawing which shows the state which corrects a presentation shape in 2nd embodiment.

  Below, the shape calibration method of 1st embodiment is demonstrated.

  The shape calibration method calibrates the presentation shape W ′ of the tactile display 1 (see FIGS. 1 and 6).

First, the configuration of the tactile display 1 will be described.
As shown in FIG. 1, the tactile display 1 reads CAD (Computer Aided Design) data and makes the surface undulation W on the CAD data be perceived in a pseudo manner. The tactile display 1 includes a liquid crystal screen 10, a tactile device 20, a dial switch 30, a foot pedal 40, a camera 50, a PC 60, and the like.

  The liquid crystal screen 10 is connected to the PC 60 and displays CAD data based on the output from the PC 60.

The tactile display 1 is used when correcting the surface distortion of the mold. That is, the CAD data is CAD data for designing the cavity surface of the mold.
However, the use of the tactile display 1 is not limited to this embodiment.

  The haptic device 20 is for giving local stimulation to the finger joint. The tactile device 20 is placed on the liquid crystal screen 10. As shown in FIG. 2, the haptic device 20 includes stimulators 21 to 29 and the like.

The stimulators 21 to 29 are provided on the upper surface of the haptic device 20. The stimulators 21 to 29 are arranged, for example, at positions facing the first to third joints of the index finger, middle finger, and ring finger, and are configured to be movable up and down by a piezo element.
As shown in FIGS. 1 and 2, the haptic device 20 is connected to the PC 60 and moves the stimulators 21 to 29 up and down based on the output from the PC 60.

The dial switch 30 is for setting a correction amount of the surface undulation W (see FIG. 3).
The skilled worker (person) operates the dial switch 30 to set the correction amount of the surface undulation W at the position corresponding to the cursor displayed on the liquid crystal screen 10.
The dial switch 30 is connected to the PC 60 and inputs the content operated by a person (the amount of correction of the surface undulation W) to the PC 60.

The foot pedal 40 is for moving the cursor. The skilled worker operates the foot pedal 40 to move the cursor to the apex of the surface undulation W (the most protruding portion in the surface undulation W), and selects the surface undulation W whose height is to be corrected.
The foot pedal 40 is connected to the PC 60 and inputs to the PC 60 the content operated by a person (position to move the cursor).

  The camera 50 detects the position of the haptic device 20. The camera 50 is connected to the PC 60 and inputs the detected position information of the haptic device 20 to the PC 60. A commercially available camera capable of measuring the position of the haptic device 20 is used for such a camera 50.

  Note that the tactile display 1 does not necessarily need to include the camera 50. That is, the position of the haptic device 20 may be specified by a sensor or the like that detects the amount of movement of the haptic device 20.

The PC 60 reads the CAD data and arranges the surface shape on the CAD data in the virtual space. The PC 60 acquires the position information of the haptic device 20 from the camera 50 and specifies the position of the haptic device 20. Then, the PC 60 moves the stimulators 21 to 29 up and down according to a position corresponding to the position of the haptic device 20 in the virtual space.
Further, the PC 60 corrects the height of the apex of the surface undulation W according to the operation of the dial switch 30. At this time, the PC 60 also corrects the height around the apex of the surface undulation W.

The skilled worker moves the haptic device 20 to trace the liquid crystal screen 10 with a palm placed on the upper surface of the haptic device 20.
At this time, the tactile display 1 presents the shape of the surface undulation W on the CAD data by moving the stimulators 21 to 29 (moving up and down) to give a stimulus to the palm of the skilled worker. Thereby, the tactile display 1 makes the skilled worker perceive the surface undulation W.

Here, as shown in FIG. 3, when the skilled worker perceives the presentation shape W ′ (hereinafter referred to as “perception shape W1”) and the presentation shape W ′ (positions of the stimulators 21 to 29) And the distance by which the stimulators 21 to 29 are moved up and down in accordance with the above.
For this reason, for example, when the same shape as the surface undulation W which is a peak of a sine curve of height H and width L is presented (when the surface undulation W and the presentation shape W ′ are the same shape), the skilled worker The surface undulation W may be perceived as a sine curve peak having a height H1 and a width L1.

  Therefore, the tactile display 1 calibrates the presentation shape W ′ using the shape calibration method (see FIG. 6).

  Note that the configuration of the tactile display 1 is not limited to this embodiment, and any configuration that can read CAD data and cause a skilled worker to perceive the surface undulation W on the CAD data in a pseudo manner. Good.

Next, the shape calibration method will be described.
In the following, for the sake of convenience of explanation, the tactile display 1 is assumed to cause a skilled worker to perceive a surface undulation W that is a peak of a sine curve having a height H and a width L as shown in FIG.

  In the shape calibration method, as shown in FIGS. 4 and 6, the presentation shape W ′ is calibrated using a neural network 60.

  The neural network 60 is an information processing model simulating the mechanism of a human brain neural circuit. The neural network 60 of the first embodiment is configured as a multilayer perceptron including an input layer 61, an intermediate layer 62 (hidden layer), and an output layer 63. Each layer 61-63 has a plurality of units.

In the unit of the input layer 61, shape data corresponding to the surface undulation W on the CAD data is input. When the surface undulation W on the CAD data is a sine curve peak as in the first embodiment, the height H and the width L of the surface undulation W on the CAD data are input layers as the shape data H · L. 61 is input.
The unit of the input layer 61 is combined with the unit of the intermediate layer 62, and the input value input to the input layer 61 is propagated to the unit of the intermediate layer 62.

  The unit of the intermediate layer 62 is coupled to the unit of the output layer 63, performs a predetermined arithmetic process on the input value propagated from the unit of the input layer 61, and propagates the output value to the unit of the output layer 63.

The unit of the output layer 63 performs a predetermined calculation process on the calculation result propagated from the unit of the intermediate layer 62, and calculates the output value.
The neural network 60 outputs the output value of the unit of the output layer 63 as the output value H ′ / L ′ of the neural network 60.

Such an i-th unit of the intermediate layer 62 and the output layer 63 calculates a variable S _{i according} to Equation 1 shown below.

In Equation 1, b _i is the bias of the i-th unit of the intermediate layer 62 and the output layer 63.
In Equation 1, w _{i, j} is a weight set for the input value. More specifically, when the i-th unit is a unit in the intermediate layer 62, the weight is set for the output value of the j-th unit in the input layer 61, and the i-th unit is a unit in the output layer 63. Is the weight set for the output value of the j-th unit of the intermediate layer 62.
In Equation 1, u _j is an output value of the j-th unit. More specifically, when the i-th unit is a unit of the intermediate layer 62, it is the output value of the j-th unit of the input layer 61, and when the i-th unit is the unit of the output layer 63, the intermediate layer 62 Is the output value of the j-th unit.

  The unit of the intermediate layer 62 and the output layer 63 calculates an output value based on the result of the mathematical formula 1 (see the mathematical formula 2 shown below). At this time, the unit of the intermediate layer 62 and the output layer 63 calculates the output value using, for example, a sigmoid function.

  In the shape calibration method of the first embodiment, learning of the neural network 60 is performed as follows in order to output appropriate values for the shape data H and L input to the input layer 61.

  First, as shown in FIG. 5, the skilled worker images a surface shape (for example, a peak of a sine curve having a height H2 and a width L2) in the head (see the image shape W2 shown in FIG. 5). The skilled worker moves the haptic device 20 (see FIG. 1).

At this time, in the shape calibration method, the skilled worker perceives the predetermined presentation shape W ′ on the tactile display 1 on a trial basis. At this time, the skilled worker confirms whether the image shape W2 and the perceived shape W1 match.
The skilled worker operates the dial switch 30 to change the presentation shape W ′ until a tactile sensation that matches the image shape W2 is obtained (until the image shape W2 and the perceptual shape W1 match).

  As a result, the shape calibration method uses the shape data H2 ′ / L2 ′ corresponding to the presentation shape W ′ and the shape data corresponding to the image shape W2 when a tactile sensation in which the image shape W2 and the perceptual shape W1 match is obtained. H2 and L2 are acquired.

In the shape calibration method, the neural network 60 is learned by the PC 60 (tactile display 1) using the acquired shape data H2, L2, H2 ′, and L2 ′ as learning data.
At this time, the shape calibration method performs learning of the neural network 60 so that an appropriate value based on the learning data can be output using, for example, a gradient method.

  Note that the shape calibration method of the first embodiment may acquire shape data corresponding to the presentation shape W ′ when the tactile sensation when the actual shape is touched and the perceived shape W1 match.

  The shape calibration method repeatedly performs such learning while changing the image shape W2. In the shape calibration method, such a learned neural network 60 is constructed in the PC 60.

  Next, the procedure of the shape calibration method of the first embodiment will be described.

  First, as shown in FIG. 4, when reading the CAD data, the tactile display 1 inputs the shape data HL corresponding to the surface undulation W on the CAD data to the learned neural network 60.

  As shown in FIGS. 4 and 6, the neural network 60 uses the shape data H ′ corresponding to the presentation shape W ′ of the tactile display 1 calibrated so that the surface undulation W on the CAD data matches the perceived shape W1.・ L 'is output.

When the skilled worker moves the tactile device 20, the shape calibration method operates the stimulators 21 to 29 in accordance with the shape data H ′ and L ′ corresponding to the calibrated presentation shape W ′ of the tactile display 1.
That is, the tactile display 1 moves the stimulators 21 to 29 up and down according to the sine curve peaks of the height H ′ and the width L ′ (see the presentation shape W ′ shown in FIG. 6).

According to this, the shape calibration method can bring the perceived shape W1 closer to the shape of the surface undulation W.
Therefore, the shape calibration method can cause a skilled worker to accurately perceive the surface undulation W on the CAD data.

  For this reason, when the mold CAD data is used as the CAD data as in the first embodiment, the surface undulation W can be accurately corrected when the surface distortion of the mold is corrected.

  As described above, the shape calibration method of the first embodiment is a static calibration method based on the shape data HL corresponding to the surface undulation W on the CAD data.

The shape data input to the neural network 60 is not limited to the first embodiment.
That is, if the surface undulation W is not a sine curve peak, for example, the height to the apex, the range of the surface undulation W, the sharpness of the surface undulation W, and the direction of the surface undulation W (which position of the surface undulation W The shape data is such that the apex is located in

Next, the shape calibration method of the second embodiment will be described.
In the following description, the tactile display 1 is assumed to allow a skilled worker to perceive a surface undulation W that is a peak of a sine curve having a height H and a width L as shown in FIG.

As shown in FIGS. 7 and 8, the shape calibration method of the second embodiment outputs the current output values H ₁ (k) to H ₉ (k) of the stimulators 21 to 29 using the neural network 160. This is different from the shape calibration method of the first embodiment.

In the output values H _n (k) of the stimulators 21 to 29, n corresponds to the first digit of the stimulators 21 to 29.
The output values H ₁ (k) to H ₉ (k) of the stimulators 21 to 29 are, for example, the heights of the stimulators 21 to 29, the applied voltage values to the piezoelectric elements, and the like.

  The neural network 160 is configured as a multilayer perceptron including an input layer 161, an intermediate layer 162, and an output layer 163.

The unit of the input layer 161, an output value H ₁ to ₉ of the current stimulator 21-29 (k), and an output value H ₁ of stimulator 21 to 29 at two time points before the current _1-9 ( k-1) · H ₁ to ₉ (k-2) are input.

In the output values H _n (k−1) · H _n (k−2), n corresponds to the first digit of the stimulators 21 to 29. In addition, the output values of the stimulators 21 to 29 at a time point k-1 is a predetermined time before the present time, and the output values of the stimulators 21 to 29 at a time point k-2 is a predetermined time before the k-1 time. The output value is shown.

The neural network 160 performs arithmetic processing as in the first embodiment in the unit of the intermediate layer 162 and the output layer 163, and outputs the current output values H ₁ to ₉ (k) of the stimulators 21 to 29.

  In the shape calibration method according to the second embodiment, the neural network 160 is learned as follows.

  First, the skilled worker operates the dial switch 30 to change the presentation shape W ′ until the tactile sensation that matches the image shape W2 is obtained in the same manner as in the first embodiment (see FIG. 5).

  Thereby, the shape calibration method acquires temporal changes of the stimulators 21 to 29 when a tactile sensation in which the image shape W2 and the perceptual shape W1 coincide with each other is obtained.

Specifically, in the shape calibration method, the output values H ₁ to ₉ (k) of the current stimulators 21 to 29 at a certain time point and the outputs of the stimulators 21 to 29 at two time points before the certain time point are used. Values H ₁ to ₉ (k−1) · H ₁ to ₉ (k−2) are acquired (see FIG. 8).
The shape calibration method uses such output values H ₁ to ₉ (k), H ₁ to ₉ (k-1), and H ₁ to ₉ (k-2) of the stimulators 21 to 29 at predetermined time intervals. To get to.

In the second embodiment, the output values H ₁ to ₉ (k) · H ₁ to ₉ (k-1) · H ₁ to ₉ (k-2) of the stimulators 21 to 29 acquired at such time intervals are obtained. , The time change of the stimulators 21-29.

In addition, the content of the temporal change of the stimulators 21 to 29 to be acquired is not limited to the second embodiment.
For example, the current output values of the stimulators 21 to 29 at a certain time point and the output values of the stimulators 21 to 29 at one time point before the certain time point may be acquired at predetermined time intervals. .
Further, the current output values of the stimulators 21 to 29 at a certain time point and the output values of the stimulators 21 to 29 at three or more time points before the certain time point may be acquired at predetermined time intervals. I do not care.

In the shape calibration method, the neural network 160 is learned by the PC 60 (tactile display 1) using the temporal changes of the stimulators 21 to 29 as learning data when the tactile sensation in which the image shape W2 and the perceptual shape W1 match is obtained. Do.
At this time, the shape calibration method performs learning of the neural network 160 so that an appropriate value based on the learning data can be output using, for example, a gradient method.

  Note that the shape calibration method of the second embodiment may acquire temporal changes of the stimulators 21 to 29 when the tactile sensation when the actual shape is touched and the perceived shape W1 match.

  Next, the procedure of the shape calibration method of the second embodiment will be described.

As shown in FIG. 7 and FIG. 8, when the skilled worker moves the haptic device 20, the shape calibration method is performed with the current output values H ₁ to ₉ (k) of the stimulators 21 to 29 and before the present. The output values H ₁ to ₉ (k−1) · H ₁ to ₉ (k−2) of the stimulators 21 to 29 at the time of are input to the learned neural network 160.

The neural network 160 outputs the output values H ₁ to ₉ (k) of the current stimulators 21 to 29 calibrated so that the surface undulation W on the CAD data matches the perceived shape W1.

In the shape calibration method, the stimulators 21 to 29 are operated in accordance with the calibrated output values H ₁ to ₉ ′ (k) of the current stimulators 21 to 29. Thereby, the shape calibration method calibrates the presentation shape W ′.

According to this, the shape calibration method can bring the perceived shape W1 closer to the shape of the surface undulation W.
Therefore, the shape calibration method can make the skilled worker accurately perceive the surface undulation W on the CAD data.

  For this reason, the shape calibration method can correct the surface undulation W accurately when correcting the surface distortion of the mold when the CAD data of the mold is used as the CAD data.

  Here, when the stimulators 21 to 29 are moved up and down in accordance with the predetermined presentation shape W ′, the perceptual shape W1 depends on how the skilled worker touches, that is, the speed and direction in which the tactile device 20 is moved. May change.

In the shape calibration method according to the second embodiment, when the touch changes, the values input to the neural network 160 (the output values H ₁ to ₉ (k) of the current stimulators 21 to 29, and the current values before the present). The output values H ₁ to ₉ (k−1) · H ₁ to ₉ (k−2)) of the stimulators 21 to 29 at the time change.

As described above, the neural network 160 calibrates the output values H ₁ to ₉ (k) of the current stimulators 21 to 29 so that the surface undulation W of the CAD data matches the perceived shape W1, The calibration results H ₁ to ₉ ′ (k) are output.

  That is, in the shape calibration method of the second embodiment, the perceived shape W1 is brought close to the shape of the surface undulation W by calibrating the presentation shape W ′ according to how the skilled worker touches.

  Thus, the shape calibration method of the second embodiment is a dynamic calibration method based on temporal changes of the stimulators 21 to 29.

  The application of the shape calibration method is not limited to this embodiment. That is, the shape calibration method allows the neural network 160 to learn the shape data corresponding to the surface undulation W on the CAD data and the correction contents of the surface distortion of the mold as learning data, thereby correcting the correction of the surface distortion of the mold. Is also possible.

  The shape calibration method does not necessarily need to be used when a skilled worker perceives the surface undulation W. That is, the shape calibration method may be used when an inexperienced worker is made to perceive the surface undulation W. In this case, the neural network 160 calibrates the presentation shape W ′ so that the inexperienced worker's perceived shape W1 matches the surface undulation W shape.

1 Tactile display 21 to 29 Stimulator 60/160 Neural network H Surface relief height (shape data corresponding to surface relief)
H 'Height of the presentation shape (shape data corresponding to the presentation shape)
L Surface relief width (shape data corresponding to surface relief)
L 'width of the presentation shape (shape data corresponding to the presentation shape)
W Surface relief W 'Presentation shape W1 Perception shape W2 Image shape (shape imaged by a person)

Claims (4)

A shape calibration method for calibrating the presentation shape of a tactile display that perceives surface undulations on CAD data by operating a stimulator to stimulate a person's palm,
The tactile display allows a person to experimentally perceive a predetermined presentation shape,
Shape data corresponding to the presentation shape when one of the shape imaged by the person and the actual shape and a tactile sensation when the person perceives the presentation shape are matched, and Using the shape data corresponding to one of the shape imaged by a person and the actual shape as learning data, learning the neural network with the tactile display,
Input shape data corresponding to the surface relief on the CAD data into the learned neural network,
From the learned neural network, output the shape data corresponding to the presentation shape of the tactile display calibrated so that the surface undulation on the CAD data and the perceived shape match,
Operating the stimulator according to shape data corresponding to the calibrated presentation shape of the tactile display;
Shape calibration method. As the CAD data, CAD data of a mold is used.
The shape calibration method according to claim 1. A shape calibration method for calibrating the presentation shape of a tactile display that perceives surface undulations on CAD data by operating a stimulator to stimulate a person's palm,
The tactile display allows a person to experimentally perceive a predetermined presentation shape,
As a learning data, a temporal change of the stimulator when a tactile sensation in which either one of the shape imaged by the person or an actual shape matches the perceived shape when the person perceives the presentation shape is obtained. , Learn the neural network with the tactile display,
Inputting the current output value of the stimulator and the output value of the stimulator at a time point before the current time into the learned neural network;
From the learned neural network, output the output value of the current stimulator calibrated so that the surface undulation on the CAD data matches the perceived shape,
Operating the stimulator according to the calibrated current output value of the stimulator;
Shape calibration method. As the CAD data, CAD data of a mold is used.
The shape calibration method according to claim 3.

Images