Hi, my name is Dan Goldrich and we're here in the tactile research lab at your master university. This is Ingrid Cans, Brian Peters and Mike Wong, and we're going to talk to you today about the tactile automated passive finger stimulator or taps. TAPS is a device for measuring passive tactile spatial acuity on the fingertip, and I developed it in Duquesne University in Pittsburgh, Pennsylvania with Ingrid cans, and we've since modified it in McMaster University here in Hamilton, Ontario with Ryan and Mike.
And this device is a method for delivering controlled tactile stimuli to the fingertip controlling for force onset velocity and stimulus duration. This video is organized into three sections. First, we provide an overview of taps functioning.
Next, we show taps in action and describe its components so that others may replicate the device. Finally, we illustrate one of many testing protocols that taps can conduct a two interval force choice grading orientation experiment with baying. Adaptive tracking taps uses gravity to press a stimulus surface against the skin.
Gravity exerts a downward force on a mass hanging from a rod. This side view shows taps with several structures removed in order to clearly illustrate the functioning of the device, the participant's arm rests comfortably in prone position on a tabletop. The fingertip lies over a tunnel in the table.
The tunnel is not shown in this diagram on a lower table hidden from the participant's view, a stepper motor rotates a disc that holds up to 40 stimulus pieces, positioning one of the pieces under the finger. A crossbar shown here in red extends from the movable carriage of a linear actuator preventing the rod from pivoting. As the actuator motors forward, the rod pivots under the influence of gravity pressing a stimulus piece upward through the tunnel and onto the skin.
Simple physics relates the speed of the actuator to the speed with which the stimulus surface rises to contact the skin and the weight of the hanging mass to the static force of the tactile stimulus. As the actuator reverses direction to return the rod to its starting position, the stimulus piece drops the disc stepper motor then rotates to position another piece under the finger and the entire process repeats. This top view with the finger removed shows another stimulus cycle.
Note the stimulus piece rising as the actuator motors forward and falling as the actuator reverses. To make the stimulus pieces, we started with half inch diameter, three inch long delrin rods on a shoreline mini milling machine. We used a fly cutter to plane, both ends reducing the rod length to 2.75 inches.
We use the fly cutter to shape two inch long rectangular shafts, 8.7 millimeters square. We cut parallel grooves into the round face to create square wave gradings with equal groove and ridge widths to cut larger grooves. We used an end mill for smaller grooves.
We used slitting sauce. We placed silicon O-rings around the shaft to reduce vibration of pieces As the disc spins, The mechanical components of taps are shown here. When the carriage of the linear actuator moves forward, the hanging mass is allowed to fall under the influence of gravity.
The weight of this mass and its distance from the pivot point along the rotating rod together determine the upward force with which the stimulus piece contacts the subject's skin. A particular piece is positioned directly under a tunnel in the upper table by a computer controlled stepper motor that spins the stimulus tray. After the subject pushes a response button, the computer commands the stepper motor to rotate the stimulus tray, positioning another stimulus piece directly under the tunnel.
In the upper table. The movement of taps is components are monitored by several sensors. A retroreflective sensor detects the presence of an elevated stimulus piece.
Note the green light going on and off. This sensor is used by taps during the initialization of the experiment to properly position the rotating rod at a specified distance below the stimulus pieces. In addition, activation of the sensor disables the step motor preventing equipment damage that would otherwise occur.
Should a stimulus piece accidentally jam in the tunnel. A U-shaped photoelectric sensor establishes the home position for the stimulus tray. This sensor is activated when a piece of stiff paper glued to the rim of the rotating tray breaks the infrared sensor beating.
All movements of the stimulus tray are subsequently calculated relative to this home position. A HU effect sensor sets the linear actuator home position. Here we can see the actuator's carriage moving forward with the pivot table will move when the carriage moves into home position.
The sensor activates note the red light going on and off as a safety feature. Two similar sensors. One of them shown here, set limits to forward and backward movement of the carriage in the unlikely event of actuator malfunction.
These limit sensors disable the actuator preventing runaway carriage movement. With the pivot table imposition and carefully aligned the crossbar that extends from the actuator's carriage contacts a vertical bar that extends down from the rotating rod at the pivot point. A nearly frictionless metal bearing serves as the pivot point and allows the bar to freely rotate.
When the actuator's carriage moves forward, the masses inserted into a Velcro basket which hangs from the rotating rod. Once the subject's arm is in prone position on the tabletop place, the distal pad of the finger to be tested over a tunnel in the table through which the stimulus surfaces will rise. To contact the skin to prevent lateral finger movement, gently place two barriers coated on the bottom with double-sided foam tape against the sides of the finger.
With the aid of a swivel position, the force sensor gently against the fingernail near the cuticle at 90 degrees to the plane of the fingernail ensure that the sensor exerts a force between 50 and 80 grams. This force is comfortable for the subject and provides the sensor with excellent sensitivity to finger movement. Fasten the swivel taps is a versatile system that can be used to apply any stimulus surfaces cut into the faces of half inch diameter rods.
For demonstration purposes, we will show you how a two interval forced choice grading orientation task is administered In our laboratory, our grading orientation task uses 20 stimulus surfaces with GR widths that range from as small as 0.25 millimeters to as large as 3.1 millimeters. Each trial consists of two sequential stimulus presentations with gradings of identical groove width, but differing 90 degrees in orientation. These are applied to the distal pad of the finger.
In one presentation, the grooves are aligned vertically or parallel to the long axis of the finger, and in the other the grooves are aligned horizontally or perpendicular to the long axis of the finger. The presentation order is chosen randomly. The subject indicates whether the horizontal orientation occurred in the first or second interval by pressing one of two buttons with the non-tested hand to indicate that horizontal was presented first, the subject would press the button on the left and to indicate that horizontal was presented.
Second, the subject would press the button on the right. Now let's take a look at what a typical subject might experience in a testing session. I want to direct your attention to the two gradings shown in the top left in this trial.
The subject was tapped on the finger first with the gradings aligned horizontally and second with the gradings aligned vertically. In this case, the correct response is to press the button on the left. In this example, the subject presses the button on the right, an incorrect response.
This is shown in the performance plot in the top right corner by a red X.The next groove width to be presented is selected by Bayesian adaptive algorithm, which we will discuss in more detail in the next segment of the video. This trial should be easier for the subject to perform. Since the groove width selected by the algorithm is larger than the groove width of the previous trial.
The subject is again presented with two gradings. In this case, the first grading is aligned vertically and the second horizontally. This time, the subject makes a correct response By pressing the button on the right, this is indicated by Blue Cross in the performance plot.
In the third trial, the group width has decreased slightly, making the task more difficult. Again, the subject makes a correct response. This sequence of events continues until the designated number of trials have been presented.
We typically run blocks of 40 trials To estimate each participant's psychometric function. We use a modified version of cevi and Tyler's Bayesian adaptive SI method. In psychophysics, a psychometric function relates a physical stimulus parameter groove width in our case to the probability of correct response.
Following Conserv and Tyler, we model D prime as the power law function of groove width. This lead to a cumulative normal psychometric function defined by three parameters, a, B and delta A, or the threshold parameter is the groove with whose orientation the subject can correctly discriminate 76%of the time, which corresponds to a DPR of one. On this task.
Here you can see that as I increase the value of the threshold, the curve shifts to the right. As I decrease the threshold, the curve shifts to the left. Subjects who are better at the task can discriminate the orientation of thinner grooves, so the curves are shifted to the left.
Notice that although the curve shifts along the X dimension, as the threshold value has changed, the slope of the curve remains the same. The slope of the curve is set by the B parameter. It tells us how fast the curve is rising.
The curve is shallow at low slope values, but you can see that the curve becomes steeper as I increase the value of the slope. You may have also noticed that the curve approaches one or 100%at large groove widths. This assumes that at large enough groove widths, the subject will always answer correctly.
In reality however, subjects sometimes lose concentration or they may accidentally push the wrong response button. Therefore, they will not always answer correctly even when presented with a large groove width. In other words, there is an upper limit to the height of the psychometric function and this is set by the delta or lapse rate parameter.
Now that we've discussed the parameters of a psychometric function, let's take a look at how the Bayesian adaptive procedure comes to an estimate of these parameters. During a testing session here, I'm running a simulation of model subject who has a threshold of 1.75 millimeters, a slope of five, an elapse rate of 4%The subject's psychometric function is represented by the purple curve. In the top left plot.
This curve is what we're trying to estimate. The plot in the top right corner displays a record of the subject's performance over a block of trials. As shown previously, a red X indicates an incorrect response and a blue cross indicates a correct response.
At the beginning of the experiment. We are uncertain about the shape of the subject's psychometric function. We express this uncertainty by considering a range of possible A, B and delta values.
We assign equal prior probability to each a b delta triplet. In other words, we assign a uniform prior probability distribution over several thousand possible psychometric functions. As the experiment progresses, we learn from the subject answers that some of these functions are more likely than others.
The uncertainty or entropy of the probability distribution decreases to determine which stimulus piece to present Next, the algorithm calculates the expected entropy for each of the groove widths after each completed trial. This is shown in the bottom left plot. The groove width with the lowest expected entropy will be presented to the subject.
This groove width is indicated by a white cross in the upper left plot. This step is crucial for the efficiency of the baying adaptive procedure because we are constantly trying to reduce the uncertainty about a subject's psychometric function by estimating and presenting the groove width, that will maximize the amount of information gained. The plot in the lower right displays the joint posterior probability distribution function or PDF of the slope and threshold parameters.
In this plot, white represents high probability and black represents low probability. To make this plot, we summed or marginalized over the lapsed rate parameter. After each trial, we come to an estimate of the subject's threshold and slope value.
The mode of this plot indicated by the Red Cross is the current best estimate of the threshold and slope. Using these values, we can plot an estimate of the subject's psychometric function. This is shown by the green curve in the top left plot.
As we collect more data, our estimation becomes refined. The joint posterior PDF becomes narrower indicating that our uncertainty of the estimated parameter values is decreasing. You can also see that with more data, the estimated curve approaches the shape of the real curve.
Now that we've finished collecting our data, we can make a better estimate of our parameters. Since our dependent measure is a threshold, we can marginalize over the slope parameter to give the most accurate estimate of the threshold parameter shown here is the posterior PDF of the threshold parameter. The mode of this PDF is our best estimate of the subject's threshold, but there is uncertainty about this estimate.
This uncertainty is represented by the width of the curve. The broader the curve, the less confident we are about our estimate of the subject's threshold and conversely, the narrower the curve, the more confident we are about our estimate of the subject's threshold. This completes our description of taps, the tactile automated passive finger stimulator.
We hope that taps will be replicated by others and serve to promote the move towards control tactile testing, which in recent years has seen promising progress. Please do not hesitate to contact us if you should have any further questions.
