\reffig{fig:tactilepattern} depicts examples of 1D tactile patterns of increased complexity.
The first one is a \emph{constant} pattern, for which the device uses the same friction command over the whole surface.
+Like every other pattern type, we can adjust the friction level.
%It corresponds to the smooth configuration in Schellingerhout's classification.
The second pattern is a \emph{step}.
It corresponds to a frontier between two adjacent zones which have a different friction.
It is the smallest building block we will use later for designing more complext patterns and textures.
+We can adjust its position and direction.
The third pattern is a \emph{shape}: a localized pattern distinct from the background.
It has two steps so that the user can move through it.
+Its parameters arre position and size (width).
The fourth pattern is a \emph{field}, a regular repetition of a shape.
With a sufficiently low size and high number of repetition, users are not able to count the item while exploring the surface.
They rather have a sensation of roughness\fixme{REF?}.
-The fifth pattern is a \emph{gradient}, repetition of a shape, with a decreased size.
+One of the parameters is the duty-cycle, which is the ratio between the signal size and the period size.
+We can also adjust the number of repetitions and the width, which together controls the density of the pattern.
+The fifth pattern is a \emph{gradient}, repetition of a shape, with a variable size.
+There are many ways to increase or decrease the size of shapes of a gradient: linearly, exponentially, etc.
Finally, the sixth pattern is a \emph{random} series of shapes.
+We can control the minimum and maximum size of shapes.
\begin{figure}[htb]
\definecolor{cellblue}{rgb} {0.17,0.60,0.99}
\begin{scope}[]
\node[x=\scale,y=\scale, anchor=center] () at (\dx/2,-10){Constant};
+ \fill[x=\scale,y=\scale,color=cellblue] (0,0) rectangle (\dx,\dy);
\draw[x=\scale,y=\scale,color=black] (0,0) rectangle (\dx,\dy);
\end{scope}
\label{fig:tactilepattern}
\end{figure}
-\begin{definition}{Tactile Texture}
+\begin{definition}{Tactile texture}
We define a \defword{Tactile texture} as combination of one of several patterns on several dimensions or at different scales.
\end{definition}
The second one repeats a field-type pattern, which is itself a repetition of a shape.
We can also see it as a frequency modulation of two signals.
We can obviously use the same structure and parameters than with vibrotactile Tactons (see \reffig{fig:lexical} and \ref{fig:syntactic}).
-However our perception of friction is not as accurate as our perception of a vibration amplitude.
-Therefore we cannot simply translate vibrotactile Tactons into tactile textures.
+However the range of friction we can produce with this device is not as wide as the range of vibration amplitudes we can produce with state-of-the-art vibrotactile actuators.
+Therefore we cannot simply translate vibrotactile Tactons into tactile textures so far.
+It would be interesting to perform JND studies with another device that produces a higher quality variable friction feedback, such as devices made by Hap2U\footnote{\href{https://www.hap2u.net/}{https://www.hap2u.net/}}.
\begin{figure}[htb]
Contrary to vibrotactile Tactons, tactile texture can easily extend to multiple dimensions.
\reffig{fig:tactilepattern} shows 1D textures, but it can extend to 2D, 3D textures or even 4D with a temporal dimension.
-This is not necessarily meant to be a systematic list of possible patterns.
-There are for instance many ways to increase the size of shapes of a gradient: linearly, exponentially, etc.
-However it is an illustration of the expressive power of tactile textures made of steps.
-
-
-What
-
-
-Le concept de textures amène donc à s’interroger sur les formes – spatiales ou temporelles – et plus généralement sur la perception de ces formes. Nous pourrions alors faire la distinction avec le concept d’icône, qui comme nous allons le voir par la suite, se base sur ces formes, mais dont l’étude pose cette fois des questions de sens, avec un objectif in- formatif. Ces deux dimensions cognitives de perception et de sens sont bien-sûr intimement liées, mais nécessitent des approches différentes pour leur étude.
+Other technologies are indeed required for three spatial dimensions, like ultrasound actuators~\cite{carter13}.
+%The pattern types presented in \reffig{fig:tactilepattern} is not necessarily cover all possible patterns.
+%There are for instance many ways to increase the size of shapes of a gradient: linearly, exponentially, etc.
+The examples of tactile patterns and textures we presented do not necessarily cover all possiblilities.
+However, it is an illustration of the expressive power of tactile textures made of elementary steps.
+We will next discuss an evaluation of users' perception of tactile patterns.
+
+\paragraph{Evaluation of tactile patterns}
+
+%The evaluation of tactile patterns is complex because the design space is large.
+The first interesting research question is whether users are able to distinguish the patterns.
+The way of evaluating this is not trivial though because the design space is large and it is difficult, not to say impossible, to present all of them with sufficient repetitions for an accurate analysis.
+Therefore we must evaluate a subset of all possible patterns
+Recent work addressed this issue with a new method for sampling the design space~\cite{demers21}.
+At the time, we multidimensional scaling (MDS) because of the multidimensional nature our patterns.
+This method consists in asking participants to group items together and use the number of times they are grouped together as a similarity metric.
+Then we compute the number of required dimensions and map each item to a position in space so that the distance between two items is proportional to their dissimilarity.
+Finally we identify clusters of similar patterns.
+This method was already succeessfully used to evaluate Tactons~\cite{enriquez06}.
+\reffig{fig:stimtacpatterns} shows the patterns we evaluated~\cite{potier16}.
+They are made of 5 \textsc{shapes} and 7 \textsc{densities}, for a total of 34 patterns (the lowest density vertical line and squares are identical).
\begin{figure}[htb]
\definecolor{cellblue}{rgb} {0.17,0.60,0.99}
\draw[x=\scale,y=\scale,color=black] (0,0) rectangle (\dx,\dy);
\end{scope}
}
- \node[x=\scale,y=\scale, align=right, text width=2cm, anchor=east] () at (0,20){vertical lines};
+ \node[x=\scale,y=\scale, align=right, text width=1.8cm, anchor=east] () at (0,20){vertical lines};
\end{tikzpicture}
\vspace{1mm}
\draw[x=\scale,y=\scale,color=black] (0,0) rectangle (\dx,\dy);
\end{scope}
}
- \node[x=\scale,y=\scale, align=right, text width=2cm, anchor=east] () at (0,20){horizontal lines};
+ \node[x=\scale,y=\scale, align=right, text width=1.8cm, anchor=east] () at (0,20){horizontal lines};
\end{tikzpicture}
\vspace{1mm}
\draw[x=\scale,y=\scale,color=black] (0,0) rectangle (\dx,\dy);
\end{scope}
}
- \node[x=\scale,y=\scale, align=right, text width=2cm, anchor=east] () at (0,20){squares};
+ \node[x=\scale,y=\scale, align=right, text width=1.8cm, anchor=east] () at (0,20){squares};
\end{tikzpicture}
\vspace{1mm}
\draw[x=\scale,y=\scale,color=black] (0,0) rectangle (\dx,\dy);
\end{scope}
}
- \node[x=\scale,y=\scale, align=right, text width=2cm, anchor=east] () at (0,20){dots};
+ \node[x=\scale,y=\scale, align=right, text width=1.8cm, anchor=east] () at (0,20){dots};
\end{tikzpicture}
\vspace{1mm}
\draw[x=\scale,y=\scale,color=black] (0,0) rectangle (\dx,\dy);
\end{scope}
}
- \node[x=\scale,y=\scale, align=right, text width=2cm, anchor=east] () at (0,20){circles};
+ \node[x=\scale,y=\scale, align=right, text width=1.8cm, anchor=east] () at (0,20){circles};
\end{tikzpicture}
\tikzexternaldisable
- \caption[Bla.]{Bla.}
+ \caption[Tactile patterns used in the MDS experiment.]{Tactile patterns used in the MDS experiment. They are made of 5 shapes and 7 densities.}
\label{fig:stimtacpatterns}
\end{figure}
-ground truth: paper?
-t
-est
-
-Tactile Textures~\cite{potier16}
-
-Exploration complète du design space compliuée
-
-Sampling~\cite{demers21}
+The secondary research question is whether the perception of tactile textures with a variable friction device is similar to the perception of a similar pattern on a physical surface.
+We can expect differences because this technology cannot produce edges that we can feel under the fingertip.
+With a squeeze film effect device, edges are produced by changing the friction of the whole surface depending on the finger position.
+Therefore in our experiment we compared tactile patterns rendered on a Stimtac device, and 250g dull coated paper cards as tangible surfaces.
+The stimulation on the paper is created with a transparent ink, which feels stickier than the paper.
+The stickiness depends on the number of printed layers, printed with a HP Indigo Digital Press® printer.
+We performed a pilot study with 3 users to define the number of layers required to match the stimulation with the STIMTAC.
+We opted for 30 layers, which makes a 0.05 mm thickness.
+Both STIMTAC and cards have the same size: 8cm wide and 4cm high.
+
+\paragraph{Discussion}
+
+I will not cover the experimental details, and the full analysis.
+Interested readers will find more details in the corresponding paper~\cite{potier16}.
+I will rather summarize the main findings and discuss further some aspects.
+
+We designed the patterns with several shapes and densities.
+Therefore we expect partitipants to group patterns of same shapes or densities.
+However the MDS analysis fails at identifying a good sets of clusters with 2 or 3 dimensions.
+This means that this is unlikely there is a consensual grouping strategy among users and conditions.
+Looking at the data, we observed participants often grouped vertical and horizontal lines per shape in both conditions.
+We hypothesize this is due to the fact that the sensation with these two shapes is different whether you explore them horizontally or vertically.
+%Participants also grouped together high-density dots, circles and squares.
+%They most likely identified them as a fine-grain
+
+%We observed that vertical and horizontal lines were often grouped per shape, whether with paper or Stimtac.
+%Our explanation is that these patterns are easy to perceive and distinguish from others.
+%The fact that they feel different whether we explore them horizontally or vertically is an essential cue for distinction.
+%Dots have a similar strategy for both Stimtac and paper.
+%They are often grouped with vertical lines and circles at low densities, with squares for high densities and with other dots of similar densities.
+%These patterns are very similar for the density ranges mentioned.
+%This does not necessarily mean users cannot distinguish them, but they feel a similar sensation, which motivates them to group them together.
+%Interestingly, circles are grouped differently for paper and STIMTAC.
+%Circles of high densities are grouped with vertical lines when explored on the paper, and with squares when explored on the STIMTAC. This difference is probably due to a difference of perception. On the left and right sides of the pattern, circles are similar to vertical lines, and at the middle there is a small portion that feels like horizontal line. The fine details are harder to perceive with the STIMTAC than with the paper. The fact that participants felt high densities circles as squares means that they felt a combination of vertical and hori- zontal patterns across the surface, rather than at specific locations on the pattern.
+
+%MDS analysis.
+
+%Initially we performed a Multidimensional Scaling (MDS) analysis. We hypothesized that participants will either group patterns per shape or per density. For a shape strat- egy we expected to see five clusters, each one for one shape, and grouping all densities together, or the inverse. We were also wondering if the grouping strategy would be similar with the paper or with STIMTAC. The distance between each sample was calculated as follows. For each phase of every participant, the distance between two sam- ple is 0 if the samples were in the same group, 1 otherwise. The overall distance is the sum of all distances in every phase of every participant.
+
+%We used the cmdscale function of the R software, which uses a principal coordinates analysis. The goodness of fit we obtain with a 2-dimensions mapping is low (GOF=0.40,0.43 for paper, GOF=0.28,0.28 for STIMTAC). An MDS analysis would require 6 dimensions to have a fair goodness of fit with paper (GOF=0.76,0.81) and 12 for STIMTAC (GOF=0.74,0.75). This means that contrary to our hypothesis, there is probably no dominant group- ing strategy among users and conditions. We use another approach to investigate further the results.
+
+%We further attempted to identify the participants' grouping strategies.
+%To do so we we analyzed every group participants made.
+%For each group, we labeled the grouping strategy as \textsc{Density}, \textsc{Shape}, or \textsc{Mix} according to two metrics.
+%The first metric is the maximum occurrence of a \textsc{Shape} and \textsc{Density} within the group
+%The second metric is the number of different \textsc{Shapes} and \textsc{Densities} within the group.
+%We normalize the values because there are 5 \textsc{Shapes} and 7 \textsc{Densities}.
+%We considered the group was made with a \textsc{Shapes} strategy if the maximum occurrence of \textsc{Shape} is greater than the maximum occurrence of \textsc{Density}, and the number of different \textsc{Shapes} is lower than the number of different \textsc{Densities}.
+%If one of the metrics is equal for both parameters, or if the two criteria give a different strategy we consider the strategy as \textsc{Mix}.
+%We analyzed this data per group size.
+
+Next, we analyzed the composition of pattern groups made by participants.
+We observed if groups contained the same \textsc{shape}, same \textsc{density} or if users \textsc{mixed} shapes and densities.
+In the paper condition, participants used more often a \textsc{shape} or \textsc{density} strategy than a \textsc{mixed} strategy.
+In particular when they made groups of 6 or 7 items, they dominantly used a \textsc{shape strategy}.
+Interestingly, in the Stimtac condition participants mostly used a \textsc{mixed} strategy, and made more groups of 2 items.
+Because the grouping strategy was different in the two conditions, we assume that they perceived the patterns differently with the paper cards and Stimtac.
+
+%We also analyzed the size of groups participants made.
+%In the first block of the experiment the participants were free to choose a number of groups between 2 and 15.
+%In the next two blocks they were instructed to make 5, 10 or 15 groups in a random order, with exception to the closest number to the number of groups they made in the first block.
+
+%First of all 92\% of groups have at most 7 items. This is certainly related to the fact that the two parameters have 5 and 7 levels. It suggests that participants tried to use either a SHAPE or DENSITY strategy. We can observe on the paper condition that groups of 6 and 7 items are mostly made with a SHAPE strategy, which means users made groups of the same SHAPE, with up to 7 different DENSITIES. We do not observe this phenomenon in the STIMTAC condition, which means that participants have more difficulties to distinguish the 7 levels.
+
+%The results also exhibit a difference in strategies between paper and STIMTAC. While the strategies for groups of 2 to 5 items are rather equal with paper, participants use more often a Mix strategy. Since participants of the paper condition do not favor a Mix strategy, there is little chance participants of the STIMTAC condition deliberately use a MIX strategy. Therefore this is most likely due to the difficulty to distinguish these patterns.
+
+%Discussion.
+
+%Overall, participants were able to identify shapes and den- sities, both with the paper and the programmable friction device. They mostly interpret patterns with both methods in a same way. They used different strategies only when patterns had fine details, which suggests that differences occur at the limit of the input and output resolution of pro- grammable friction devices. The whole device produces the same sensation across the device regardless the finger location. This means that the output refresh rate, input rate and input resolution contribute to the signal actually produced. Our results suggest that these limits have an impact on the interpretation of the patterns represented.
+
+Limitation of MDS: not similar does not mean we cannot distinguish
+
+\subsubsection{Conclusion}
+
+On a methodological point of view we see that except at the limits of the device capabilities, the interpretation of simple patterns with a programmable friction device or with a physical surface is similar. This means for example that paper cards can be used for prototyping such interactions for a best case scenario.
+
+On a perception point of view, we remind that this result stands with a one finger exploration.
+This means we do not exploit the richness of large surfaces and the ability to feel different sensations at the same time on different fingers.
+We are not aware of any technology able to produce such effects.
+Many studies, especially with visually impaired users take advantage of this capability, with embossed paper for example.
+We also remind that programmable friction devices like the STIMTAC we used cannot produce edge sensations, because it requires to feel two different sensations under the finger.
+This limitation did not have a particular effect on our results. It could be an issue for a realistic simulation of physical textures. In our case we do not require a realistic sensation.
-MDS\cite{enriquez06}
\subsection{Vibrotactile widgets}
\label{sec:printgets}
projects / hdr.git / commitdiff