%\subsection{The haptic pitfall path}
-The design of haptic systems requires joint efforts between specialists in several scientific domains as depicted in Figure~\ref{fig:hapticpath}.
+The design of haptic systems requires joint efforts between specialists in several scientific domains as depicted in \reffig{fig:hapticpath}.
This pipeline has two steps on the system side and two steps on the human side.
Both the system and humans have a step in the physical world and one outside the physical world.
The objective of this pipeline is to transmit information to users through their sense of touch.
\paragraph{Sense of touch}
The objective when producing a physical effect with a haptic system is to stimulate the users' sense of touch.
-On Figure~\ref{fig:hapticpath}, I separate the pure sensing part from the interpretation part.
+On \reffig{fig:hapticpath}, I separate the pure sensing part from the interpretation part.
The body has sensors in the skin, muscles, tendons and articulations~\cite{bolanowski88}.
These sensors are different type of nerves and similarly to electronic sensors they transform physical effects into electric signals.
This document will not cover this in detail.
For example, the funneling illusion, also called phantom sensations, produces a virtual vibration in between two points\cite{alles70,vonbekesy58}.
To do so, one must place two actuators, one at each end.
The amplitude of the signal on both actuators is a ratio corresponding to the desired position of the virtual vibration.
-For example, let $A$ be the maximum amplitude and two actuators $a$ and $b$ placed $4cm$ apart, as depicted on the left part of Figure~\ref{fig:illusions}.
+For example, let $A$ be the maximum amplitude and two actuators $a$ and $b$ placed $4cm$ apart, as depicted on the left part of \reffig{fig:illusions}.
If the amplitude of actuator $a$ is $A(a) = \frac{A}{4}$ and the amplitude of actuator $b$ is $A(b) = \frac{3A}{4}$, then the users feel a vibration between $a$ and $b$, $1cm$ away from $a$ and $3cm$ away from $b$.
Tactile saltation, also called the cutaneous rabbit illusion, gives the illusion of a sequence of equally spaced vibration~\cite{geldard72}.
The tactile stimulation however is a repeated vibration on a smaller subset of locations.
-For example, with three actuators $a$, $b$, and $c$ equally spaced on a straight line, the stimulation is three vibrations on $a$, three on $b$, and three on $c$, as shown on the right part of Figure~\ref{fig:illusions}.
+For example, with three actuators $a$, $b$, and $c$ equally spaced on a straight line, the stimulation is three vibrations on $a$, three on $b$, and three on $c$, as shown on the right part of \reffig{fig:illusions}.
The person feels nine equally spaced stimulations between $a$ and $c$.
Israr \etal combined both funneling and saltation to produce 2D tactile motions, not only in straight line but also on curves~\cite{israr11}.
It also reveals the many pitfalls at all levels that can make users perceive something different than what was intended.
Any step in this pipeline potentially introduces a drift with respect to the original message.
For example the resolution of commands could be too low, the mechanical response to the command can be non-linear, the quality of the contact between the haptic system and the user can be sub-optimal, the mechanical effects can be out of the perceptual range, and haptic illusions can alter the interpretation of sensations.
-While the work of researchers in specialized domains mentioned in Figure~\ref{fig:hapticpath} is to avoid, or at least minimize such drifts in their part of the pipeline, my work as an HCI researcher consists in selecting, adjusting, combining, designing, implementing and evaluating such parts to build \defword{interactive systems}.
+While the work of researchers in specialized domains mentioned in \reffig{fig:hapticpath} is to avoid, or at least minimize such drifts in their part of the pipeline, my work as an HCI researcher consists in selecting, adjusting, combining, designing, implementing and evaluating such parts to build \defword{interactive systems}.
My research is therefore fundamentally interdisciplinary, and I combine, complement or replace expertise of different domains, depending of the needs of each research project, and the expertise of my collaborators.
My objective is to make interactive systems better than the sum of their parts.
\end{figure}
The \emph{lexical level} defines the basic vocabulary of the device.
-For example, we can control vibrations in frequency, amplitude, shape, and duration (Figure~\ref{fig:lexical}).
+For example, we can control vibrations in frequency, amplitude, shape, and duration (\reffig{fig:lexical}).
Hence the command for a vibrotactile actuator is an electrical signal made of these elements.
The research challenge here is to bridge the knowledge gap between the engineering of actuators and human factors, and find tactile parameters that users can perceive and distinguish.
Such research is typically useful to guide the design of tactile actuators.
The \emph{syntactic level} combines lexical items to form haptic phrases.
They eventually combine different modalities.
%, into higher level representations.
-The figure~\ref{fig:syntactic} shows several examples of such haptic phrases with vibrotactile feedback.
+The \reffig{fig:syntactic} shows several examples of such haptic phrases with vibrotactile feedback.
Frequency and amplitude modulation create new kinds of feedback, that Brown \etal describe as roughness \cite{brewster04}.
They also use sequences of vibrations to form rhythms.
The challenge here is to find combinations of parameters that users are able to interpret together.
The \emph{semantic level} represents the mapping between the haptic effect and its associated meaning.
For example, if we create Tactons for meeting alerts, we can encode the kind of meeting with a rhythm, the importance with roughness, and the delay with spatial location.
The combination of both parameters enables encoding every level of urgency for every caller ID.
-Figure~\ref{fig:semantic} illustrates this example of mapping between multi-parameters Tactons and hierarchical information.
+\reffig{fig:semantic} illustrates this example of mapping between multi-parameters Tactons and hierarchical information.
\begin{figure}[htb]
\centering
As activity performace is generally evaluated as a percentage or as a value on a scale, we created vibrations corresponding to the values 1 to 10, with the objective of representing 10\% increments.
Because of our choice of using a single basic vibration actuator, we encode the vibrations using the duration and rhythm parameters only.
Since there was no prior encoding of discrete numbers found in the literature using duration and rhythm only, we first had to determine the best encoding pattern for ActiVibe.
-%We designed a total of six patterns (Figure~\ref{fig:activibesets}) and in our subsequent evaluations used the pattern with the highest accuracy rate for ActiVibe.
+%We designed a total of six patterns (\reffig{fig:activibesets}) and in our subsequent evaluations used the pattern with the highest accuracy rate for ActiVibe.
-Figure~\ref{fig:activibesets} shows a visual representation of the pattern sets that we evaluated.
+\reffig{fig:activibesets} shows a visual representation of the pattern sets that we evaluated.
Each individual squiggly line represents a single short pulse, while a long line represents a longer vibration.
We first designed the series of vibration sets (A-E) that were evaluated in a first laboratory setting.
The results of the first study helped us to design pattern F, which was then compared in a second laboratory study to the best sets from the first study (A, C, and E).
We explored two possibilities: 1) represent the actual value only; 2) represent the value as well as the scale.
%\paragraph{Duration Only}
-When representing the actual value only, the duration of the vibrotactile pattern depends on the value it represents (Figure~\ref{fig:activibesets}, sets A \& B).
+When representing the actual value only, the duration of the vibrotactile pattern depends on the value it represents (\reffig{fig:activibesets}, sets A \& B).
The pattern has a short average duration, and thus it may be hard to understand the distance to the end of the event represented.
We distinguish two variations.
In set A, each value is represented by a series of short pulses, separated by short pauses.
The disadvantage of representing only the value is that even if the user has an idea of the current value, there is no clue about the distance between this value and the maximum value.
Introducing a scale enables the positioning of a value relative to the beginning and the end of a progression.
Sets C-E represent both the value and the scale of a progression in several ways.
-Either the current value is represented by a series of short vibrations and the scale by filling the sequence with a long vibration, either before or after the value (Figure~\ref{fig:activibesets}, sets C \& E), or the current value is represented by a long vibration and the scale is represented by filling the sequence with a series of short vibrations (Figure~\ref{fig:activibesets}, set D).
+Either the current value is represented by a series of short vibrations and the scale by filling the sequence with a long vibration, either before or after the value (\reffig{fig:activibesets}, sets C \& E), or the current value is represented by a long vibration and the scale is represented by filling the sequence with a series of short vibrations (\reffig{fig:activibesets}, set D).
Set F was defined using the results from the first laboratory study as a combination of short pulses and long vibrations.
\begin{table}[htb]
\subsubsection{Perception of programmable friction}
The connection between the command and what users perceive through their fingers is not trivial though.
-This is a typical example of the phenomenon discussed at the beginning of this chapter, and illustrated on Figure~\ref{fig:hapticpath}.
+This is a typical example of the phenomenon discussed at the beginning of this chapter, and illustrated on \reffig{fig:hapticpath}.
Both the relation between the command and the vibration amplitude and the relation between the amplitude and the sensation of slickness are not linear~\cite{biet07}.
They are affected by many environmental factors such as air moisturre, temperature; surface properties such as material, cleanliness; and fingers aspect: tribology, cleanliness, moisture.
There are differences in performance between prototypes.
The first study in which I was involved consisted in evaluating the Just Noticeable difference (JND) of friction between two adjacent zones.
The objective is to measure the useful resolution of the device.
With this information, our colleagues can optimize the power consumption of the device, and we can design more complex patterns such as textures.
-This work was not published, therefore I will get in more details below.
+This work was not published, therefore I will get in more details below with a new statistical analysis.
\paragraph{Methodology}
-%We experimented six reference levels, spread out linearly between the minimum and maximum command ($0^\circ$ to $180^\circ$ phase shift).
-%For the first three values ($0^\circ$, $36^\circ$, and $72^\circ$) we searched the minimum greater value for which participant could feel a step.
-%For the last three values ($108^\circ$, $144^\circ$, and $180^\circ$) we searched the minimum lower value for which participant could feel a step.
+%We experimented the perception of a difference of friction between two adjacent zones.
+%The left and the right part of the surface was split into two equal sized areas.
+%We were interested
+
We used an one-up/two-down adaptative method, which enables a faster convergence to the JND value~\cite{leek01}.
-It consists in reducing the intensity of the signal after two good answers, and augmenting it after one wrong answer (Figure~\ref{fig:adaptativeprocedure}).
+It consists in reducing the intensity of the signal after two good answers, and augmenting it after one wrong answer (\reffig{fig:adaptativeprocedure}).
The evaluation of each trial used a 3-alternatives forced choice (3-AFC).
This means participants were presented 3 configurations.
One of them had a signal: a difference of friction between the left and right side of the device.
\definecolor{cellred}{rgb} {0.98,0.17,0.15}
\definecolor{cellblue}{rgb} {0.17,0.60,0.99}
\def\sx{4.7mm}
- \def\sy{15mm}
+ \def\sy{10mm}
\def\revthickness{1.5pt}
\newcommand{\good}[1]{
\begin{tikzpicture}
%Axis
\draw[x=1mm,y=1mm, <->]
- (-5,70) -- (-5,0) -- (150,0);
+ (-5,45) -- (-5,0) -- (150,0);
\node[x=\sx,y=\sy] at (0,-0.3) {1};
\node[x=\sx,y=\sy] at (9,-0.3) {10};
\node[x=\sx,y=\sy] at (19,-0.3) {20};
\end{figure}
The first trial of each block used the largest difference for the chosen reference value.
-For example with a reference command of $0^\circ$ the other value was $180^\circ$.
+For example with a reference level of $0^\circ$ the other value was $180^\circ$.
Then, the value decreased of $1.8dB$ after two good answers in a row, and increased of $1.8dB$ after a wrong answer.
After 3 reveresals we reduced the increment/decrement to $1.2dB$ since participants went closer to the perception threshold.
-The block ended at the $13\nth$ reversal.
+The block ended at the \nth{13} reversal.
We estimated the threshold by averaging the values corresponding to the last 10 reversals.
-PArticipants typically performed between 30 to 50 trials per block.
-
-
-Six valeurs de référence ont été choisies pour le déphasage fixe utilisé dans un bloc, équitablement réparties sur la plage de fonctionnement du dispositif, correspondant à 0\%, 20\%, 40\%, 60\%, 80\% et 100\% de cette plage. Ces six valeurs correspondent aux conditions de l’expérience. Pour les trois premières, le déphasage variable est progressivement abaissé de 180° vers 0° pour réduire l’intensité du signal (approche descendante). Pour les trois dernières, il est au contraire augmenté de 0° à 180° pour obtenir le même effet (approche montante).
-
-Le dispositif utilisé pour cette expérience était un STIMTAC standalone semblable à celui de la Figure 2. La plaque vibrante de ce dispositif (75 x 40 mm de cuivre-béryllium) est recouverte d’un film plastique transparent filmolux® easy clear matt et 36 céramiques piézoélectriques sont collées dessous, 35 servant à la faire vibrer et la dernière mesurant l’amplitude de vibration. Ce dispositif était alimenté et connecté par USB à un PC sur lequel était exécuté une application spécifique implémentée avec la librairie tIO précédemment mentionnée. Un cache en plastique (non visible sur la Figure 2) était par ailleurs placé au-dessus de la plaque pour limiter la surface à explorer à une zone de 75 x 20 mm et réduire ainsi la variabilité possible de l’amplitude de vibration.
-
-Les sujets devaient s’asseoir et il leur était demandé de n’utiliser que l’index de leur main dominante. Les sujets portaient également un casque réducteur de bruit afin d’éviter qu’ils ne cherchent des indices sonores durant la tâche. La surface à explorer était nettoyée à l’alcool avant chaque bloc. La vitesse d’exploration était laissée libre, mais pour s’assurer que la marche soit explorée dans les deux sens, il était demandé aux sujets de réaliser au moins 3 aller-retours entre les zones gauche et droite. La procédure d’auto-ajustement de la fréquence de référence de tIO était appelée par le programme avant chaque phase d’exploration de chaque essai. Une représentation visuelle de la force attendue et de celle exercée sur la surface était présente à l’écran durant les phases d’exploration, les sujets ayant pour consigne de maintenir une force constante.
-L’expérience commençait par une phase de familiarisation où différents signaux étaient présentés au sujet afin qu’il comprenne ce qu’il allait devoir chercher et la force qu’il allait devoir appliquer. Un signal faible était ensuite présenté pour évaluer sa capacité à réaliser les 6 blocs prévus (aucun sujet n’a été écarté). Une fois cette phase de familiarisation terminée, l’expérience pouvait réellement commencer.
-Des éléments de jeu ont été utilisés pour rendre la tâche à réaliser plus attractive .
-Trois fantômes d’apparence visuelle semblable étaient présentés à l’écran (Figure 3), le but étant d’attraper celui présentant une différence au toucher. Le sujet pouvait passer d’un fantôme au suivant en appuyant sur la touche espace du clavier, mais ne pouvait pas revenir en arrière. Le temps passé sur chaque fantôme était de 12 secondes au plus pendant lesquelles le sujet pouvait explorer la surface et essayer de détecter le signal (la marche tactile). Le fantôme exploré était visuellement différencié des deux autres et un son (1, 2 ou 3 bips) était produit à chaque changement. Une fois les trois possibilités considérées, le sujet devait indiquer le fantôme ayant produit selon lui un retour tactile en appuyant sur la touche correspondante (1, 2 ou 3) du clavier. Aucun retour sur la qualité de la réponse n’était alors produit.
-
-12 sujets ont pris part à l’expérience (tous droitiers ; 9 hommes ; d’un âge moyen de 27,7 ans ; aucun n’ayant a priori de problème de sensibilité tactile). Le passage de chaque sujet était organisé en 2 sessions de 3 blocs entrecoupés de pauses, les sessions se déroulant sur deux jours différents. De 30 à 50 essais ont été réalisés par bloc, chaque bloc ayant duré moins de 20 minutes et leur ordre pour un sujet ayant été déterminé par un carré latin.
+%Participants typically performed between 30 to 50 trials per block.
-\paragraph{Results}
+We experimented six reference levels, one for each block.
+The values were spread out linearly between the minimum and maximum command (\degr{0} to \degr{180} phase shift).
+For the first three values (\degr{0}, \degr{36}, and \degr{72}) we searched the minimum greater value for which participant could feel a step.
+For the last three values (\degr{108}, \degr{144}, and \degr{180}) we searched the minimum lower value for which participant could feel a step.
-Les résultats de l’expérience ont été analysés dans l’espace des valeurs de déphasage (espace de la commande) ainsi que dans l’espace de l’amplitude de variation mesurée par la céramique piézoélectrique (Figure 4). Les seuils de détection (hauteurs minimales de la marche tactile) ont été analysés entre conditions à l’aide d’un test de Kruskal-Wallis pour données non-paramétriques.
+The Stimtac device version we used in this experiment is depicted on \reffig{fig:stimtac}.
+The surface is $75 \times 40mm$, made of Copper-Beryllium, covered with a \emph{Filmolux® easy clear matt} plastic sheet.
+36 piezo-electric ceramics are glued underneath.
+35 of them are actuated to vibrate the surface, anf the last one is used as a sensor to measure the vibration amplitude.
+The device was connected to a PC through USB, on which the experimentall application was running.
+A plastic cover, not visible on the picture, was placed over the surface to reduce the interactive surface to $70 \times 20mm$ to reduce the variability of amplitude and guide the user on a lateral movement.
-Nos analyses montrent un effet significatif de la valeur de référence dans l’espace de déphasage (p < 0,001), mais pas dans l’espace d’amplitude de vibration (p = 0,485). Un test de comparaison deux à deux montre une différence entre les conditions 20% et 0%, entre 80% et 100%, ainsi qu’entre 40% et 80%.
-La hauteur de marche minimum pour être détectée (MOY sur la Figure 4) varie suivant les sujets et la valeur de référence choisie. Ces variations peuvent s’expliquer en partie par des non linéarités entre la commande (le déphasage souhaité), la tension d’alimentation des céramiques et l’amplitude de vibration résultante. Nous pouvons toutefois proposer des recommandations générales pour le STIMTAC standalone. Pour être détectée à coup sûr, une marche de déphasage doit avoir une hauteur d’au moins 100°. Et toute marche inférieure à 10° a de grandes chances de ne pas pouvoir être détectée. La variabilité observée dans l’espace de l’amplitude de vibration s’explique sans doute également par celle de la pression exercée par les sujets, malgré sa représentation visuelle, et par l’échauffement du dispositif, malgré la procédure d’auto-ajustement de sa fréquence de référence. Pour palier à ces problèmes, il faudrait idéalement que le dispositif puisse être contrôlé en termes d’amplitude de vibration, et non de déphasage, et que les mesures faites par la céramique spécialisée soient utilisées pour asservir ce contrôle.
-Nous avons comparé les vitesses d’exploration 100 ms avant et après la traversée de la marche pour les 4 premiers essais de chaque bloc, où les marches sont les plus hautes. Une ANOVA à un facteur n’a montré aucun effet de la présence effective de la marche sur la vitesse (F(1,622) = 0,336 ; p = 0,562).
-L’absence de variation de vitesse entre les deux côtés des marches signifie que l’effet perçu n’est pas dû à des variations de mouvement mais plus probablement à la déformation de la peau, comme on l’observe sur des éléments tactiles classiques en relief. Dans ces conditions, la technologie STIMTAC pourrait-elle être utilisée pour simuler des reliefs et des creux par simples variations de frottement ? La nature même de cette technologie la rend a priori pour le moment incompatible avec la simulation de tels éléments physiques : le frottement étant modifié de la même manière sur toute la surface, le doigt ne pourra jamais toucher en même temps deux états différents, comme lorsqu’il traverse une arrête.
-
-\paragraph{Conclusion}
-
-Les résultats obtenus fournissent des informations utiles pour l’amélioration des dispositifs STIMTAC et la conception de retours tactiles par frottement variable. Les expériences pilotes et celle décrite ci-dessus ont mis en lumière une importante variabilité liée aux nombreuses non-linéarités entre la commande (déphasage) et l’effet produit (réduction de frottement), aux conditions environnementales (température, humidité), à la précision de la mesure de la position du doigt sur la plaque, à la sensibilité des personnes, etc. Les évolutions matérielles de la technologie STIMTAC (sous-projet 2) et les solutions intégrées développées dans le projet (sous-projets 3 et 4) permettront sans doute de réduire ces sources de variabilité dans un futur proche et d’étudier dans de meilleures conditions d’autres aspects comme la distance minimale nécessaire dans l’espace et le temps pour distinguer deux signaux tactiles élémentaires. Nos prochaines études expérimentales porteront sur des objets tactiles plus complexes : des textures. Le travail dans ce sens a déjà commencé . Nous étudions notamment en ce moment les moyens de nous abstraire (au moins en partie) de la technologie visée pour ce qui concerne la compréhension des dimensions caractéristiques des textures à base de frottement variable
-
-
-Figure~\ref{fig:stimtacmarches}) shows .
+\begin{figure}[htb]
+ \centering
+ \includegraphics[height=6cm]{figures/stimtac}
+ \caption[Stimtac device]{The Stimtac device produces a variable friction haptic feedback.}
+ \label{fig:stimtac}
+\end{figure}
+We instructed participants to use the index finger of their dominant hand only.
+They wore a noise cancelling headset to avoid gueses with audio cues.
+We cleaned the surface with isopropyl alcohol before each block.
+To make sure participants felt the signal in both directions, we instructed them to perform at least 3 back-and-forth between the left and right zones of the surface.
+%Before each trial, the device measured the amplitude of vibration and adjusted the signal to ensure a consistent effect over each trial.
+The experiment application showed a pressure bar and participants were instructed to remain in a specified range.
+
+The experiment started with a training phase in which signals of different intensity were presented to participants.
+We presented them a low intensity signal to make sure they will be able to perform all 6 blocks.
+All the participants felt this signal.
+The application showed three visually identical items, representing the three configurations of the trial.
+Participants could switch between them with the \Space key, but could not go back to the previous ones.
+The current item was highligted, and we limited the exploration time for each item to $12s$.
+After exploring the three items, participants had to indicate which one presented the signal with the \keys{1} \keys{2} \keys{3} keys.
+Participants did not receive any feedback whether their answer was good or wrong.
+
+12 participants took part of this experiment, all of them were right-handed, their mean age was $27.7$ years old, and none of them had a known tactile sensitivity issue.
+Participants performed 2 sessions of 6 blocks with pauses between blocks and the two sessions happened on a different day.
+Each block typically consisted in 30 to 50 trials and lasted about 20 minutes.
+The order of blocks was balanced with a Latin square.
+
+\paragraph{Results and discussion}
+
+In this analysis, we used 10 \textsc{Reversals} $\times$ 6 \textsc{Levels} $\times$ 12 \textsc{Levels} = 720 trials.
+The \reftab{tab:jndstimtac} shows the mean JND value and standard deviation for each reference level.
+The \reffig{fig:stimtacmarches} shows two representations of the results.
+On the left, for each level the bars represent JND value.
+%On the right, for each level the bars represent the absolute values of both the reference level and the JND.
+On the right, for each level the bottom of the bars represent the reference level and the height of the bar is the JND value.
+It gives an absolute view of the phase-shift associated with the reference level and the JND.
+For both charts, the error bars are 95\% confidence intervals.
+On the right side the error bar is on the variable end because the reference level did not change.
\begin{figure}[htb]
\centering
- \includegraphics{figures/stimtac_marches}
- \caption[Stimtac marches]{Stimtac marches}
+ a)
+ \includegraphics{figures/stimtac-psychophysics2}
+ \hfill
+ b)
+ \includegraphics{figures/stimtac-psychophysics}
+ \caption[JND between two friction values]{Results of the experiment about the JND between two adjacent zones. a) The relative phase shift JND corresponding to the difference between the reference level and the variable value. b) The absolute values of the reference level and the JND. Error bars are 95\% confidence intervals.}
\label{fig:stimtacmarches}
\end{figure}
-issue with JND between 80\% and 100\%
+\begin{table}[htb]
+ \renewcommand{\arraystretch}{0.7}
+ \centering
+ \begin{tabular}{c c c}
+ \toprule
+ \textbf{Reference level} & \textbf{Mean JND} & \textbf{Standard Deviation} \\
+ \midrule
+ \degr{0} & \degr{60.0} & \degr{33.8}\\
+ \degr{36} & \degr{29.8} & \degr{16.6}\\
+ \degr{72} & \degr{34.1} & \degr{17.0}\\
+ \degr{108} & \degr{39.8} & \degr{15.5}\\
+ \degr{144} & \degr{57.6} & \degr{16.8}\\
+ \degr{180} & \degr{64.6} & \degr{33.7}\\
+ \bottomrule
+ \end{tabular}
+ \caption[Variable friction mean JND and standard deviation]{Mean JND values and standard deviation for each reference level.}
+ \label{tab:jndstimtac}
+\end{table}
+
+A Shapiro-Wilk normality test shows the data does not follow a normal distribution (\p{0.0001}).
+We performed a boxcox correction ($\lambda=0.18$), however the data distribution remained not normal (\p{0.0001}).
+Therefore we analyzed our data with a Kruskal-Wallis rank sum test, which showed significant differences (\chisquares{5}{204.45})
+The post-hoc analysis with pairwise Wilcoxon rank sum tests shows two groups of reference levels with significant differences between the groups.
+The difference between the JND and the \degr{0}, \degr{144}, and \degr{180} reference levels was significantly higher than with the \degr{36}, \degr{72} and \degr{104} (\p{0.0001} for all differences, except between \degr{72} and \degr{108} for which \p{0.05}).
+
+%Shapiro-Wilk normality test
+%data: res
+%W = 0.93491, p-value < 2.2e-16
+
+% => not normal
+
+% boxcox correction: $\lambda=0.1818182$
+% => does not work
+
+%> kruskal.test(StepRaw ~ PhaseFixRaw, data=datatr)
+% Kruskal-Wallis rank sum test
+%data: StepRaw by PhaseFixRaw
+%Kruskal-Wallis chi-squared = 204.45, df = 5, p-value < 2.2e-16
+
+% Pairwise comparisons using Wilcoxon rank sum test with continuity correction
+%data: datatr$StepRaw and datatr$PhaseFixRaw
+% 0 36 72 108 144
+%36 6.3e-13 - - - -
+%72 1.8e-08 0.23509 - - -
+%108 0.00011 2.2e-06 0.01177 - -
+%144 1.00000 < 2e-16 < 2e-16 1.8e-14 -
+%180 1.00000 < 2e-16 9.6e-15 1.3e-09 1.00000
+%P value adjustment method: holm
+
+% PhaseFixRaw mean.var sd.var n.var se.var lower.ci.var upper.ci.var
+% <fct> <dbl> <dbl> <int> <dbl> <dbl> <dbl>
+%1 0 60.0 33.8 120 3.08 53.9 66.1
+%2 36 29.8 16.6 120 1.52 26.7 32.8
+%3 72 34.1 17.0 120 1.55 31.0 37.2
+%4 108 39.8 15.5 120 1.41 37.0 42.6
+%5 144 57.6 16.8 120 1.54 54.6 60.7
+%6 180 64.6 33.7 120 3.08 58.5 70.7
+
+First of all, it is important to remind that the squeeze film effect reduces friction.
+Therefore, when the command is high, the surface is more slippery than when the command is low.
+Therefore on the figures, the left and bottom sides correspond to high friction and the right and top sides correspond to low friction.
+Therefore we explain the fact that the higher difference between the reference levels and the JND are on the edge values (\degr{0}, \degr{144}, and \degr{180}) with the non-linearity of both the effect produced by the command (effect on \degr{0}) and the perception of the mechanical effect (effect on \degr{144}, and \degr{180}).
+
+The overall, and conservative, recommendation here is to use differences of phase shift greater than \degr{100} to create patterns with this implementation of Stimtac (larger mean JND plus standard deviation).
+In any case it is unlikely users can perceive difference of phase shift lower than \degr{10} (lower mean JND minus standard deviation).
+Unfortunately, at the time I am writing this I have no access to the amplitude measurement data.
+This makes impossible to draw more general recommendations.
+The measurement with a laser vibrometer of a \degr{180} commang gives a $2.2\mu m$ amplitude of vibration.
+It is however difficult to measure an actual friction value because it depends on the force applied on the surface, which varies when users move their finger to explore the surface.
+
+
\subsubsection{Output vocabulary: Tactile Textures}
Motion is an essential aspect of interaction with peripherals.
Pointing devices rely on movement measurements.
Keyboards use binary key positions as input data.
-In the Métamorphe project~\cite{bailly13}, we actuated the keys so that they can either be up or down (Figure~\ref{metamorphe}, left).
+In the Métamorphe project~\cite{bailly13}, we actuated the keys so that they can either be up or down (\reffig{metamorphe}, left).
Keys can still be pressed, whether the key is up or down.
\begin{figure}[!htb]
\end{figure}
This shape changing keyboard has new properties compared to regular keyboards.
-When a key is up, the use can push it in four directions, or even pinch it (Figure~\ref{metamorphe}, right).
+When a key is up, the use can push it in four directions, or even pinch it (\reffig{metamorphe}, right).
With a touch sensor all around it, the key could be used as an isometric pointing device such as a trackpoint.
Our previous studies showed that raising keys eases eyes-free interaction with the keyboard.
Other users turned their screen either to show visual content to somebody, or to show something in the room in a video conference with the camera affixed to the screen.
It is also frequent to move the mouse and keyboard to make space on the desk for something else.
-In the Living Desktop project\cite{bailly16}, we actuated a mouse, keyboard and screen (Figure~\ref{livingdesktop}):
+In the Living Desktop project\cite{bailly16}, we actuated a mouse, keyboard and screen (\reffig{livingdesktop}):
\begin{itemize}
\item The mouse can translate in the $x,y$ plane directions.
\item The keyboard can rotate, and translate in the $x,y$ plane directions.
projects / hdr.git / commitdiff
