The design outlined here is intended to aid people to mix better cocktails. The specificity of the task for which the tools are optimised allows for a simple interface at the cost of generality (as discussed in the previous lecture). I wanted to create an interface for the many glasses and liquids that are needed to mix cocktails, to allow
anyone or group of
people to easily mix drinks that normally require tedious instructions
or skilled barstaff (the first is unwanted and the second usually
unavailable). The system plays on the use of semitransparent containers
(bottles and glasses) as a possible means of implementing the tools
with reasonably simple technology. The interface consists of a bar or
table on which moving images can be presented, a series of normal
cocktail glasses and bottles of spirits, a database of cocktail recipes
and an image capture and object segmentation system.Indication of tasty mixes
My aim is to provide
simple guides to combining spirits and mixers in the form of digital
shadows, that attach themselves to a drinker's glass. I wanted several
modes of operation. The first is a simple guide to what goes well with
what is already in your 
glass, based on an a
priori database of cocktail recipes. If someone picks up a container of drink,
having selected a glass, and not having indicated a preferred drink, I
want the system to provide feedback as to how well previous drinkers
believe the liquid you are about to add to your glass will compliment
what is already there. This information is indicated by appending a
cluster of arrows (pointing towards the glass) and crosses, that become
visible as the bottle is picked up. The glass of interest is inferred
by the software from the glass that was last moved. The arrows show a
good match, the crosses a disgusting one. A mixture of the two symbols
indicates some debate among tastes in classic cocktails and previous
drinkers in the database. As positive matches are encouraged, the
drinker's attention is drawn to the ambient information clustered
around the glass by having the arrows appear and move outwards from the
base of the glass as if they are bubbling or fizzing. This energy and
movement draws the eye to the glass. In contrast a bad match should
only be subtly indicated until a last minute warning is given. In this
case crosses slowly fade into view around the glass as the bottle is
picked up. If the bottle is brought close to the glass and tipped, the
crosses vibrate a small amount, with increasing speed and become
darker. This noisy visual feedback should draw attention to the warning
at the last minute. Inventing your own drinks
Of course, part of the joy of mixing drinks is making your own. Therefore, if the user ignores the taste information and continues to mix what the system considers unsuitable, and then proceeds to drink the result, the system should infer that the drink is to that persons taste, and adjust the weight for that combination of liquids for future users (this way the system can adapt, but reflects generally popular tastes). If a new combination of ingredients is formed, the system will create a 'cocktail-like' name and assign it to the drink so that the recipe is not lost.
Making a specific drink
In the second mode of use a user can select a recipe to follow, this is done using a drink coaster. When placed on the table, a new coaster will bring up a suitable menu system. A toolglass-like menu seems appropriate. By dragging the coaster slightly
into
one of the adjacent squares, an option can be
selected, and a new grid of options becomes available (dragging back in
the direction, the menu has just moved should move the user back up the
menu). Two ways of selecting a recipe are available, the user can
navigate by taste (sweeter, fruitier, etc.) or by category (classic
-> martinis -> etc.). Two switch between the two modes, the
coaster is flipped over. With the coasters, the interface becomes
slightly less obvious than simply pouring drinks, it is hoped that the
metaphor is sacrificed for further function.
In both modes of operation, a drinker can reveal the name of the drink, or the names of all possible drinks that can be made with the currently mixed ingredients, by reaching out to the glass. If the list of possible drinks is large, a smaller list of super-categories is listed instead.
Guide to pouring
As a bottle is brought close to a glass to pour, a bar indicating the volume of liquid already in the glass is slowly revealed. The bar should stretch into view from the base of the glass as the bottle is moved and tipped toward the glass, forming a
shadow-like
impression. The bar
should extend out from the glass in the direction of the bottle, making
it easy to see for the pourer, and consistent in it's location relative
to the user's view point. Lines are place across the bar at stop points
for available recipes, and the fill level should rise as liquid is
added. This provides a simple visual method for judging amount added
(and no need to tilt your head).
A further idea I have explored is ways of visually linking a glass to the ingredients still required for a recipe. In a multiuser case, I think this would lead to a confusing weaving of connections.
Implementation
I feel that with an overhead view of the mixing table most if not all the required information could be collected. Glass and bottle shapes are limited in
number, and should be
relatively easy to identify. If it is
possible to infer bottle angle from it's outline, the amount of liquid
poured could be calculated. This information could be supplemented by
viewing through the glass of the bottle at the level
of liquid inside as it is tilted. User interactions
would be clear from the
dark shadows of a users arms (relative to the lit table below). The table could be projected onto from
below. For image capture, either a light above and a camera for the
shadows below the table, or a camera above, and subtraction of the
projected image from the camera signal to provide a clean image of the
glasses, bottles and arms in view.
To introduce redundancy of information, thus making the system more robust, a series
of horizontal
perspectives through the glasses (against a light background), would
provide information on the level of liquid in each glass (rather than
having to keep a running total (which might lead to errors). In the
design shown here, multiple cameras are mounted in the edge of the bar,
facing back across it. In the multi user case (such as a cocktail party), I think the system would be able to maintain track of all the glasses and users if enough line of sight could be maintained, and if the users were seated (which could be confirmed through chair sensors, etc.) which allows the system to infer who's hand is whose by following the arm off the table to it's origin. The system
allows people to
add liquid to some else's glass (a natural action), because the fill
bar appear as a bottle is brought close to a glass. Mixing tips are
provided only for the bottle you are holding and the glass you have
most recently touched.