 |
In the Smart Shelf
approach we augment products by RFID transponders to give them an unique
identification. Furthermore we embed a transponder detection system into
a shelf. In order to obtain the position of the product we declare the position
of the transponder detection as the position of the product. The challenge
in our prototype implementation was to have a very fine-grained detection
and to be able to differentiate between various items standing close to
each other.
Design
The shelf is decomposed into the antenna system, the reader units and a
central communication unit. It has to perform the following tasks: detect
all transponders in range and read their identification, map the position
of detection to an absolute position on the shelf, communicate the retrieved
Ids and their position to an external system by request.
The detection and reading task is done by the antenna system and the reader
units. The antenna system consists of a 6x6 grid of 36 smaller coils antennas
based on an experimental design creating an electromagnetic detection field
by overlapping of their separate fields, which covers a surface of 400 cm²
almost uniformly. The core of a reader unit is RFID reader and a embedded
microcontroller. It can detect and read 125kHz EM Marin compliant transponders
augmenting items. Although a rather old technology such transponders are
cheap, wide spreaded and very robust. A RFID reading unit is surrounded
by 12 switches to address 12 separate antennas of the antenna system one-by-one.
This multiplexing capability was introduced because a single chip reader
costs an order of a magnitude more than the coil antenna with the switching
circuit for this antenna. The reading of a transponder takes 130ms in average.
Robustness (>>99%) came from an algorithm that is able to detect misreadings.
However, in order to achieve an acceptable response time of the entire shelf
there are 3 reading units working in parallel, but each for a separate detection
area of the shelf. They operate most of the time independently from the
central unit, except for the time of synchronization. Latter is needed because
is has to be avoided that two antennas beside of each other are used to
detect transponders simultaneously. Their detection fields would overlap
and the attempt of reading would result in collision when there is a transponder
on one of them. The central communication unit is responsible for synchronization.
It also performs the task of gathering all detection data from the reader
units and hold the pre-configured information of the position of each coil
antenna. As a consequence the central unit is able to create the overview
of all items with their position on the shelf. An external application can
interfere the shelf's operation by a request on a standard RS232 interface.
The shelf respective the central unit then replies with the last recent
detections and their positions.
|
|
As mentioned before the Smart Shelf
can report the three basic actions take, return and remove to a backend system. In this
section we give an overview of the applications that can be built on top of these actions.
Basically these applications can be classified into three categories: data mining, store
management, and recommendation systems.
Data Mining
Obviously the actions can serve as additional input for data mining and business
intelligence systems. Hence they can provide a means for more detailed analyses of, for
example, trends in customer behavior or the impact of advertising campaigns. In particular
the interesting moment when the customers make their decision about buying a product or
not can be revealed to such systems using the Smart Shelf technology. For example, we can
observe how often a product is taken from the shelf without being actually bought or how
long a customer holds a product in his hands before returning it to shelf. Thus we can
deploy metrics similar to the click-buy-ratio or the time-on-page, which are widely used
in e-commerce systems. Thus we can provide a means to quantitatively evaluate the effect
of advertising campaigns and other measures undertaken to increase the sales of a certain
product.
Store Management
Additionally the actions can be used to build various store management applications like a
shelf out-of-stock watchdog or a plan-o-gram compliance check. That shelf management is a
non-trivial task can be seen from figures telling that approximately 8% of all
out-of-stock situations occur although there are enough products in the back-store, just
because of missed shelf replenishments.
The Shelf Watch application that we built integrates both an out-of-stock watchdog and a
plan-o-gram compliance check. This application manages a virtual model of the shelf
representing its current state, i.e. the current location of the products on the shelf.
The application processes the messages received from the Smart Shelf. As described above
these messages contain for each position on the shelf the identifier of the product
located there or "0" for the positions where no product is located. By receiving
a message describing the complete status of the shelf every 1.5 seconds, the model can be
updated almost in real-time.
The application is connected to a backend system that stores the product information and
the mapping of product identifiers to the appropriate product information. This backend
system also holds the plan-o-gram stating which type of product is expected at each shelf
position.
The application shows a graphical representation of the shelf's state (see Figure 3) and
checks the compliance of that state with the plan-o-gram. If there are any misplaced
products they will be marked by a red frame in the graphical representation allowing an
operator to easily recognize misplacements. If necessary a misplacement alert message can
be sent out to other applications. Similarly the Shelf Watch application can generate an
out-of-stock alert message, if a certain type of product is missing on shelf. Empty fields
and fields with an unknown product on them are also especially marked in the graphical
representation.
Recommendation Systems
The Smart Shelf allows a direct interaction with customers. When a customer performs a
certain action, like picking a product from a shelf, a backend system can immediately
react to this action and trigger an appropriate response action. In this section we will
show how this direct customer interaction can be used to realize recommendation systems in
retail stores.
For such systems we also need some technology to communicate recommendations to the
customers. The technology we have in mind here are electronic price labels as they are
already available on the market today. Such labels consist of a LCD and a wireless
communication interface. They are mainly intended to ease price adaptations, since new
prices just have to be transferred to the electronic labels over the wireless
communication interface, what makes a manual relabeling obsolete. Besides the price
information today's labels can also display small texts, e.g. 2 rows of 16 characters. Of
course it would be technologically no problem to have more advanced labels that can also
display graphics and play sounds.
In the e-commerce world recommendation systems are a well-known technique to increase
cross-sales and customer loyalty. Consequently a lot of such systems are already in use
today. The best known is probably Amazon.com's "Customers who bought" -system.
For almost each kind of e-commerce recoommendation system introduced there, we can find a
reasonable equivalent in the retail world that can be implemented using Smart Shelves and
electronic price labels.
For so-called non-personalized recommendation systems, i.e. systems that recommend the
same products to all customers independent of their own personality, we would not even
need a Smart Shelf but only the electronic price labels with a backend integration. The
price labels could, for example, display a special message, if the respective product is
under the three best selling products of a certain category and thereby recommending the
customers to buy this product. However, if we also use a Smart Shelf we can direct the
customer's attention to such messages exactly at the moment when he is picking the product
from the shelf, e.g. with a small beep emitted by the price label. This will certainly
increase the number of customers who read the message.
Item-to-item correlation systems are another type of recommendation systems. They base
their recommendations on the knowledge about a customer's interest in one or more
products. In order to realize such systems in a retail store we can use the Smart Shelf to
recognize when a user is picking a certain product from the shelf what certainly indicates
an interest in this product. At that moment the backend system can identify related
products and recommend them to the customer through the electronic price labels. For
example, we can inform customers picking a mobile phone from a shelf about the appropriate
cases that are also available in the store. Less obvious correlations between items can be
determined by analyzing the data provided by the checkout systems, i.e. the output set in
our model. From this data we can see which products are often sold together and should
therefore be considered as correlated.
|
