Editors: Frédéric Thiesse • [email protected] Florian Michahelles • [email protected] Building the Internet of Things Using RFID Internet of Things Track Things of Internet The RFID Ecosystem Experience At the University of Washington, the RFID Ecosystem creates a microcosm for the Internet of Things. The authors developed a suite of Web-based, user-level tools and applications designed to empower users by facilitating their understanding, management, and control of personal RFID data and privacy settings. They deployed these applications in the RFID Ecosystem and conducted a four-week user study to measure trends in adoption and utilization of the tools and applications as well as users’ qualitative reactions. Evan Welbourne, he rapid proliferation of passive cal properties, origin, ownership, and Leilani Battle, Garret Cole, RFID tags in the past decade has sensory context (for example, the tem- Kayla Gould, Kyle Rector, T given rise to various concepts that perature at which a milk carton is being Samuel Raymer, integrate the physical world with the stored). When ubiquitous and available Magdalena Balazinska, virtual one. One of the most popular is in real time, this information can dra- and Gaetano Borriello the Internet of Things (IoT), a vision in matically streamline how we manufac- University of Washington which the Internet extends into our ev- ture, distribute, manage, and recycle eryday lives through a wireless network our goods. It can also transform the of uniquely identifiable objects. Given way we perform everyday activities by numerous predictions that we’ll have giving applications current and detailed hundreds of billions of RFID-tagged ob- knowledge about physical events. This jects at approximately five cents per tag “real-life” context can unlock the door by 2015,1 we’re not only approaching to various business, environmental, such a world, we’re on its doorstep. personal, and social contexts hitherto In this type of RFID system, each inaccessible to Internet applications. physical object is accompanied by a The incredible amount of informa- rich, globally accessible virtual object tion captured by a trillion RFID tags that contains both current and histori- will have a tremendous impact on our cal information on that object’s physi- lives. However, questions remain if we 48 Published by the IEEE Computer Society 1089-7801/09/$25.00 © 2009 IEEE IEEE INTERNET COMPUTING Building the Internet of Things Using RFID are to use RFID in the IoT. How do we transform low-level RFID data into meaningful, high-level information? Can we design and build applica- tions that are truly useful and not just novelties? If so, will their utility outweigh the potential loss of privacy, and how can we help users un- derstand and control their privacy settings? At the University of Washington, we’re exploring these issues first-hand with a building- scale, community-oriented research infrastructure called the RFID Ecosystem (http://rfid.cs.washington.edu). This infrastruc- ture creates a microcosm for the IoT in which we can investigate applications, systems, and social issues that are likely to emerge in a re- alistic, day-to-day setting. We’ve developed a suite of Web-based, user-level tools and appli- cations for the IoT and deployed it in the RFID Ecosystem. We’ve also conducted a four-week user study to investigate patterns of adoption and utilization of our tools and applications as well as users’ subjective reactions. We present Figure 1. The RFID Ecosystem. RFID reader antennas are mounted the results of this study, focusing on tool and on cable trays (upper left) and in custom-built wooden boxes application usage. (lower left). An RFID kiosk (upper right) lets users associate one of three types of tags (lower right) with a personal object. The RFID Ecosystem We built the RFID Ecosystem around an Elec- tronic Product Code (EPC) Class-1 Generation-2 one tag-read event (TRE) per tag per antenna RFID deployment that spans all seven floors of per second, a tuple with the schema (tag ID, our 8,000-square-meter computer science and antenna ID, time). For example, if tag A is engineering building (see Figure 1). The deploy- detected by reader antenna X at time stamp t, ment includes 44 RFID readers (each equipped then the custom reader software will generate with up to four antennas for a total of 161) posi- and send the following TRE to the server: (tag tioned at the building’s entrances, on the stair- A, antenna X, t). Each reader also runs the wells, and throughout the corridors. Readers network time protocol to synchronize its clock run embedded Linux and have wired or wireless with the rest of the system. Gigabit Ethernet over which they report their We store all TRE data in a central SQL RFID data to a central server. Volunteers carry Server database (www.microsoft.com/SQL). This RFID tags as badges and attach tags to personal database also contains metadata about the de- objects. Because most everyday objects aren’t ployment, including each antenna’s latitude yet manufactured with tags embedded, and be- and longitude and a symbolic antenna name cause manufacturer-assigned metadata might (for example, “front entrance,” “4th floor stair- not be personally meaningful, we create the well,” or “Room CSE 405”). We wrote software tag-object association manually. For this pur- to transmit data between the readers and the pose, we created a special kiosk where users can database in Java using Apache’s Multipurpose select an RFID tag, physically attach it to an Infrastructure for Network Applications library object, and create a corresponding association for efficient, secure networking. This software between the tag and that object. implements various privacy policies2,3 and runs All readers in our deployment run custom the Cascadia system4 to support application de- software that processes new RFID data before velopment and execution. Finally, our tools and streaming it to the central server. This soft- applications are entirely Web-based, imple- ware continuously polls the reader hardware mented with the Google Web Toolkit (http:// for newly detected RFID tags and generates code.google.com/webtoolkit/), and hosted with MAY/JUNE 2009 49 Internet of Things Track presents a highly interactive set of menus, ta- bles, and Web forms for creating and managing metadata on a user’s tags and personal objects. The Tag Manager interfaces with the RFID ki- osk so that when users are at the kiosk, they can associate one or more physical tags with an object. For example, a new user can use the Tag Manager at the kiosk to register several per- sonal tags. Later, the same user can access the Creates object: keys Creates place: database laboratory (a) (b) Tag Manager from his laptop to delete objects and review or edit object metadata (such as name, type, image URL, or where the object’s tags were last seen). A second tool, the Place Manager, supports creating and editing high-level location in- formation items, called places.5 A place in the RFID Ecosystem is a set of one or more RFID (c) Creates event: “Evan enters database laboratory with keys” antennas with a label. For example, the two an- tennas in the corridor on either side of a user’s Figure 2. Metadata management tools. (a) The Tag Manager office door might be grouped and labeled “my creates and manages virtual objects to which tags can be bound; office.” The Place Manager displays each RFID (b) the Place Manager groups antennas into places; and (c) Scenic antenna’s location as an icon in a Google Map uses objects and places to specify how to transform a user’s tag- mashup of the RFID Ecosystem deployment. Us- read events (TREs) into higher-level events. ers can create or edit a place by clicking on an- tenna icons to select or deselect antennas and by entering the place label in a text box (see Apache and Tomcat (www.apache.org) on a sep- Figures 2a and 2b). arate server. Once the Tag and Place Managers define metadata that binds tags to objects and anten- User-Level Tools nas to places, respectively, applications and The RFID data our deployment supplies offer other system components can use that data to low-level location information in terms of tags generate higher-level information that’s person- and antennas. TREs — such as (tag A, an- alized and more directly meaningful to users. tenna X, t) or (tag B, antenna Y, t + An additional third tool, Scenic, lets users spec- 1) — are helpful to IoT users only if a middle- ify what higher-level events they would like ware can transform them into more meaning- to have extracted from their TREs (see Figure ful, high-level information about events that 2c). Scenic uses an iconic visual language and applications and users can directly consume a storyboard metaphor to describe how people (for example, Ana is leaving the office with her and objects enact an event through a sequence purse). Furthermore, because such high-level of movements between places. Specifically, the events are personal and potentially sensitive, Scenic interface lets users drag and drop icons users must be able to precisely control all in- representing people, objects, places, and basic formation disclosure to avoid privacy breaches. relationships (such as inside, outside, near, or As such, we’ve developed several secure, Web- far) onto one or more panels, or “scenes,” in a based tools that let users directly control how storyboard to specify a movement sequence their RFID data is transformed and disclosed in corresponding to an event. Thus, to specify an the RFID Ecosystem. event, users simply “tell the event’s story,” scene by scene.