INVISIBLE COMPUTING

Interactive Workspaces

images; and facilitate group interaction. Figure 1 shows our prototype experimental facility, the iRoom.

Experimental scenario One study, conducted jointly with the Center for Integrated Facilities Engineering (CIFE) in the university’s Department of Civil and Environmental Engineering, looked at enhancing the productivity of meetings for large construction projects and exemplifies the type of work we are striving to support.

Brad Johanson, Terry Winograd, and Armando Fox Stanford University

W

e are rapidly entering a world in which people equip themselves with a small constellation of mobile devices and pass through environments rich in embedded technology. However, we are still a long way from harnessing the power of this technology in a way that seamlessly and invisibly assists users in their day-to-day activities. Bringing pervasive computing to maturity requires innovations in systems integration and human-computer interaction (HCI). To investigate these challenges, Stanford University’s Interactive Workspaces project (http://iwork.stanford. edu/) is exploring team-based collaboration in technology-augmented environments. These workspaces are designed to allow groups of five to 10 people to work together using computing and interaction devices on many scales. Rather than making the environment itself “smart” and able to anticipate user needs, this project integrates an array of computing appliances that provide access to simulations or data and lets social protocols determine which tools team members use at any given time. Our experimental hardware and software testbeds include • large, interactive, high-resolution wall-mounted and tabletop displays; • specialized I/O devices such as ceiling-mounted tabletop scanners, pan-and-tilt cameras, and wireless LCD displays; and

A cross-disciplinary project is exploring new ways for people to collaborate in technologyrich spaces.

• personal mobile computing devices such as laptops and PDAs connected through a wireless LAN. Building on previous research in multimodal interaction, computer-supported cooperative work, scientific visualization, pervasive computing, and distributed systems architectures, we have experimented with a number of work and educational scenarios in conjunction with research groups at the university and from around the world. Application projects include construction project management, interactive learning, and product design.

INTERACTIVE WORKSPACES Our current focus is on augmenting a dedicated space—a meeting room rather than an individual’s office or a teleconnected set of spaces—to support task-oriented collaboration such as brainstorming sessions and design reviews. A higher-level operating system that links together the various devices and applications in the room makes it easy to add new display and input devices; create, modify, share, and display all sorts of data and

These meetings involve the coordination of many people—the project owner, the architects, various building contractors—who each have large quantities of information in various forms such as maps, construction drawings, project plans, and resource-planning spreadsheets. In these meetings, the parties must reach agreement on how to modify the construction activities to cope with the inevitable delays, changes, and problems that arise. Studies show that participants spend most of their time searching for relevant information when using traditional paper-based materials and consume the next largest portion of time trying to make this data understandable and accessible to the others. As a result, relatively little time remains to explore alternatives and come up with solutions. In the CIFE interactive workspace, users can share electronic information from individual laptops, manipulate it on large shared screens, and cross-link it to provide multiple interrelated views such as a 3D model linked to a site map and project plan elements. The interactive tools let them quickly April 2003

99

Invisible Computing

Figure 1. Interactive room. The iRoom features rear-projection blackboard-like displays with touch-sensitive SmartBoards, a bottom-projected table display, and a PC cluster hosting various I/O devices.

locate and share relevant data so that they can focus on the creative aspects of problem solving.

Software infrastructure

100

Design goals We identified three main design goals related to systems integration: • multiplatform support and integration of legacy applications; • application extensibility and portability; and • failure tolerance and minimal administration.

In developing the software infrastructure for this and other scenarios, we recognized that monolithic applications requiring a specific pattern of user interaction are generally not realistic in an interactive setting. Rather, just as a craftsman chooses from a collection of tools and materials and combines them as needed for a given task, collaborators prefer to run applications pertinent to their own domain on one or more of the devices in the workspace. As the interaction proceeds, these applications cooperate by updating shared data and appropriately responding to changes in one another.

Underlying all of these goals is the desire to provide a software infrastructure that places a minimal set of requirements on devices and applications interacting in the workspace. The system also must manage tradeoffs between flexibility and efficiency, especially for events that require short latency to provide effective interaction.

SYSTEMS INTEGRATION

Event Heap

Given our emphasis on tools, the key systems challenge in designing an interactive workspace is the dynamic integration of mobile and permanent computing devices, various applications, and data drawn from a number of sources including individual laptops, the Web, and online databases.

We addressed the issue of integration primarily through our design of a centralized event exchange system for workspace devices. The Event Heap is a derivative of the tuplespace model, originally proposed by David Gelernter and Nicholas Carriero, and is similar to JavaSpaces and TSpaces.

Computer

Applications post events, which are tuples of attributes, to the Event Heap server and can request events that match a pattern of attribute values. Controlled persistence on the server provides robustness in the face of transient failures of any room components. The Event Heap lets participating applications run under Windows, Macintosh OS X, and Linux with C++, Java, and Python client libraries. The Event Heap uses a simple set of communication primitives that can provide broadcast, remote procedure call, and various other communication patterns. Providing a uniform coordination mechanism facilitates support of different platforms and integration of legacy applications. In addition, by consolidating the matching and routing logic on the Event Heap server machine, we can keep client libraries simple and easy to port as well as reduce the administration needed on each client device. Finally, the Event Heap spatially and temporally decouples applications— they always communicate through the server and, because events persist for a short time, need not run simultaneously. This indirect communication provides extensibility by allowing intermediary applications to intercept and translate events—a process called interposition—so that applications not designed to directly communicate with one another can do so. Applications can likewise snoop on a “conversation” in progress and react to it by monitoring communications through the server. For example, a user could set up applications on a laptop to shadow changes on the interactive workspace screens by reacting to the same events causing those changes. The spatial level of indirection provided by the system also enhances failure tolerance because indirectly communicating applications tend to be programmed to be less interdependent, thereby minimizing the chance that failure in one application will impact others. In addition, the temporal indirection helps mask transient failures.

HUMAN-COMPUTER INTERACTION For many years, HCI design has focused on a single user working at a PC or workstation outfitted with a display, pointing device, and keyboard. Collaboration among users is accomplished over the network via e-mail, shared files, or groupware. The few existing integrated multidevice computer environments tend to be highly specialized and based on applicationspecific software. In contrast, interactive workspaces are designed to enhance the free-flowing collaborative activity characteristic of traditional work settings, where participants flexibly and quickly draw on information from paper, models, whiteboards, and other physical material as well as computing devices. This approach raises several HCI issues.

Distraction-free interfaces Unlike users working alone, participants in collaborative activity are distracted when they have to pause to interact with their machines. To facilitate brainstorming in an interactive workspace, we developed the PostBrainstorm system, a low-attention interface that simplifies computerrelated activities as much as possible. For example, we mounted a digital camera in the ceiling to photograph a tabletop region bounded by physical crop marks. To scan a document, participants need only place it within this region and issue a simple command that captures the image and automatically makes it available for use on a digital whiteboard.

Multiperson interaction To address the need for multiperson interaction, we have created simple cross-device, cross-platform mechanisms such as PointRight that let any user in the room control the keyboard and pointer input of the common displays as if they were a large virtual desktop. Meeting participants can use a dedicated wireless mouse and keyboard or their own laptop’s pointing device. Multiple persons can simulta-

neously control the public screens as long as they do not control the same machine at the same time. In addition, multibrowsing lets users move Web pages and application files among the various displays and machines in the workspace. For example, users can browse for content on their laptops and export it to the public displays, or they can import information from the public displays to their laptops for private perusal. In the case of both PointRight and multibrowsing, we chose not to establish fine-grained floor control but instead to let social protocols dictate operational control as in normal physical settings. As we move our investigations to multiple linked rooms, we plan to maintain our design philosophy of minimizing coordination overhead and preserving the awareness and intimacy of face-to-face interaction as much as possible.

Multidevice interaction Unlike traditional interactions with a PC or workstation that serves as a portal to a virtual world, workspace interactions involve devices that are part of a collaborative effort in a real physical environment. For example, iStuff is a set of wireless I/O devices we have developed—including buttons, sliders, speakers, and lights—that users conceptualize as belonging to the workspace rather than to particular computers. We have implemented simple mechanisms to associate actions on any iStuff device with an intended activity. For example, the Workspace Navigator captures a record of meeting activity by incorporating electronic activities such as browsing and opening files with periodic overview pictures of the whole workspace. The tool’s browser includes a timeline of these pictures; participants can use their visual memory to find and reuse materials from a previous session by choosing a picture and then clicking on the appropriate computer. We modified the Workspace Navigator server to enable users to add an annotation to the time-

line tagged with their name by simply pressing on their iStuff button. This makes it easier for participants to search for important moments in the timeline. ince the formation of the Interactive Workspaces project three years ago, we have deployed and tested our software, which is open source and available for downloading from our Web site, in a variety of environments. Although we have made progress in addressing several systems integration and HCI issues, many challenges lie ahead. One major goal is to extend our mechanisms to integrate multiple workspaces in a unified way, along with mobile remote interaction. We want to maintain the simplicity and robustness of the current infrastructure while dealing with the problems of security and synchrony that arise in more distributed environments. Another challenge is to expand the range of devices integrated into the workspace, including multimodal input using voice and vision as well as sensors that can respond to changes in the location of people and objects. Finally, we plan to develop new techniques that will let a group of technologically savvy users set up new types of collaboration using applicationsized building blocks in just a few minutes before a session, obviating the need for developers to program at the system level. ■

S

Brad Johanson is a postdoctoral student working in the Stanford University Computer Science Department. Contact him at bjohanso@graphics. stanford.edu. Terry Winograd is a professor in the Stanford University Computer Science Department. Contact him at winograd@ cs.stanford.edu. Armando Fox is an assistant professor in the Stanford University Computer Science Department. Contact him at [email protected]. April 2003

101