In 2000, the Massachusetts Institute of Technology launched an ambitious project to transform the way the world uses computers. The old model: a box, a monitor and keyboard. The new: computers as pervasive and invisible as the air we breathe. They called it Project Oxygen.
For an overview of the Project’s goals, and a Q&A with its founders, see “Project Oxygen’s New Wind”.
Now, nearly two years out, the first technologies are rolling out of the labs. Project leaders-Laboratory for Computer Science chief Victor Zue, associate director Anant Agarwal and Artificial Intelligence Laboratory director Rodney Brooks-insist that Project Oxygen is about an idea, not products. But corporate sponsors-among them Hewlett Packard, Nokia and Philips-eagerly await their results. Technology Review went into the labs to get a sneak peak at three facets of Oxygen that show particular promise: Cricket, a location-aware computing system; the Intelligent Room, an high-tech office that doubles as a vision-interface research lab; and the Raw microprocessor, a low-power, ultra-programmable chip designed to power the handheld devices of the 21st century. Together these technologies, their creators say, will put computers everywhere-and nowhere.
A Raw Deal
Handheld computers have come a long way since Apple unveiled its Newton in 1993. Once little more than a glorified Rolodex, handhelds today rival the performance and range of applications of desktop PCs. But higher speeds and multiple, specialized processors have made them power-hungry, and battery life continues to be a limiting factor. To address the power problem, Oxygen researchers, led by Agarwal, are building a more flexible, less power-intensive chip they call the Raw Architecture Workstation, or Raw. “Today, people build custom [chips] for video, graphics, networking and so on,” says Agarwal. “We have a single processor that can do all these things.”
Not only does this optimize performance-especially for tasks like video processing, which bog down in memory-but it saves power, an essential feature for any small, battery-powered device. And the programmability extends not only to integrating discrete functions. It could open up exciting breakthroughs in areas such as software radios, which can easily switch between multiple cellular protocols.
By making the data paths highly programmable, Raw avoids centralized memory and register systems. “In a typical processor you may have to bounce a piece of data around. But with Raw, it goes straight to where I want it to go,” says Agarwal.
The Raw architecture resembles a network of tiles, each containing features for instruction, switch instruction, data memory, logic units, registers and a programmable switch. “We pay a lot of attention to the interconnect, to the wires,” says Agarwal. “If you expose the interconnect to the software you can customize how data flows through the chip. You can orchestrate the flow of data. Now my software can match up the hardware with the application.”
|Oxygen’s handheld computer prototype, the Handy 21, uses a camera (top) to perform face recognition.|
The first device the chip will power will be Oxygen’s model handheld, what they call the Handy 21. Prototype Handys integrate voice recognition, wireless communications and video-power-hungry applications that would benefit from Raw’s all-in-one design. A prototype of the Raw processor, being developed with IBM Microelectronics, is expected to arrive sometime this year.
Cricket Chirps Up
At Project Oxygen, researchers believe a mobile computer can be more helpful if it knows where it is, and what’s around it. Enter the Cricket Indoor Location System, a network of wireless transmitters that provides mobile devices such as Handy 21s with information about their physical location, which they can use to find static devices such as printers or exits as well as other people.
Location-tracking is a hot topic now in light of the Federal Communications Commission’s “Enhanced 911” requirements that call for 95 percent of all cell phones to include automatic location identification technology such as the Global Positioning System by the end of 2005.
The goal, says LCS associate professor Hari Balakrishnan, is to develop an indoor alternative to satellite-based GPS tracking, which rarely works inside buildings and often fails outside near tall buildings.
Inside buildings, multipath and magnetic interference disrupt traditional locational devices. “Getting something to work indoors is particularly challenging,” says Balakrishnan. “The goal for us is to get linear distances of within a few centimeters so you can tell where you are within a foot or so.”
Cricket’s trick is to have each beacon continually transmit two signals: one radio and one ultrasound signals. Because radio zips along at the speed of light and ultrasound pulses travel at the speed of sound, the Cricket software that governs the listening device built into a piece of hardware can calculate the timing difference between the two to determine location. “So if there’s a gap of ten millisecondsthen you’re about ten feet away,” says Balakrishnan.
The low-cost, battery-powered Cricket beacons can be “slapped” on ceilings quickly without calibration, thus making for easy scalability. They’re placed so any listening device can receive signals from three or four devices at once to further localize position. Cricket beacons can also send other information beyond location coordinates, for example, transmitting the identity of key resources in its purview.
The Oxygen team is also working on a “Cricket Compass” prototype that can determine which direction the listening device is facing. By equipping each listening device with several ultrasound receivers placed very closely together, they can compare the minute differences between the reception times, thus determining orientation. This capability could help direct a computer to send information to the nearest facing display, or it could enhance informational and point-of-sale applications. For example, shoppers could point their handheld toward a store display to find out about nearby sales, or museum visitors could download information on a nearby exhibit. Cricket is not wed to a particular radio frequency, and Balakrishnan says they may switch to Bluetooth if the technology takes off. Sensitive to Cricket’s big-brother undertones, researchers are also designing intricate protections for user privacy.
Cricket’s greatest impact may come in embedded systems that track not people in an office, but parts through a warehouse. In fact, Balakrishnan’s group is experimenting with a wired library, in which every book features a radio tag tracked by a Cricket-like system. Better tracking of goods throughout their manufacture and delivery could save billions in theft, loss and inefficiency, while avoiding the privacy worries attendant to the tracking of people.
The Intelligent Room
If, as LCS director Victor Zue suggests, Project Oxygen is a “big playground,” then the Intelligent Room is the cool new jungle gym in the middle. The room hosts a variety of projects exploring new collaborative tools and audio/visual interfaces. For Oxygen, the Artificial Intelligence Laboratory is focusing on voice and vision recognition technologies that will help to shape Oxygen’s Enviro 21, a room-controlling device that lets users interact naturally with the computer.
At first glance, the Intelligent Room looks like a typical meeting room, albeit with a surfeit of computer projected “live board” displays on the wall. You interact with the displays via voice, light pen, gesture, or, if all else fails, a touch panel. The ceiling is studded with an array of 32 microphones, two standard video cameras and two stereoscopic video cameras.
A basic goal is to improve communications between microphones and cameras so that the computer can determine who to pay attention to. The task of identifying speakers is important both for controlling videoconferences and for letting the computer respond to user commands without getting confused. Eventually, such communications, which are orchestrated via Oxygen’s innovative networking software, Metaglue, will also help the computer customize responses for each individual.
“In traditional vision systems you have mono cameras trying to detect objects by extracting the prerecorded background, but changing the lighting fools the camera,” says Krzysztof Gajos, an A.I. Lab research scientist and technical director of the Intelligent Room. “With the stereo cameras, we can not only record the background image, but the background shape. It’s much more robust.”
Oxygen is also interested in what people are looking at, for example to help the computer decide which displays to use for optimal viewing. Software tracks the way a user is looking by combining face-recognition software with the 3D information provided by a stereo camera. To identify orientation, the head-post algorithm keys in on how facial features change during movement. Among other applications, the researchers hope to mount the tracking system on robots to improve navigation.
Researcher Harold Fox demonstrated SAM, an animated computer display that shows different emotions to reveal its state. Instead of prefacing commands by saying “computer,” which can be confusing in meetings, the user just looks at the graphic, and Fox’s prototype knows to listen up. When the user looks away, SAM disengages.
Whether SAM or Cricket or Raw ever find their way into the conference rooms, hallways and handhelds in everyday business is a question that will not be answered for years. But one thing is for certain: the concepts they inspire undoubtedly will.