The model rendered Di Bernardo in two dimensions, the way the camera saw him. In order to graduate to 3-D, the duo used four cameras, decked Gonçalves with Christmas lights, and made a video of him walking around the room. Recalls Di Bernardo, "We'd kick everybody out for the night, move all the furniture, clean up the area, take down the divider, and basically take over the lab."

The walking-around model in its most basic form is a stick figure with a flat, triangular head that looks like a bipedal praying mantis, so they fleshed it out with some off-the-shelf animation software. In either case, the model stands in a box representing the room. You click on the floor wherever you want to step, rather like those learn-to-dance diagrams, and the model walks in your footsteps. Or rather, it plods dispiritedly-not only does it capture Gonçalves's gait, its posture conveys his emotional state as well." Thats exactly how I was feeling," he says."It was three in the morning. I walked back and forth for a couple of hours with those markers." Wondering how much nuance was available, they went back and tried it again. "So the original walk was me dying and walking at the same time, and then another night, I pretended I was happy. It learned the happy walk, too, and you can see the difference." At this point, the duo realized that they had stumbled across an excellent way to create realistic motions for a variety of purposes, and incorporating the model into the whole-body tracking system got shelved in preference to exploring the model. "We still havent figured out the general model for all motions," says Di Bernardo. "We just have models for particular classes of motions." Adds Gonçalves, "But we can apply our algorithms to learn any action we want—to act like certain people, or act happy, or drunk, or whatever." Gonçalves is graduating soon, so the pair are forming a company, called realMOVES, to animate joystick-driven characters for the video-game industry. Response from game developers is enthusiastic, says Gonçalves. "They said they had never seen something that was computer-generated and interactive look so realistic." The duo is off to a good start they shared first place (and won $10,000 in seed money) in the second annual 10K Business Plan Competition, run by Caltech's Industrial Relations Center.

Let's shift our focus to the hand. We often pick up a pen in order to convey our thoughts, so why not let the computer watch as we write? Grad student Mario Munich (MS 94) is taking a real-time look at handwriting. Current systems are touch-based, like palmtop computers or the electronic pads at some stores that allow you to sign for a credit-card purchase electronically. (You'll notice, however, that the clerk still compares your signature to the one on the back of the card.) There are other systems that look at handwriting—such as the zip-code scanners the post office uses—but they work after the fact, looking at writing that's already been written.