Chair becomes personalized posture coach.

Researchers in the heartland are developing a device that may achieve what our mothers couldn't--getting us to sit up straight.

The prototype apparatus is an office chair fitted with a pressure-sensitive blanket. A computer analyzes pressure data from the blanket's sensors.

Last week, the chair's developers, Lynne A. Slivovsky and Hong Z. Tan of Purdue University in West Lafayette, Ind., reported that their sensitive chair can distinguish among 10 different sitting positions.

That's a first step toward a posture-monitoring system that can "figure out what the [seated] person is doing currently and what the person's intention is," says Tan. For instance, a teleconferencing system might incorporate signals from the chair to automatically adjust camera angles in accordance with the sitter's body language. The researchers presented their findings Nov. 9 at the 2000 International Mechanical Engineering Congress & Exposition in Orlando, Fla.

To Tan and others in a field known as ubiquitous computing, or ubicomp, the data-gathering chair could become one of countless input devices eventually pervading people's environments (SN: 11/20/99, p. 330). A primary goal of ubicomp researchers is to equip computers with enough smarts to recognize, from body language and other cues, what people are doing and thinking.

The new chair is "the kind of sensor we're going to need in the future smart environment," says Scott Hudson of Carnegie Mellon University in Pittsburgh.

In particular, the chair's type of sensing could prove useful in automobiles, aircraft, and ergonomics labs, Tan says. For instance, auto seats are increasingly being designed to detect occupants' size and weight. With those data, airbags can automatically tailor the force with which they open (SN: 9/26/98, p. 206).

With its dawning awareness, the Purdue chair detects postures by first recording a pressure profile from the sensor data. Using algorithms adapted from face-recognition technology, the computer then compares those calculations with a library of similar data on prerecorded, averaged profiles for specific postures, such as leaning left, right-leg crossed, and slouching.

To help the chair recognize a sitter's fidgeting and other motions between specific postures, Tan and Slivovsky are now adapting voice-recognition software, which normally tracks changes in an audio signal over time.

This development could prove a boon for postural studies, says Peter A. Hancock of the University of Minnesota at Minneapolis St. Paul. Not a static condition, a person's posture changes subtly many times per minute, he notes.

The Purdue chair's performance so far seems almost maternal: It sometimes doesn't recognize good posture. But if a person slouches, the team reports, the chair catches this bad posture with 99.7987 percent accuracy--almost as well as Mom.