INSTALL OF THE MONTH
Aug 1, 2003 12:00 PM, Fred Nicolaus
In recent years, there's been much buzz about the advance of the technologically “intelligent” residence; levels of automation and control that were previously considered sci-fi fantasy are not far from becoming everyday reality. Companies such as Smarthome now offer products that enable a home owner to adjust the temperature of his or her living room from a wireless PDA, most newly built homes are wired for some level of integration, and the future is going to bring only more advancements. Michael Mozer, a professor of computer science at the University of Colorado, has taken a big step forward in this residential technology revolution. For a research project funded by the National Science Foundation, he transformed his own home into an adaptive environment, a system capable of learning and adjusting to the habits of its owners.
The house monitors its inhabitants using a network of more than 75 sensors that are capable of detecting light intensity, temperature, motion, and sound. Those sensors relay the information they collect over a low-voltage network to a PC, which is equipped with software that Mozer and his students designed. In turn this software processes the sensor data and uses it to make decisions about basic comfort settings (HVAC, water heating, lighting) in the house, which are controlled by a network of X10 devices. However, the automation isn't limited to simply turning on a light when an inhabitant walks into a room. The house actually trains itself through a system of reinforcement learning, by which the program is “punished” whenever its inhabitants change a setting; every time someone turns on a light that was off, the software loses points and must take steps to regain them. The system is also programmed to conserve energy, and in its ongoing quest to avoid punishment and save power, the house establishes a stable routine.
Mozer's system is also programmed to try new things. Occasionally, it will turn a light off or lower the temperature a few degrees as an experiment. If the inhabitants don't punish the system by correcting the change, then the program assumes that the new condition is favorable and incorporates it into the daily schedule. In that fashion, Mozer's automated home learns its owners' habits, even as those habits change.
Despite such high-tech results, the actual installation process was fairly minimal. Mozer and a team of graduate students put in the sensor array, wiring network, and control mechanisms themselves. The tricky part of the process was fine-tuning the software so that the house would operate without human interference. If it was set to experiment too much, the constant shifting of temperature and light intensity was exasperating. However, if the system didn't experiment enough, it wouldn't learn quickly. Actually living in the house, Mozer noted, provided an invaluable perspective into what worked and what didn't. “When a light turned on in the other room, it was me who had to get up and turn it off,” he says. “I was personally invested in getting it right.”
Though Mozer's primary interests lie in the computer-science aspect of the project, he observes that the adaptive house offers a solution to one of the major dilemmas in the integrated residential technology market: although a high degree of automation is possible, most consumers don't have the patience and technical know-how required to program their own systems. Of course, professionals can be hired to perform the same function, but that process is often inconvenient and costly. The advantage of the adaptive house, Mozer says, is it programs itself.
As networked houses become more commonplace, Mozer predicts that the applications for adaptive, smart technology will become widespread and more varied. With a greater number of sensors and more intricately integrated appliances, self-teaching automation could be extremely specific and could save a great deal of energy. However, he's doubtful about the possibility of a completely automated home. Though Mozer considered incorporating various media systems into the adaptive house network, he ultimately decided to limit the automation to climate and light control. “People don't have a hard time turning on their favorite TV show,” he says. “It was hard to justify why you'd need to automate something like that.”
Subscribe to S&VC EXTRA!
Breaking industry news in your e-mail inbox every other week! Subscribe at www.svconline.com.