The smart homes coming out of Washington State University have a higher IQ than your average cyber-enhanced abode. Their homes can learn.
WSU’s Center for Advanced Studies in Adaptive Systems (CASAS) is developing smart home technology that harnesses machine learning in an effort to help older people live with greater independence and remain in their homes longer.
Additionally, the technology could give loved ones a better sense of how their aging parents or grandparents are doing, whether they live down the street or across the country.
The WSU researchers, led by electrical engineering and computer science professor Diane Cook, are building a system that can monitor a senior’s activities and then use machine learning to determine specifically what they’re doing, how it compares to their usual behavior and respond as needed.
The demand for innovation that assists older people is growing. The U.S. Census Bureau predicts that the number of Americans age 85 or older will triple from 2010 to 2050 — and there likely won’t be enough nursing facilities or in-home care providers to assist them all.
“We’re going to need some very creative solutions to heath care,” said Maureen Schmitter-Edgecombe, a psychology professor who’s part of the project. “We need to be thinking out of the box to come up with solutions that are going to be helpful for improving the quality of life for our aging population.”
Eight years ago, WSU created CASAS to tackle this sort of challenge.
One of their first projects was recruiting 400 older volunteers and asking them to engage in normal, day-to-day activities in apartments on the Pullman campus that were outfitted with sensors that tracked their movements. They interviewed and gave cognitive tests to the volunteers and began assembling baseline information that correlated health and activity.
From there they built “Smart Home in a Box” — an easy-to-install system of roughly 30 sensors that detect movement, temperature, or doors opening and closing, as well as other hardware to collect and store that information.
Four years ago, the group began installing the equipment in the homes of older volunteers in Seattle and Spokane. Forty residences are now being monitored.
Helen Dennis, age 88, is one of the volunteers.
“I joined for very a selfish reason,” Dennis said. “I wanted to find out as I age how I’m declining.”
Dennis, who lives in a house in a senior community in Spokane, is eager to remain self-sufficient and to spare her children and grandchildren the obligation of taking care of her. So about one-and-a-half years ago, WSU researchers installed sensors that track when she cooks, works on a book documenting her family history, or takes Chloe, her 52-pound standard poodle, out for a walk.
“I don’t know if they know the difference between the dog and me,” Dennis said of the sensors. “If they don’t, they’ll think I’m awfully active.”
This is where the machine learning becomes essential.
One way to analyze the data collected by the sensors is to have a programmer write lots and lots of rules describing different behaviors. Code sophisticated enough to tell the difference between the comings and goings of a large poodle and a person.
Then it must also differentiate between ambiguous actions. Imagine the sensors detect that a resident is on the floor. Did the person fall, or are they doing yoga or cleaning? Has the programmer considered all of these options?
The WSU system doesn’t care. It looks at the pattern of behavior — has it happened before, what time of day did it happen, how long did the event last, what was the nature of the movement — and deduces what’s going on.
We would rather “let the computer design the complex rules,” said Cook, director of CASAS.
To help check their work, WSU psychologists interview the 40 study participants monthly by phone and twice a year in person to correlate what the sensors are detecting and the machines are learning to make sure they’re getting it right.
Combining all of this information, the researchers are trying to come up with algorithms that will take sensor data and assess a variety of health-related conditions, including how well someone is sleeping, the amount they’re socializing and their mood.
The hope is that with early detection of changes in these conditions, a provider could step in early to correct a problem. If a senior started taking new medications and there was a change in her sleep patterns, for example, a doctor could be notified and intervene to fix it.
“We can be more proactive and preventive with our health care,” Schmitter-Edgecombe said.
And preliminary data suggests that while the monitoring systems look at physical activities, they can also be informative about an older person’s cognitive abilities. It appears that when people begin to develop dementia, their routines become more erratic.
One suite of tools the CASAS researchers are developing are prompts to help people with memory loss. If an older person is supposed to take medication with food, the system could detect when they’re having breakfast and alert them it’s time to take their pill.
“We’re trying to make use of the smart environment not just for heath assessment and earlier interventions,” Schmitter-Edgecombe said, “but also for actually assisting people.”
The CASAS team is already licensing aspects of their products for commercial distribution, namely the Smart Home in a Box kit with monitoring software. But the more elaborate tools — the algorithms to accurately recognize complex behaviors and the prompts triggered by specific actions — are still under development.
“It’s a lot to take on, but it’s fun to work on and it’s really compelling,” Cook said, then added with a chuckle, “and I want it to be available to help me.”
