For more than a decade, roboticists have worked on systems for the elderly, hoping to extend the amount of time that seniors can live at home and improve their quality of life. Now MIT researchers have built a humanoid robot with a special motion-tracking system and spring-loaded actuators that make it better equipped to deal with household chores. The robot, named Domo, can size up an object by shaking it in its hand and then put it away in a cupboard.
“Demographics are changing, particularly in Japan, Europe, and the U.S.,” says Aaron Edsinger, a lead researcher on the Domo project and a postdoctoral student at MIT’s computer-science and artificial-intelligence lab. “There are a lot of people that are getting older and not a lot of young people to take care of them.”
But developing a multipurpose robot for the elderly hasn’t been easy because the home environment is so unpredictable. Industrial robots, which are widely used in manufacturing, work with parts that come in standard shapes and sizes. Food, however, does not. So a simple task such as putting away groceries can become quite complicated.
Domo takes that variability into account. Instead of preprogramming the robot so that it only knows how to deal with cans and boxes with certain dimensions, Edsinger has Domo size up each item–one at a time–before deciding how it should be stored.
The shelving process begins when a human puts an item in one of the robot’s hands. The robot then determines the object’s dimensions based on grip and video analysis. First, the robot wiggles the object in its hand while video cameras in the robot’s head record the movement. The robot knows how much force it applied with the wiggle, so it knows how much the object it’s holding should move. Using special motion-capture software, Domo finds the object in the video that moves as predicted and assumes it is the item in its hand.
Now that the robot has identified the item to be shelved, Domo must determine its shape and size. If it’s a small object that fits in the robot’s hand, it can determine the object’s size based on its grip. For long objects, the robot must perform more video analysis.
Knowing that the tip of a long object will wiggle quicker than the rest, the software isolates the part of the object moving the fastest and considers it to be the point farthest away from the robot’s hand. Once the robot knows the object’s dimensions, it can determine how best to place it in the cupboard. “If it’s a pack of spaghetti, it will lay it on its side instead of trying to stand it upright,” Edsinger says.
When designing an embedded system choosing which tools to use often comes down to building a custom solution or buying off-the-shelf tools.