At the Georgia Institute of Technology, a robot is said to slide hospital gowns successfully onto people’s arms after it taught itself using simulated instances. The machine, a PR2, depends on the forces it feels, instead of using its eyes, to guide the gowns onto a person’s arm and shoulder. Almost 11,000 simulated instances were analyzed by the PR2 to teach itself how to dress people in just one day. Some instances were spectacular failures while others were flawless. Using these instances, the robot was able to step into the shoes of the human receiving assistance. This helped its neural network to learn how to calculate the forces applied to the person.
Still Several Steps to Take to Enable Robot to Fully Dress a Person, say Researchers
Simulated instances analyzed by the PR2 helped it to anticipate the end results of moving the hospital gown in different ways. Some motions slid the gown effortlessly along the human’s arm. Other movements caused the gown to be stretched tightly against the person’s body. Using these anticipations, the PR2 is able to pick motions that dress the arm comfortably. It attempted to dress people after achieving success in simulations. Wallace H. Coulter Department of Biomedical Engineering associate professor, Charlie Kemp said that the robot is always thinking ahead.
The robot’s timing was modified by the researchers and allowed to think as much as 1/5th of a second ahead of time while it strategized about its following move. According to Kemp, robots could provide more effective, comfortable, and safer assistance by anticipating physical consequences of their actions.