Help robots 'understand' people

More and more robots are designed to serve people, from providing entertainment to patient care. However, the interaction between robots and humans is difficult and slow.

In fact, robots need to recognize and respond to human behavior. This is a big problem because each person has different ways of waving their hands. Therefore, scientists must develop intelligent computer algorithms to train robots to understand the general pattern of motion. Researchers Rui Yan and colleagues at the A * STAR Research Institute in Singapore have developed a new software system that helps robots recognize gestures quickly, accurately, and inexpensively.

Rui Yan and colleagues tested the software by integrating it with the Shape Tape - a special jacket that uses optical fibers and inertial sensors to track bending, twisting of the hands and arms. They programmed the Shape Tape to deliver data 80 times per second on three-dimensional orientations at the shoulders, elbows and wrists, and apply different speed thresholds to detect when gestures begin.

In the experiment, five people were put on the Shape Tape and used it to control the virtual robot through the movement of the arm to make commands such as forward, backward, step faster or slower. The researchers found that the system translated 99.15 percent of gestures. The next step in improving the gesture recognition system is to allow people to control the robot without having to carry any special equipment. Yan and colleagues solve the problem by replacing the Shape Tape with a motion-sensing camera.