Researchers at the University of Southern California have developed a method to train robots to predict
When working with people, the robot must constantlyguess what the person will do next. For example, if the robot thinks the human will need a screwdriver to assemble the next part, it can pick it up ahead of time so the human doesn't have to wait. In this way, the robot can help people finish assembly much faster.
Heramb Nemlekar, study co-author
The researchers found similarities in howa particular person collects different products. For example, if he starts with the hardest part when assembling an IKEA sofa, it is highly likely that the same approach will be used when assembling a crib, the scientists explain.
So instead of "showing" the robot how to perform a complex task, they created a small assembly task (the "canonical" task) that humans can doIt is a combination of parts of a simple model airplane, such as wings, tailand a propeller.
The robot “observed” how the person performed the task,using a camera positioned directly above the build area. To detect parts being moved by humans, the system used AprilTags, which are similar to QR codes attached to the parts. The robot then used machine learning to determine the person's preferences based on the sequence of actions performed on the canonical task.
Robot helps man assemble airplane model after training
The analysis showed that on the basis of such trainingthe system was able to predict the actions people would take when performing other tasks with an accuracy of about 82%. The researchers note that the classical approach requires training the robot to solve each specific task. The ability to predict user actions by analogy will make it possible to create universal devices that will be effective assistants in production and for people with disabilities.
Read more:
Scientists have figured out the nature of strange radio signals from a planet similar to Earth
Experts have predicted how many people will live on Earth by 2100
New type of black hole found lurking in Earth's 'cosmic backyard'