2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
Robot PR-2
Researchers at the University of California, Berkeley have created a robot that learns from its mistakes when performing simple tasks involving fine motor skills. Details can be found on the university's website. Demonstration of the robot's capabilities will take place on May 28 at the ICRA 2015 conference.
A PR2 robot named BRETT (Berkeley Robot for the Elimination of Tedious Tasks) independently tries to complete the task, while taking into account unsuccessful attempts and correcting its movements. With the help of a self-learning algorithm, the robot successfully learned, for example, how to twist a cork on a bottle, assemble a toy airplane, and was even able to hang a hanger on a bar.
The neural network used in training the robot constantly evaluates how much a particular movement has helped it get closer to the goal. Thus, after the successful completion of the task, the robot remembers the most optimal movements and then uses them first of all when solving the task again.
“We have demonstrated a new approach to teaching robots,” explains Professor Pieter Abbeel. "The key point is that we don't have to reprogram the robot every time it encounters something new."
The authors of the project believe that it is self-learning robots that are able to independently navigate in space that will be able to help a person in everyday life and at work in the future.
