Massachusetts Institute of Technology (MIT) researchers have developed a machine learning method that enables robots to acquire new skills using a handful of human examples.
The technique allows a robot to pick up and place objects in never-before-seen random poses, within about 15 minutes.
It involves a Neural Descriptor Field neural network designed to reconstruct the three-dimensional geometry of objects, whose knowledge the system taps to grasp new objects similar to those seen in demonstrations.
The researchers used simulations and a robotic arm to show that the system can manipulate never-before-seen mugs, bowls, and bottles arranged randomly, using just 10 examples.
"Our major contribution is the general ability to much more efficiently provide new skills to robots that need to operate in more unstructured environments where there could be a lot of variability," said MIT's Anthony Simeonov.
From MIT News
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA