Communications of the ACM
AI Researchers Devise Cheap Data Collection Method to Scale Training Robots
Researchers have created a low-cost method to train robots in an effort to scale the collection of training data and enable widespread adoption.
Credit: VentureBeat
Researchers at Carnegie Mellon University, Facebook AI Research, New York University, and the University of California, Berkeley have created a low-cost method to train robots in an effort to scale the collection of training data and enable widespread adoption.
The Demonstrations Using Assistive Tools (DAT) framework enables the collection of training data from essentially anywhere.
The researchers used reacher-grabber devices on a pole, and a GoPro camera mounted on the pole recorded 1,000 attempts to move objects or complete tasks.
The videos were then used to train a convolutional neural network that was applied to a robotic arm fitted with a camera and a two-finger grasping clamp.
The system was 87.5% successful in pushing objects across a table to a target spot and 62.5% in stacking objects.
From Venture Beat
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found