Communications of the ACM
Helping Robots Learn Quickly in New Environments
RoboCLIP dramatically reduces how much data is needed to train robots.
Credit: iStock
An algorithm developed by computer scientists at the University of Southern California (USC) reduces the amount of data necessary to train robots and enable them to perform tasks after a single video demonstration or language description.
The RoboCLIP algorithm performed two to three times better than other imitation learning methods after one video or textual demonstration of a task.
RoboCLIP trains robots using video-language models (VLMs), which are pretrained on large numbers of video and textual demonstrations.
From USC Viterbi School of Engineering
View Full Article
Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA
No entries found