Our robot future is approaching, but there's a lot that still stands in the way of a complete machine takeover.
One issue is in how humans currently communicate with robots; a lot of data and energy goes into a robot recognizing and properly executing real-world tasks, often of the simplest nature.
But, researchers from Nvidia may have come up with a solution: have robots learn by watching humans.
Essentially, the researchers made it so its neural networks can infer what's being done in a video-fed stream of a human performing a task (in the researchers' demonstration, stacking blocks).
The robot then produces a step-by-step description of how it would execute the task, which a human can read to verify the robot will perform the task correctly. Then, the robot carries out the task.
You can watch a full breakdown of how the researchers pulled off their achievement below. This is a promising step forward in how robots and humans will work together, but let's just hope the robots don't get too smart.
Via Engadget
- Driverless cars explained: everything you need to know about the futuristic tech
source http://www.techradar.com/news/researchers-develop-a-way-for-robots-to-learn-by-watching-humans
No comments:
Post a Comment