YELLING commands at humanoid robots could help them learn more efficiently.
That's according to a new study that focused on a humanoid robot doing menial tasks.
The extra verbal commands were said to make the robot 20 percent more successful.
Researchers from UC Berkeley and Stanford University worked on the study.
It can be found on the online archive and it's titled Yell At Your Robot, Improving On-the-Fly from Language Corrections.
They found that commanding an AI robot to do tasks differently "leads to significant performance improvement in long-horizon, dexterous manipulation tasks without the need for any additional teleoperation."
The dexterous manipulation tasks involved in the experiment included a robot scooping candy and cleaning plates.
Lucy Xiaoyang Shi from Stanford University in California and her colleagues worked on the study.
According to The New Scientist, Shi said she thinks it's important that a way of "naturally interacting" with AI robots and giving quick feedback is created.
The scientist created two AI models and asked a robot to change its method during tasks.
Most read in News Tech
Demands would be given to the robot via a microphone.
They included phrases such as "I want some M&Ms” or “clean the right side”.
Numerous humanoid robots are being worked on in the hope that one day humanoids could help at home or even care for the elderly.
Elon Musk is working on his own Optimus robot that aims to do laborious work for humans.
There's also a bot called Draco 3 that is learning to complete menial tasks and could one day care for the elderly.
Right now, there are several different methods used to train robots.
It's common to input commands or teleoperate a robot from afar.
Verbally communicating with the humanoids could make teaching robots easier.
The researchers do admit that there are some limitations to their work.
Read More on The Sun
"Our system is not well-equipped to handle such non-verbal communication.
"Looking forward, we hope that future research can further enable robots to improve with natural forms of human supervision, ultimately towards empowering anyone to help teach robots," the researchers wrote.