MIT researchers have develop a system that allows robots to learn by observing humans, reported MIT News on March 5.
While the most popular approach in training robots is reinforcement learning, which is like a trial-and-error, MIT system uses “linear temporal logic” that enables robots to form “beliefs” for the potential outcome of actions, which is closer to how humans learn, explained Ankit Shah from Interactive Robotics Group:
“The robot is essentially hedging its bets in terms of what’s intended in a task, and takes actions that satisfy its belief, instead of us giving it a clear specification”
The learning system has been tested on the basic task of setting the table. While researchers compiled the data set with various variables, the system has observed humans performing the task, and as a result the automated robotic arm has succeeded.
The end goal of the research project is to create system which would enable anyone to teach a robot, without any programming skills, contrary to current engineering approaches, added Ankit Shah:
“That way, robots won’t have to perform preprogrammed tasks anymore. Factory workers can teach a robot to do multiple complex assembly tasks. Domestic robots can learn how to stack cabinets, load the dishwasher, or set the table from people at home”
Now researchers plan to further enhance the system with the ability to understand verbal instructions and corrections. At the end, automated robots might not require even demonstrations.
As Future Time previously reported, last week Google researchers have presented a robot that taught itself to walk with deep reinforcement learning.
Image credit: MIT News