Workplace robot learns new tasks overnight

by L&D24 Mar 2016
The Japanese company Fanuc has invented a robot which will spend the night working out how to complete a task. By the time morning has arrived, the robot would have figured out how to do the job as well as if an expert had programed it.

This is a departure from typical industrial robots which usually need to be programed to do something like grab an object.

What's the trick to the robot learning how to do it?

The technique is called 'deep reinforcement learning' which allows the robot to accurately train itself without the help of an expert.

This might work by picking up objects while capturing video footage of the process. Every time it gets the task right or wrong, the robot remembers how the object looked.

This then becomes knowledge that is used to create a deep learning model that controls its actions.

Indeed, this technique has proved to be an effective approach in pattern recognition in recent times.

Shohei Hido, chief research officer at the machine learning company Preferred Networks, said that it has been a very effective advancement. 

“After eight hours or so it gets to 90% accuracy or above, which is almost the same as if an expert were to program it,” Hido was quoted as saying by the MIT Technology Review.

Hido added that one of the major potential benefits of the learning approach is that it can be sped up if multiple robots work in parallel and then share what they have learned.

Indeed, researchers are testing reinforcement learning as a way to simplify and speed up the programing of robots that can complete factory work.

Meanwhile, in another development, a humanoid robot named ‘Nadine’ is being created by researchers in Singapore.

This social robot is designed to express a range of emotions, mimic gestures, play with children, care for dementia patients and could even appear in workplaces in the future.