A few years back, for most people the threat of a robot stealing your job seemed a distant prospect, easily dismissed by the fact it took a bot 25 minutes to fold a towel.
However, some academics are now predicting that by 2040 almost half of current jobs could be automated, with new research demonstrating how robots could eventually replace human workers.
Historically, for a job to be performed by a robot it needed to be to repetitive, such as welding together car parts on a production line. Such roles could be automated because they could be carried out in controlled and predictable environments, where the car parts that needed welding were found in the same place each time.
Now researchers at the Massachusetts Institute of Technology (MIT) have demonstrated how robots could build a far wider range of items, by removing their need to work in such a locked-down environment.
The researchers created software that allowed three mobile robotic arms to assemble a chair from four parts strewn around an arena.
In the past, it would have taken hours for a robotic system to work out how to piece together the parts, due to the complexity of working in an unfamiliar setting and coordinating the actions of multiple bots.
The MIT team devised an algorithm that significantly reduces the planning time, to the point where the bots were able to work together to build the chair in about six-and-a-half minutes.
“We’re really excited about the idea of using robots in more extensive ways in manufacturing,” said Daniela Rus, the Andrew and Erna Viterbi Professor in MIT’s Department of Electrical Engineering and Computer Science, whose group developed the new algorithm. “For this, we need robots that can figure things out for themselves more than current robots do. We see this algorithm as a step in that direction.”
Learning on the job
To be able to approach the flexibility of human workers robots also need to be able to learn.
At UC Berkeley, researchers have devised a bot that can work out how to perform simple tasks within as little as 10 minutes.
BRETT (Berkeley Robot for the Elimination of Tedious Tasks) uses a technique called reinforcement learning to master tasks ranging from putting clothes on a hanger to building a toy plane and screwing a top on a bottle.
BRETT hasn’t been programmed on how to carry out these actions, but instead learns for itself how to carry each job out. The bot attempts to perform each task and is given feedback from an algorithm on how close it is to successfully completing it. This feedback enables BRETT to learn which of its movements are suited to finishing its given task and to gradually improve its performance.
BRETT relies on a deep learning program, which creates “neural nets” in which layers of artificial neurons process overlapping raw sensory data, whether it be sound waves or image pixels. This helps the robot recognise patterns and categories among the data and, in the case of BRETT, learn which movements are better suited to carrying out its assigned task.
“With more data, you can start learning more complex things,” said Professor Pieter Abbeel of UC Berkeley’s Department of Electrical Engineering and Computer Sciences.
“We still have a long way to go before our robots can learn to clean a house or sort laundry, but our initial results indicate that these kinds of deep learning techniques can have a transformative effect in terms of enabling robots to learn complex tasks entirely from scratch. In the next five to 10 years, we may see significant advances in robot learning capabilities through this line of work.”