Deep learning holds key to boost robot intelligence

Science fiction movies aside, robots are mostly composed of chips and circuits, integrated together by a set of hardware, and are made to function by dedicated algorithms and software. In short, they operate solely as a result of the commands generated by the software and therefore lack the capacity of humans to analyse and respond to various scenarios, generally.

Although robots are getting more agile and automation systems are becoming more complex, developments in robotics and automation are seeing the need to increase their intelligence. Machines in automation are increasingly able to analyse huge amounts of data. They are often able to see, speak, even imitate patterns of human thinking. Researchers at the automation company, European Automation, call this deep learning.

European Automation sees deep learning as a set of algorithms that machines use to model high level abstractions in data. This allows computers to see objects and understand what they are. These calculation are similar to the processes used by the human brain. Deep learning refers to the intelligence that allows machines to independently analyse the data they come across and to extract patterns from this data, whether it's images, videos, human speech, different languages or codes, and then respond to the results of the analysis.

As an example, Darren Halford, a group sales manager at European Automation, revealed a video that shows a six-axis robot "seeing" boxes stacked in an irregular fashion. The robot analyses the shape of each box and chooses which one to pull out while also calculating the unusual angle of the box and figuring out how to adjust for the angle.

Interestingly, the robot in the video is not moving at its full capability. The process could actually be speeded up for greater efficiency. "The video shows the robot working at 60 per cent speed, probably about the same pace as a human," said Halford. "The added benefit of the machine is that it can't hurt itself moving boxes, it doesn't require breaks and even at 75 per cent speed, it is faster than a human."

Halford noted that the original 3D vision technology deployed in the robot was developed by Industrial Perceptions Inc. Intelligent technology such as the robot's ability to understand a stack of boxes is showing up in more and more applications. "The idea of computers seeing and recognizing objects is being used by Amazon in its Flow app. Flow recognises items via their shape, size, colour, box text and general appearance," he said. "In a shop, home, or even out and about, if you hold your smartphone up to an item with the Flow app open, within seconds of the camera seeing it (if the item is recognisable), Flow will identify the item and add it to your Amazon cart."

Thinking out of the box: Unleashing deep learning potential

Halford used the term deep learning to describe any machine intelligence technology that mimics the human brain in its ability to gather data and sort the data into distinguishable forms. This can be the vision data needed to recognise and respond to shapes, or it can be the ability of a computer to recognise cats in YouTube videos. European Automation uses deep learning to analyse customer information. "Our system is able to use extensive customer data to predict what type of equipment our customers need to purchase and when," said Halford. "It most definitely helps us anticipate our client's needs."

As well as using deep learning to anticipate customer needs, European Automation is interested in deep learning because of its potential in the automation sector. "At the moment, there is an emphasis on archiving and analysing as much data as possible to predict future trends. This is called big data," said Halford. "If a customer's plant could perform self-analysis and self-diagnosis, it could predict parts that need replacing, order the correct part automatically, and thus reduce costly downtime. That's super efficiency."

- Rob Spiegel
??DesignNews