The US state of Massachusetts has felt the effects of industrial disruption in the past, as textile factories flourished, then went South in search of cheaper labour. But it’s now home to a cluster of companies working on a new industrial revolution.

One such is Desktop Metal, founded by Ric Fulop. He was born in Venezuela, and came to America as a young man, where he started a number of companies, from wireless communications to batteries.

His latest venture is now valued at around $1.5bn (£1.1bn). It aims to make it possible to produce any metal part in hours from a computerised design. 3D printing in metal has been possible for years, of course, but it has been expensive and slow. Desktop Metal plans to make it quicker and cheaper than current production techniques, such as milling or casting.

These metal parts would be the essential building blocks of any of today’s most important machines - cars, washing machines, even aircraft.

The company’s first mass production system, which can make up to 300 tonnes of parts a year, is now being delivered to the first customers.

It builds objects from thin layers of metal powder, stuck together with a glue-like binder which is printed on to it using technology similar to office inkjet printers.

These layers form three-dimensional shapes, which are removed from loose powder, heated to burn off the binder - and become solid metal, strong enough to perform the demanding functions of modern machinery.

To show how this could change the things around us, he holds up a piston head, a vital part of a car engine. From the outside, it looks much like any other piston head – a flat cylinder about 10cm across. But the interior, instead of being solid metal, is a curious network of fibrous-looking elements which looks more like something you’d find in a living creature than an automobile.

It’s designed by artificial intelligence software to be the most efficient shape possible, as strong as a solid part but much lighter, therefore cheaper to make, and allowing the engine to burn less fuel.

The Siemens factory in Bad Neustadt, central Germany, resounds to the regular whirring noises of this revolution in full swing.

Its bright, airy halls echo with the sounds of robots making motors for other robots, gyrating in an endless dance in their glass cages.

Parts are sorted and stored in an automated storehouse. Driverless forklifts whoosh up and down the aisles. Finished motors are wrapped by machine.

With more than 30,000 different kinds of motors to make, this factory has to be particularly flexible, so Siemens has made it a showcase for its manufacturing automation technologies - the electronic brains of the factories of the future.

It uses a system called “digital twin” to manage almost every aspect of the factory, from designing parts, figuring out how to build them, to monitoring performance.

Using digital models or “twins” of two robots, they can test how they would work together on a new assembly. A digital twin of a drill bit can shape a block of virtual metal, allowing engineers to find the most efficient way to create a shape, without having to leave their desks.

The factory has to fight for orders and keep costs down to remain competitive - and new efficiencies were becoming difficult to find.

But using digital twin enabled them to shave up to 20% off the cost of making some of their parts, says Siemens’s Peter Zech. And the output of the factory has grown - from 600,000 to 750,000 units per year without expanding the workforce.

The scale of investment in new industrial robots is impressive - sales have more than doubled in the past five years, according to the International Federation of Robotics.

But the range of tasks robots can accomplish is still limited. They are good at repetitive jobs, and never get bored. So in some factory tasks, such as welding identical car chassis in the exact same way day after day, robots beat humans hands down.

But present them with a messy, disordered world, and they are flummoxed - which limits the tasks they can do in factories.

In a basement in central Tokyo, a tech start-up is trying to fix that. Preferred Networks is applying a species of artificial intelligence called deep learning to teach robots how to deal with disordered objects - or things they have never seen before.

The most eye-catching example of their work is a system where two robots tidy a child’s bedroom.

It would be an easy task for even a six-year-old human - given the necessary motivation - but tough for robots.

They have to identify every object, crumpled and strewn in random orientations, figure out how to pick them up, and place them in the appropriate bins or boxes.

The robots work slowly and are easily defeated by things they don’t know. They fail to recognise this reporter’s sock, which is larger and more colourful than the socks it has encountered before.

Nonetheless, the company’s founder and chief executive, Toru Nishikawa, hopes to start selling his tidy-up robots within five years.

And the firm’s investors include Fanuc, the Japanese company whose lemon yellow robot arms toil away in many of the world’s most advanced factories.

Fanuc and Preferred Networks have had a research alliance since 2015, applying deep learning technology to the refinement and improvement of Fanuc’s industrial robots.

Elsewhere in the Preferred Networks demonstration room, a Fanuc robot arm scans a basket of goods, figures out how to pick them up, and deposits them in another basket.

It can pick up items it has never seen before, including our cameraman’s smartphone.

In a factory of the future, this technology could sort metal parts for another robot to assemble, or scan and pack groceries at a checkout.

You leave their offices amazed at the things robots can do, but also struck by how far they have yet to develop to match the abilities of humans. If some of the smartest robots in the world are still bamboozled by a colourful sock, perhaps humans do have a future after all.

Last year Elon Musk, struggling with production delays at his Tesla electric car factory in California, admitted that excessive automation had been a mistake.