On the front line of automation, where robots truly are poised to displace humans from their jobs, some cutting-edge technology is testing the best minds in Japanese industry.

It turns out it is remarkably hard to automate the bento lunchbox.

Making a bento — little portions of rice, fish, meat, pickles and other delicacies packed in a plastic box and sold for a few hundred yen — is currently a miserable job done by hand on grim midnight production lines, so that Tokyo stores can be filled with weather-appropriate lunches by the morning.

“The thing we at Yaskawa Electric are wrestling with now is the convenience store lunch,” explains Akira Kumagae, head of motion control at Japan’s number two maker of factory robots, with a market capitalization of $7bn. “We think maybe this is our next business chance.”

Tofu or vegetables are soft and irregularly shaped. “They’re not necessarily small, but are extremely hard to grasp. For example, to pick up a single bean with chopsticks and put it in — that’s tough even for a human,” says Mr. Kumagae. “There are so many things that robots still can’t do.”

After stagnation in the early 2000s, sales of industrial robots have boomed in the past few years, thanks to Chinese car factories and the successful use of robots in electronic assembly. The International Federation of Robotics forecasts average growth of 13 per cent a year to 2019.

Automators such as ABB, Fanuc and Keyence are stock market darlings, often posting operating margins in excess of 30 per cent. For all the alarm about robots taking over the world, however, that is not what the people who make them anticipate.

Yaskawa and its rivals expect a slow but steady expansion into new types of manufacturing, such as food; new roles in manufacturing, such as quality control; and into adjacent niches of distribution or agriculture.

For Masaki Hojo, chief executive of Daifuku, the cutting-edge question is how to automate the picking of products in an ecommerce warehouse. Daifuku is one of the world’s biggest suppliers of warehouse automation, baggage handling systems for airports and conveyors for vehicle assembly lines.

“We’re now in a world where you need to handle products down to an individual pen,” he says. If every product is placed in a separate box — black pens in one box and blue pens in another — then it is feasible for a robot to pick them. But moving and warehousing so many different containers is inefficient, hence picking is still largely done by hand.

“If you have a box with products X, Y and Z in it, then a robot needs to distinguish one from another. It needs to pick a soft item or a hard item without dropping it,” says Mr. Hojo. Image processing and deep learning may make it possible for robots to do that work — but not yet.

Other challenges for Daifuku include unloading a truck and automatic baggage drops at the airport. As Mr. Hojo points out, a future self-driving truck will not be much use if it needs a driver to unload the contents with a forklift on arrival.

Within manufacturing, a growth area is inspection. Mobile phones may roll off a robot assembly line, only for a team of humans to check them by hand, pressing the on/off switch and making sure everything works.

“Inspection is an extremely tiring job. It takes a lot of concentration,” says Masaru Takeuchi of Omron, which specializes in sensing and control for factory robots. He suggests that image sensors and artificial intelligence could be used to check a mobile phone screen for scratches or marks.

Smaller factories also make little use of robots. “It takes a lot of people, time and money to get robots working. The engineering can cost three to eight times, sometimes even 20 times, more than the hardware,” says Mr. Takeuchi.

That makes sense on a car assembly line or when manufacturing at the scale of the iPhone, but not when producing in small batches to rush deadlines. Complexity explains why robotics is such a high-margin business: factories need a lot of help from the supplier to make their production line work.

“There’s a big difference between what’s technically possible and what you can use on the factory floor. If you have half a year to prepare, and top-class graduate engineers, you can do all sorts of things. For robots to spread, we need that to be one week with no specialists,” says Yasuhiko Hashimoto, head of robotics at Kawasaki Heavy Industries.

Achieving that, he says, will need big improvements in the user interface, in safety and in the ability to train robots in an intuitive way, rather than programming their every motion.

The industry is cheerful about its prospects but nonplussed at the idea it will soon displace humanity. Japan installs 30,000 robots a year, points out Mr. Hashimoto, compared with a decline in the working-age population of 640,000

“If the population of Japan drops to 80m as forecast, which means a fall of about 50m, then automation will be no use at all,” says Mr. Hojo. “You might be able to cover for 2m people with robotics and automation. Not 50 million.”

What robots cannot do

Even more revealing than asking a robotics company what they can do is to ask what they cannot.

“The simplest thing we cannot do is exert large force,” says Akira Kumagae, head of motion control at Yaskawa.

“An Olympic weightlifter can lift several times their own body weight, but for every kilogram you move with a robot, the robot will weigh 10kg.”

Pound for pound, then, even the average human is ten times stronger than a factory robot. “Robots need lots of space and cost lots of money. You need to prepare for them from the start, put in big power supplies,” he says.

Despite advances in computer vision, says Mr. Takeuchi of Omron, the technology still needs to get faster.

“Robots can pick something up and move it very quickly. But distinguishing what an object is and deciding how to move it is still a bottleneck.”

The processing power built into a factory robot is not on a level with big AI systems. Put the robot in a network and you have a security problem.

For Mr. Hashimoto at Kawasaki, one of the challenges is greater agility. Most robots, he points out, are single-handed.

“When we talk about robots we can make the arm but we can’t really make the hand,” he says. “University professors will tell you they can do all kinds of things, but as a maker who actually has to deliver, there’s still a long way to go.”

That lack of agility is linked to one of the biggest practical problems with factory robots: the difficulty of reprogramming them to do something else.

“There was a time when people tried to go to full automation on vehicle assembly lines, but it didn’t work because the vehicle changes. For example, if you wanted the same model in the same color for 10 years you could fully automate it. But it wouldn’t sell,” says Mr. Hojo of Daifuku.

“For most products in the manufacturing industry, 100 per cent automation is impossible.”

Source: Financial Times August 8, 2017 | By Robin Harding