The Economist — © B. Cristofani

# Usines 4.0 I Bots for Basics…

La suite du “grand re-make” avec cet article très intéressant publié dans The Economist, qui remet en perspective la question de l’utilisation des robots dans l’industrie textile.

Après le dossier publié dans le dernier numéro d’Interface sur l’arrivée des premiers robots dans l’univers de la confection (lire le dossier sur le site du R3iLab), cet article de The Economist remet les choses en perspective : l’intégration des robots sur les chaînes de confection est une opération complexe. Les Bots ne permettront vraisemblablement pas, à terme, d’automatiser plus de 25% de la production mondiale.

Ce qu’il faut retenir :

  • Sur les 1,63 millions de robots industriels en service dans le monde en 2015 (la dernière année pour laquelle les chiffres sont disponibles), seuls 1 580 étaient en service dans l’industrie textile.
  • Pour un robot, le textile reste un matériau complexe parce qu‘il est souple, mou et imprévisible dans son comportement. Il pourrait être plus facile d’automatiser les activités d’un créateur de mode que d’automatiser les tâchent de ceux qui cousent des vêtements.
  • Un robot Sewbot de SoftWear peut être 17 fois plus productif qu’un opérateur humain, mais le coût minimum du travail aux États-Unis est plus de 18 fois plus élevé qu’au Bangladesh, sans compter le coût du robot.
  • Le patron de Sewbot lui-même reste prudent : “Certain s’imagine que les robots vont prendre le relais et tout automatiser. Nous pensons que les Sewbots resteront minoritaires, même dans 20 ou 30 ans. et permettront d’automatiser environ 20–25% de l’industrie de l’habillement” .
  • Ceux qui gèrent les chaînes d’approvisionnement pour les détaillants de vêtements sont sceptiques quant au rôle des robots dans l’industrie : “Il y a beaucoup de gens qui ont fait de la semi-automatisation. Mais les usines de confection entièrement automatisées, nous n’en avons pas vu … nous sommes probablement dans des années”, explique Spencer Fung, directeur général de Li & Fung à Hong Kong.

Special Report I Emerging Market (Eng)

Lire la traduction française.

BANGLADESH EXPORTS 60% more ready-made garments than India, a country with over eight times its population. On the busy roads of Dhaka, Bangladesh’s capital, white vans nose through the traffic on “Emergency Export Duty”, according to the ambulance-like letters painted on their sides. The success of this quintessentially labour-intensive industry helped make Bangladesh a lower-middle-income country in 2014, according to the World Bank’s classifications.

But some think that Bangladesh’s garment industry now faces a new problem almost as grave as the traffic: the threat of automation. Robots are already common in other kinds of manufacturing, but still rare in clothes-making. Of the 1.63m industrial robots in operation worldwide in 2015 (the latest year for which figures are available), only 1,580 were in textiles, apparel and leather, says the International Federation of Robotics (IFR). Robots find garment-making so hard because its basic materials are so soft. When fabric is picked up or put down, it loses its form, creasing, crumpling, folding and draping in unpredictable formations. That can make it hard for a robot to keep track of what it is handling and where to apply itself. It might be “easier to automate the activities of a fashion designer than to automate the people who sew clothing”, suggests Michael Chui of the McKinsey Global Institute.

Jian Dai, a robotics professor at King’s College London, once described the formidable feats of robotic dexterity required even to iron a garment, something any teenager is quite able (if not always willing) to do. It requires miniaturised infra-red sensors to find the edge of the fabric, which must then be squeezed between robotic fingertips in an “impactive grip”. The robot also has to maintain the tension and smoothness of the material and align its seams. To manage all this, Mr Dai writes, the robot needs several moving parts linked in a “multiple kinetic chain”.

One firm that is tackling similar challenges is SoftWear Automation, based in Atlanta, Georgia, 8,000 miles from Dhaka. Its Sewbot uses high-speed cameras to keep track of the fabric, vacuum nozzles to pick up and rotate pieces, and rotating balls embedded in a worktop to move the fabric along. “Our technologies enable the micro-manipulation and macro-manipulation of the fabric to mimic what a seamstress could do,” says Palaniswamy Rajan, the company’s CEO. His machines can already make simple items like pillows and bath mats on a commercial scale. Next year the company hopes to offer a T-shirt production line. It says that a single Sewbot operator will be able to produce 1,142 T-shirts in an eight-hour shift, 17 times the number a traditional garment worker could make in that time.

These intriguing advances in Atlanta reflect broader progress in robotic technology. The machines are becoming cheaper, safer, more versatile and easier to instruct, notes Mr Chui. Unlike many existing industrial robots, which are kept in cages, the latest generation are safe enough to be used in crowded workspaces. They can also be easily “programmed”, a word Mr Chui puts in quotation marks, since no coding is required. These pieces of kit are no longer the preserve of high-income countries like Japan or Germany. Of all the industrial robots shipped in 2015, a third ended up in middle-income countries, where they were mostly used in carmaking and electronics, according to IFR. China was the world’s biggest single buyer.

The rising efficiency of robots has made economists question some of their traditional prescriptions for success in development. Work by Simon Kuznets in the 1960s and 1970s suggested that modern economic growth requires moving resources out of agriculture into industry, then out of industry into services. This arc of industrialisation is supposed to carry poor countries into prosperity before eventually turning down as sophisticated services take over.
But what if robots, not people, fill the factories? The McKinsey Global Institute calculates that it would be technically possible (if not necessarily economically sensible) to automate 67% of India’s manufacturing employment. It came up with similar figures for Indonesia and Thailand. If poor countries cannot move enough workers into industry, the benefits of productivity gains in manufacturing cannot spread widely through their economies. Their opportunities for development will be squeezed by automation’s impactive grip.

Indeed, the arc of industrialisation has already changed, according to Dani Rodrik of Harvard University. In today’s emerging economies, industry’s share of employment is peaking at a lower level than it used to do, and at an earlier point in their development. This trend towards premature deindustrialisation is “not good news for developing nations”, he notes.
But Mr Rodrik’s results are not as depressing as they seem. Asia, as he points out, has so far defied premature deindustrialisation. The same, in aggregate, is true of Sub-Saharan Africa. It is chiefly in Latin America that the arc of industrialisation has lost height and reach. This Latin deindustrialisation may reflect the gradual abandonment of “import substitution” after the 1960s, when governments lowered the tariffs protecting local alternatives to foreign industrial goods. It may also reflect the arrival of China as a manufacturing superpower in recent decades. But it probably has little to do with robots, which are no more prevalent in Latin America than elsewhere.

Some researchers have questioned whether the developing world as a whole has deindustrialised. They argue that manufacturing employment became geographically more concentrated after 1990, but no less important. Nobuya Haraguchi of the United Nations Industrial Development Organisation (UNIDO), Charles Fang Chin Cheng of the University of New South Wales and Eveline Smeets, a consultant, have painstakingly pieced together employment data on over 100 developing countries, going back to 1970. They find that the average of each country’s manufacturing-employment ratio has indeed declined since the early 1990s, as Mr Rodrik showed. But when they look at developing countries in aggregate, the share of manufacturing employment is higher than in earlier decades. These results are not as contradictory as they seem. To see why, imagine that the world contained only two countries: Colombia and China. In Colombia, industry accounted for about 30% of employment in 1990, according to the International Labour Organisation. That fell to around 20% in 2015. For China, the opposite was true. On average, then, industry’s share remained roughly the same in both years: around 25%. But in aggregate, industry’s share increased enormously, because a tenth of China’s workforce is a much bigger number than a tenth of Colombia’s.

China’s takeover of manufacturing employment may itself have peaked. The number of Chinese working in industry started falling in 2013, and China’s share of world clothing exports has also stagnated since then. This represents a “historic opportunity” for countries like India, notes Arvind Subramanian, chief economic adviser to India’s government. Countries such as Bangladesh, Indonesia and Vietnam are seizing it, but India itself is lagging. Its handicaps are largely self-imposed. The country’s garment-makers pay high duties to import the man-made fibres that now dominate the industry. Exporters can get a refund, but the procedure is cumbersome. Perhaps it could be automated.

As things stand, regulatory barriers are far more damaging for South Asia’s garment-makers than automation. Indeed, the practical people who orchestrate supply chains for clothing retailers are somewhat sceptical about the role of robots in the industry.

“There are many people who have done semi-automation. But fully automated garment factories, we have not seen any…we’re probably years away,” says Spencer Fung, chief executive of Li & Fung in Hong Kong. The company’s chairman, William Fung, agrees. E-commerce may have transformed retailing, but “the supply chain that supplies this highly digitised consumer market is actually analogue.”

Automation can speed things up, but it also adds to costs. The operator of one of SoftWear’s Sewbot lines may be 17 times as productive as a traditional garment-worker, but the typical cost of labour in the United States, even on the minimum wage, is more than 18 times as much as in Bangladesh. And that does not count the cost of the bot.
Bots for basics

SoftWear Automation itself is surprisingly measured in its claims for its technology. “There’s the perception that robots will take over and automate everything,” says Mr Rajan, the firm’s boss; but he believes the sewbots will remain in the minority even 20–30 years into the future. “I expect we’ll probably automate about 20–25% of the apparel industry,” he predicts. Robots will take care of “the high-volume basics”. But “the higher fashion, lower batch sizes are always going to be done by people.”
SoftWear Automation occasionally receives calls from Bangladeshi garment-makers, but the company serves only the American market.

“If you are looking to deploy our technology because you think you can save labour costs, then it’s the wrong reason to do it,” says Mr Rajan.

Instead, his company aims to minimise transport costs, reduce environmental strains and relieve acute American labour shortages. One of their principal customers supplies America’s armed forces, whose uniforms are required by law to be made within the country. This anachronistic legislation is supposed to preserve America’s industrial capacity to make the things its army needs, but “the average age of seamstresses in America is 56,” Mr Rajan points out.
For the foreseeable future, then, the Sewbot is not a threat to the abundant labour in countries like Bangladesh. Its existence owes a lot to some cutting-edge innovation and more than a little to some long-standing American protectionism. Unfortunately, more examples of such protectionism are on the way.

Pour aller plus loin :

Lire l’article ci-dessus en Français.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.