Rise of the Robots

Chapter 1

THE AUTOMATION WAVE

A warehouse worker approaches a stack of boxes. The boxes are of varying shapes, sizes, and colors, and they are stacked in a somewhat haphazard way.

Imagine for a moment that you can see inside the brain of the worker tasked with moving the boxes, and consider the complexity of the problem that needs to be solved.

Many of the boxes are a standard brown color and are pressed tightly against each other, making the edges difficult to perceive. Where precisely does one box end and the next begin? In other cases, there are gaps and misalignments. Some boxes are rotated so that one edge juts out. At the top of the pile, a small box rests at an angle in the space between two larger boxes. Most of the boxes are plain brown or white cardboard, but some are emblazoned with company logos, and a few are full-color retail boxes intended to be displayed on store shelves.

The human brain is, of course, capable of making sense of all this complicated visual information almost instantaneously. The worker easily perceives the dimensions and orientation of each box, and seems to know instinctively that he must begin by moving the boxes at the top of the stack and how to move the boxes in a sequence that won’t destabilize the rest of the pile.

This is exactly the type of visual perception challenge that the human brain has evolved to overcome. That the worker succeeds in moving the boxes would be completely unremarkable—were it not for the fact that, in this case, the worker is a robot. To be more precise, it is a snake-like robotic arm, its head consisting of a suction-powered gripper. The robot is slower to comprehend than a human would be. It peers at the boxes, adjusts its gaze slightly, ponders some more, and then finally lunges forward and grapples a box from the top of the pile.* The sluggishness, however, results almost entirely from the staggering complexity of the computation required to perform this seemingly simple task. If there is one thing the history of information technology teaches, it is that this robot is going to very soon get a major speed upgrade.

Indeed, engineers at Industrial Perception, Inc., the Silicon Valley start-up company that designed and built the robot, believe the machine will ultimately be able to move a box every second. That compares with a human worker’s maximum rate of a box roughly every six seconds.¹ Needless to say, the robot can work continuously; it will never get tired or suffer a back injury—and it will certainly never file a worker’s compensation claim.

Industrial Perception’s robot is remarkable because its capability sits at the nexus of visual perception, spatial computation, and dexterity. In other words, it is invading the final frontier of machine automation, where it will compete for the few relatively routine, manual jobs that are still available to human workers.

Robots in factories are, of course, nothing new. They have become indispensable in virtually every sector of manufacturing, from automobiles to semiconductors. Electric-car company Tesla’s new plant in Fremont, California, uses 160 highly flexible industrial robots to assemble about 400 cars per week. As a new-car chassis arrives at the next position in the assembly line, multiple robots descend on it and operate in coordination. The machines are able to autonomously swap the tools wielded by their robotic arms in order to complete a variety of tasks. The same robot, for example, installs the seats, retools itself, and then applies adhesive and drops the windshield into place.² According to the International Federation of Robotics, global shipments of industrial robots increased by more than 60 percent between 2000 and 2012, with total sales of about $28 billion in 2012. By far the fastest-growing market is China, where robot installations grew at about 25 percent per year between 2005 and 2012.³

While industrial robots offer an unrivaled combination of speed, precision, and brute strength, they are, for the most part, blind actors in a tightly choreographed performance. They rely primarily on precise timing and positioning. In the minority of cases where robots have machine vision capability, they can typically see in just two dimensions and only in controlled lighting conditions. They might, for example, be able to select parts from a flat surface, but an inability to perceive depth in their field of view results in a low tolerance for environments that are to any meaningful degree unpredictable. The result is that a number of routine factory jobs have been left for people. Very often these are jobs that involve filling the gaps between the machines, or they are at the end points of the production process. Examples might include choosing parts from a bin and then feeding them into the next machine, or loading and unloading the trucks that move products to and from the factory.

The technology that powers the Industrial Perception robot’s ability to see in three dimensions offers a case study in the ways that cross-fertilization can drive bursts of innovation in unexpected areas. It might be argued that the robot’s eyes can trace their origin to November 2006, when Nintendo introduced its Wii video game console. Nintendo’s machine included an entirely new type of game controller: a wireless wand that incorporated an inexpensive device called an accelerometer. The accelerometer was able to detect motion in three dimensions and then output a data stream that could be interpreted by the game console. Video games could now be controlled through body movements and gestures. The result was a dramatically different game experience. Nintendo’s innovation smashed the stereotype of the nerdy kid glued to a monitor and a joystick, and opened a new frontier for games as active exercise.

It also demanded a competitive response from the other major players in the video game industry. Sony Corporation, makers of the PlayStation, elected to essentially copy Nintendo’s design and introduced its own motion-detecting wand. Microsoft, however, aimed to leapfrog Nintendo and come up with something entirely new. The Kinect add-on to the Xbox 360 game console eliminated the need for a controller wand entirely. To accomplish this, Microsoft built a webcam-like device that incorporates three-dimensional machine vision capability based in part on imaging technology created at a small Israeli company called PrimeSense. The Kinect sees in three dimensions by using what is, in essence, sonar at the speed of light: it shoots an infrared beam at the people and objects in a room and then calculates their distance by measuring the time required for the reflected light to reach its infrared sensor. Players could now interact with the Xbox game console simply by gesturing and moving in view of the Kinect’s camera.

The truly revolutionary thing about the Kinect was its price. Sophisticated machine vision technology—which might previously have cost tens or even hundreds of thousands of dollars and required bulky equipment—was now available in a compact and lightweight consumer device priced at $150. Researchers working in robotics instantly realized the potential for the Kinect technology to transform their field. Within weeks of the product’s introduction, both university-based engineering teams and do-it-yourself innovators had hacked into the Kinect and posted YouTube videos of robots that were now able to see in three dimensions.⁴ Industrial Perception likewise decided to base its vision system on the technology that powers the Kinect, and the result is an affordable machine that is rapidly approaching a nearly human-level ability to perceive and interact with its environment while dealing with the kind of uncertainty that characterizes the real world.

A Versatile Robotic Worker

Industrial Perception’s robot is a highly specialized machine focused specifically on moving boxes with maximum efficiency. Boston-based Rethink Robotics has taken a different track with Baxter, a lightweight humanoid manufacturing robot that can easily be trained to perform a variety of repetitive tasks. Rethink was founded by Rodney Brooks, one of the world’s foremost robotics researchers at MIT and a co-founder of iRobot, the company that makes the Roomba automated vacuum cleaner as well as military robots used to defuse bombs in Iraq and Afghanistan. Baxter, which costs significantly less than a year’s wages for a typical US manufacturing worker, is essentially a scaled-down industrial robot that is designed to operate safely in close proximity to people.

In contrast to industrial robots, which require complex and expensive programming, Baxter can be trained simply by moving its arms through the required motions. If a facility uses multiple robots, one Baxter can be trained and then the knowledge can be propagated to the others simply by plugging in a USB device. The robot can be adapted to a variety of tasks, including light assembly work, transferring parts between conveyer belts, packing products into retail packaging, or tending machines used in metal fabrication. Baxter is particularly talented at packing finished products into shipping boxes. K’NEX, a toy construction set manufacturer located in Hat-field, Pennsylvania, found that Baxter’s ability to pack its products tightly allowed the company to use 20–40 percent fewer boxes.⁵ Rethink’s robot also has two-dimensional machine vision capability powered by cameras on both wrists and can pick up parts and even perform basic quality-control inspections.

The Coming Explosion in Robotics

While Baxter and Industrial Perception’s box-moving robot are dramatically different machines, they are both built on the same fundamental software platform. ROS—or Robot Operating System—was originally conceived at Stanford University’s Artificial Intelligence Laboratory and then developed into a full-fledged robotics platform by Willow Garage, Inc., a small company that designs and manufactures programmable robots that are used primarily by researchers at universities. ROS is similar to operating systems like Microsoft Windows, Macintosh OS, or Google’s Android but is geared specifically toward making robots easy to program and control. Because ROS is free and also open source—meaning that software developers can easily modify and enhance it—it is rapidly becoming the standard software platform for robotics development.

The history of computing shows pretty clearly that once a standard operating system, together with inexpensive and easy-to-use programming tools, becomes available, an explosion of application software is likely to follow. This has been the case with personal computer software and, more recently, with iPhone, iPad, and Android apps. Indeed, these platforms are now so saturated with application software that it can be genuinely difficult to conceive of an idea that hasn’t already been implemented.

It’s a good bet that the field of robotics is poised to follow a similar path; we are, in all likelihood, at the leading edge of an explosive wave of innovation that will ultimately produce robots geared toward nearly every conceivable commercial, industrial, and consumer task. That explosion will be powered by the availability of standardized software and hardware building blocks that will make it a relatively simple matter to assemble new designs without the need to reinvent the wheel. Just as the Kinect made machine vision affordable, other hardware components—such as robotic arms—will see their costs driven down as robots begin scaling up to high-volume production. As of 2013, there were already thousands of software components available to work with ROS, and development platforms were cheap enough to allow nearly anyone to start designing new robotics applications. Willow Garage, for example, sells a complete mobile robot kit called TurtleBot that includes Kinect-powered machine vision for about $1,200. After inflation is taken into account, that’s far less than what an inexpensive personal computer and monitor cost in the early 1990s, when Microsoft Windows was in the early stages of producing its own software explosion.

When I visited the RoboBusiness conference and tradeshow in Santa Clara, California, in October 2013, it was clear that the robotics industry had already started gearing up for the coming explosion. Companies of all sizes were on hand to showcase robots designed to perform precision manufacturing, transport medical supplies between departments in large hospitals, or autonomously operate heavy equipment for agriculture and mining. There was a personal robot named “Budgee” capable of carrying up to fifty pounds of stuff around the house or at the store. A variety of educational robots focused on everything from encouraging technical creativity to assisting children with autism or learning disabilities. At the Rethink Robotics booth, Baxter had received Halloween training and was grasping small boxes of candy and then dropping them into pumpkin-shaped trick-or-treat buckets. There were also companies marketing components like motors, sensors, vision systems, electronic controllers, and the specialized software used to construct robots. Silicon Valley start-up Grabit Inc. demonstrated an innovative electroadhesion-powered gripper that allows robots to pick up, carry, and place nearly anything simply by employing a controlled electrostatic charge. To round things out, a global law firm with a specialized robotics practice was on hand to help employers navigate the complexities of labor, employment, and safety regulations when robots are brought in to replace, or work in close proximity to, people.

One of the most remarkable sights at the tradeshow was in the aisles—which were populated by a mix of human attendees and dozens of remote-presence robots provided by Suitable Technologies, Inc. These robots, consisting of a flat screen and camera mounted on a mobile pedestal, allowed remote participants to visit tradeshow booths, view demonstrations, ask questions, and otherwise interact normally with other participants. Suitable Technologies offered remote presence at the tradeshow for a minimal fee, allowing visitors from outside the San Francisco Bay area to avoid thousands of dollars in travel costs. After a few minutes, the robots—each with a human face displayed on its screen—did not seem at all out of place as they prowled between booths and engaged other attendees in conversation.

Manufacturing Jobs and Factory Reshoring

In a September 2013 article, Stephanie Clifford of the New York Times told the story of Parkdale Mills, a textile factory in Gaffney, South Carolina. The Parkdale plant employs about 140 people. In 1980, the same level of production would have required more than 2,000 factory workers. Within the Parkdale plant, “only infrequently does a person interrupt the automation, mainly because certain tasks are still cheaper if performed by hand—like moving half-finished yarn between machines on forklifts.”⁶ Completed yarn is conveyed automatically toward packing and shipping machines along pathways attached to the ceiling.

Nonetheless, those 140 factory jobs represent at least a partial reversal of a decades-long decline in manufacturing employment. The US textile industry was decimated in the 1990s as production moved to low-wage countries, especially China, India, and Mexico. About 1.2 million jobs—more than three-quarters of domestic employment in the textile sector—vanished between 1990 and 2012. The last few years, however, have seen a dramatic rebound in production. Between 2009 and 2012, US textile and apparel exports rose by 37 percent to a total of nearly $23 billion.⁷ The turnaround is being driven by automation technology so efficient that it is competitive with even the lowest-wage offshore workers.

Within the manufacturing sector in the United States and other developed countries, the introduction of these sophisticated labor-saving innovations is having a mixed impact on employment. While factories like Parkdale don’t directly create large numbers of manufacturing jobs, they do drive increased employment at suppliers and in peripheral areas like driving the trucks that move raw materials and finished products. While a robot like Baxter can certainly eliminate the jobs of some workers who perform routine tasks, it also helps make US manufacturing more competitive with low-wage countries. Indeed, there is now a significant “reshoring” trend under way, and this is being driven both by the availability of new technology and by rising offshore labor costs, especially in China where typical factory workers saw their pay increase by nearly 20 percent per year between 2005 and 2010. In April 2012, the Boston Consulting Group surveyed American manufacturing executives and found that nearly half of companies with sales exceeding $10 billion were either actively pursuing or considering bringing factories back to the United States.⁸

Factory reshoring dramatically decreases transportation costs and also provides many other advantages. Locating factories in close proximity to both consumer markets and product design centers allows companies to cut production lead times and be far more responsive to their customers. As automation becomes ever more flexible and sophisticated, it’s likely that manufacturers will trend toward offering more customizable products—perhaps, for example, allowing customers to create unique designs or specify hard-to-find clothing sizes through easy-to-use online interfaces. Domestic automated production could then put a finished product into a customer’s hands within days.

There is, however, one important caveat to the reshoring narrative. Even the relatively small number of new factory jobs now being created as a result of reshoring won’t necessarily be around over the long term; as robots continue to get more capable and dexterous and as new technologies like 3D printing come into widespread use, it seems likely that many factories will eventually approach full automation. Manufacturing jobs in the United States currently account for well under 10 percent of total employment. As a result, manufacturing robots and reshoring are likely to have a fairly marginal impact on the overall job market.

The story will be very different in developing countries like China, where employment is far more focused in the manufacturing sector. In fact, advancing technology has already had a dramatic impact on Chinese factory jobs; between 1995 and 2002 China lost about 15 percent of its manufacturing workforce, or about 16 million jobs.⁹ There is strong evidence to suggest that this trend is poised to accelerate. In 2012, Foxconn—the primary contract manufacturer of Apple devices—announced plans to eventually introduce up to a million robots in its factories. Taiwanese company Delta Electronics, Inc., a producer of power adapters, has recently shifted its strategy to focus on low-cost robots for precision electronics assembly. Delta hopes to offer a one-armed assembly robot for about $10,000—less than half the cost of Rethink’s Baxter. European industrial robot manufacturers like ABB Group and Kuka AG are likewise investing heavily in the Chinese market and are currently building local factories to churn out thousands of robots per year.¹⁰

Increased automation is also likely to be driven by the fact that the interest rates paid by large companies in China are kept artificially low as a result of government policy. Loans are often rolled over continuously, so that the principal is never repaid. This makes capital investment extremely attractive even when labor costs are low and has been one of the primary reasons that investment now accounts for nearly half of China’s GDP.¹¹ Many analysts believe that this artificially low cost of capital has caused a great deal of mal-investment throughout China, perhaps most famously the construction of “ghost cities” that appear to be largely unoccupied. By the same token, low capital costs may create a powerful incentive for big companies to invest in expensive automation, even in those cases where it does not necessarily make good business sense to do so.

One of the biggest challenges for a transition to robotic assembly in the Chinese electronics industry will be designing robots that are flexible enough to keep up with rapid product lifecycles. Foxconn, for example, maintains massive facilities where workers live onsite in dormitories. In order to accommodate aggressive production schedules, thousands of workers can be woken in the middle of the night and set immediately to work. That results in an astonishing ability to rapidly ramp up production or adjust to product design changes, but it also puts extreme pressure on workers—as evidenced by the near epidemic of suicides that occurred at Foxconn facilities in 2010. Robots, of course, have the ability to work continuously, and as they become more flexible and easier to train for new tasks, they will become an increasingly attractive alternative to human workers, even when wages are low.

The trend toward increased factory automation in developing countries is by no means limited to China. Clothing and shoe production, for example, continues to be one of the most labor-intensive sectors of manufacturing, and factories have been transitioning from China to even lower-wage countries like Vietnam and Indonesia. In June 2013, athletic-shoe manufacturer Nike announced that rising wages in Indonesia had negatively impacted its quarterly financial numbers. According to the company’s chief financial officer, the long-term solution to that problem is going to be “engineering the labor out of the product.”¹² Increased automation is also seen as a way to deflect criticism regarding the sweatshop-like environments that often exist in third-world garment factories.

The Service Sector: Where the Jobs Are

In the United States and other advanced economies, the major disruption will be in the service sector—which is, after all, where the vast majority of workers are now employed. This trend is already evident in areas like ATMs and self-service checkout lanes, but the next decade is likely to see an explosion of new forms of service sector automation, potentially putting millions of relatively low-wage jobs at risk.

San Francisco start-up company Momentum Machines, Inc., has set out to fully automate the production of gourmet-quality hamburgers. Whereas a fast food worker might toss a frozen patty onto the grill, Momentum Machines’ device shapes burgers from freshly ground meat and then grills them to order—including even the ability to add just the right amount of char while retaining all the juices. The machine, which is capable of producing about 360 hamburgers per hour, also toasts the bun and then slices and adds fresh ingredients like tomatoes, onions, and pickles only after the order is placed. Burgers arrive assembled and ready to serve on a conveyer belt. While most robotics companies take great care to spin a positive tale when it comes to the potential impact on employment, Momentum Machines co-founder Alexandros Vardakostas is very forthright about the company’s objective: “Our device isn’t meant to make employees more efficient,” he said. “It’s meant to completely obviate them.”¹³ * The company estimates that the average fast food restaurant spends about $135,000 per year on wages for employees who produce hamburgers and that the total labor cost for burger production for the US economy is about $9 billion annually.¹⁴ Momentum Machines believes its device will pay for itself in less than a year, and it plans to target not just restaurants but also convenience stores, food trucks, and perhaps even vending machines. The company argues that eliminating labor costs and reducing the amount of space required in kitchens will allow restaurants to spend more on high-quality ingredients, enabling them to offer gourmet hamburgers at fast food prices.

Those burgers might sound very inviting, but they would come at a considerable cost. Millions of people hold low-wage, often part-time, jobs in the fast food and beverage industries. McDonald’s alone employs about 1.8 million workers in 34,000 restaurants worldwide.¹⁵ Historically, low wages, few benefits, and a high turnover rate have helped to make fast food jobs relatively easy to find, and fast food jobs, together with other low-skill positions in retail, have provided a kind of private sector safety net for workers with few other options: these jobs have traditionally offered an income of last resort when no better alternatives are available. In December 2013, the US Bureau of Labor Statistics ranked “combined food preparation and serving workers,” a category that excludes waiters and waitresses in full-service restaurants, as one of the top employment sectors in terms of the number of job openings projected over the course of the decade leading up to 2022—with nearly half a million new jobs and another million openings to replace workers who leave the industry.¹⁶

In the wake of the Great Recession, however, the rules that used to apply to fast food employment are changing rapidly. In 2011, McDonald’s launched a high-profile initiative to hire 50,000 new workers in a single day and received over a million applications—a ratio that made landing a McJob more of a statistical long shot than getting accepted at Harvard. While fast food employment was once dominated by young people looking for a part-time income while in school, the industry now employs far more mature workers who rely on the jobs as their primary income. Nearly 90 percent of fast food workers are twenty or older, and the average age is thirty-five.¹⁷ Many of these older workers have to support families—a nearly impossible task at a median wage of just $8.69 per hour.

The industry’s low wages and nearly complete lack of benefits have drawn intensive criticism. In October 2013, McDonald’s was lambasted after an employee who called the company’s financial help line was advised to apply for food stamps and Medicaid.¹⁸ Indeed, an analysis by the Labor Center at the University of California, Berkeley, found that more than half of the families of fast food workers are enrolled in some type of public assistance program and that the resulting cost to US taxpayers is nearly $7 billion per year.¹⁹

When a spate of protests and ad hoc strikes at fast food restaurants broke out in New York and then spread to more than fifty US cities in the fall of 2013, the Employment Policies Institute, a conservative think tank with close ties to the restaurant and hotel industries, placed a full-page ad in the Wall Street Journal warning that “Robots Could Soon Replace Fast Food Workers Demanding a Higher Minimum Wage.” While the ad was doubtless intended as a scare tactic, the reality is that—as the Momentum Machines device demonstrates—increased automation in the fast food industry is almost certainly inevitable. Given that companies like Foxconn are introducing robots to perform high-precision electronic assembly in China, there is little reason to believe that machines won’t also eventually be serving up burgers, tacos, and lattes across the fast food industry.*