We know two things about the future. And both point to the need for many more and very different kinds of robots than now exist.

We know that even as economies increasingly digitalize and become ever more service- and software- centric, demand will still increase for the kinds of things that are produced by the “hard” industries. That reality was made clear during the disruptive lockdowns. Miners are needed to access minerals to supply the manufacturers that, in turn, fabricate physical stuff, from computer chips to medical devices, and from fertilizers to pharmaceuticals. All of that is critical for creating and operating all the services that make modern life possible.

We also know, since “demography is destiny,” that the world in general and the United States and Europe in particular will experience increasing shortages of skilled workers for all of the hard industries.ⁱ In today’s America, the nearly 50-year-old average age of those in the skilled trades is far older than the overall population average.ⁱⁱ

There are only a few options for ensuring a sufficient supply of skilled labor needed for the hard industries.

The primary option, thus far, has been to find those people and industries elsewhere—that is, the de facto policies of increasingly importing goods from other, younger nations. That option, if pushed too far, has its own geopolitical and economic challenges and is likely to face constraints now since many policymakers are embracing reshoring initiatives.

Then there’s the option of importing a younger labor pool. Setting aside the politics of immigration, it still takes a long time for any rising generation to develop the necessary skills and experience, even if the targets (whether natives or immigrants) have the requisite interest in the first place.

Which brings us to the only other option, amplifying the effectiveness of those, of any age, with skills. Industrial automation is a longstanding solution for amplifying labor, whether by outright taking over some jobs to free up a human for higher-skill tasks or by increasing the productivity of the skilled person (faster, safer execution of tasks). However, contrary to the popular narrative, there is far less automation across industries than most imagine, especially when it comes to robots.

Surveys reveal that you won’t find any industrial robots in over 90 percent of America’s manufacturing enterprises.ⁱⁱⁱ

Yes, there are millions of industrial robots in the world, but the majority are found in the minority of businesses and performing a small minority of the universe of tasks. Even for firms of significant size, over 500 employees, just half those have industrial robots. Today’s robot population is found mainly in the big businesses that produce large quantities of similar products (especially automobiles). As the size of the firm shrinks, and the variety of tasks rises, the robot share shrinks faster. The skilled workforce and automation challenges for small business impact large businesses because the latter depend on supply chains of small firms.

That automation yields more output per employee is intuitively obvious and borne out by the data.^iv Fifty years ago, large firms with economies-of-scale achieved, on average, about 25 percent greater output per worker compared to small firms. But today, large- firm adoption of industrial automation has led to per- person output nearly double that of the small ones.

The dramatic automation schism between large and small firms is explained by a simple fact. Robots haven’t yet been good enough to be deployed widely. Industrial robots, in the main, are used in a fixed location performing high-volume, repetitive tasks, well-suited to big manufacturers. Up until recently, there wasn’t any prospect for finding robots that both match or exceed human performance and can also be reassigned to a new task. Both those metrics need to be met with robots that can work (safely) alongside people, instead of isolated and bolted down, or limited to fixed tracks. And, of course, costs matter, not just the purchase price, but the cost of integration which, so far, can double or triple the initial cost.

What hard industries need in the near future distills to affordable, useful anthropomorphic robots, ones with skills. Ones that can perambulate or at least easily roll around in the same environment as people.

Long Time Coming

For centuries, engineers have designed machines with embodied controls that can automatically react to a change. A simple example might be a tank that registers a filling liquid and reacts accordingly: Once the fluid level reaches a certain point, it flips a lever connected to a simple valve, stopping the flow. But far more clever mechanical automation than this dates back to ancient days. In the first century, Hero of Alexandria built automatic doors and the like, powered by compressed air or running water, and even steam. Hero also invented a coin-operated drink- dispenser, as well as animated puppets controlled by ropes connected to weights, amongst dozens of other ingenious, automatic machines.^v

The idea of an automaton itself, a robot, is also old. We can trace the idea of an automaton to a time before Hero of Alexandria, to 250 BC in the epic Greek myth of “Argonautica,” (to become Hollywoodified as “Jason and the Argonauts”) wherein Apollonius imagined a giant, man-like bronze robot called Talos.^vi For the 20 centuries that have followed, robots have been a staple in fiction, usually of the dystopian variety.

When Czech playwright Karel Čapek wrote his 1920 play, “Rossum’s Universal Robots,” he imagined automatons replacing humans for manual labor. Čapek invented the word “robot” from the Czech “robota,” which translates as forced labor or drudgery. Even though the word is now used rather elastically to include everything from an automated pick-and-place machine to a clothes washer, what we really mean by “robot” is a truly autonomous, ambulatory machine, and one that can be anthropomorphic—even human- like in appearance and mechanical function.

In 1939, Westinghouse built a kind of “Wizard of Oz” Tin Man, a stunt robot for the New York World’s Fair. But it could only walk stiffly and had a recorded voice that would say: “My brain is bigger than yours.” (Westinghouse wanted to show off its automated switchgear used for electrical grid controls.) More famously, it was the scientist-turned-writer Isaac Asimov who created the modern archetype for robots in his iconic 1950 science fiction book, I, Robot.

A key feature of an ambulatory general- purpose robot is that it can navigate in the same environments as people. Over recent decades, there have been myriad pretenders in the race to produce a real robot, from Sony’s toy robodog circa 2000 (while looking dog-like, it was not close to being able to emulate animal ambulation) to Honda’s contemporaneous walking and stair-climbing Asimo, to name only two amongst dozens.^vii All were engineering demonstrations or toys. Few could perform functions other than walk or dance awkwardly.

But in the past few years—because of radical innovations in sensors, AI, materials and batteries— engineers are finally building anthropomorphic robots, even if most are not yet commercially viable. With remarkable prescience, for the occasion of the 1964 World’s Fair, Asimov made some forecasts, amongst which he wrote that “robots will neither be common nor very good in 2014, but they will be in existence.”^viii

We know what Asimov was referring to with robots: the difference that distinguishes automation and automatons.

Boston Dynamics human-like Atlas robot meets that definition of an automaton, a true anthropomorphic robot. But it’s not commercially available and is rumored to cost about $1 million.^ix Nonetheless, Atlas has demonstrated human-like running, jumping, back-flipping and autonomous navigation. Atlas’s prowess vastly exceeded the best efforts of the teams that competed only as recently as 2015 in a DARPA (the U.S. Defense Advanced Research Projects Agency) Grand Challenge. The contest involved simply having an untethered robot perform easy tasks of ascending a staircase, opening a door and turning a valve—all tasks inherent to operating usefully within a typical human environment, and all previously beyond the capabilities of any general-purpose robot.

In early 2020, Boston Dynamics offered for commercial sale (base price at $74,500) its autonomous, ambulatory four-legged automaton, Spot.^x That such robots are now being offered for sale is more than mere curiosity. It is a pivot in history comparable to the first automobile, the 1896 horse- carriage-looking Duryea Wagon which was, in its day, magical because it was self-propelled.

Spot can walk, run and get up after falling, open doors and fetch objects. Spot, at least initially, is being hired for such things as roaming safety surveillance at construction sites, on farms, offshore oil platforms, pharmaceutical factories. Such services are critical not just for confirming that equipment is operating optimally but also for safety. Such tasks are by nature inherently repetitive and often become the kind of drudgery that lends itself to error and oversight. And freeing people from those tasks makes them available for upskilling; it amplifies human labor.

There are today at least two dozen companies designing and building pre-commercial anthropomorphic robots, ranging from start-ups to industrial giants such as Toyota and Hyundai. And, last year Elon Musk announced that Tesla would soon commercialize a walking robot called Optimus.

Useful Biomimicry

As with the automobile, the (true) robot is made possible by the confluence of a suite of technologies. For the age of the car to launch, it took the independent maturation of high-strength steel, internal combustion and oil refining. For robots, it’s the arrival of powerful micro-motors, vision “chips,” enabled by AI, and lithium batteries.

Advances in the suite of sensors, vision and location systems have followed a progression similar to the often-noted Moore’s Law for computer chips. Roboticists now have available an array of tiny, powerful cameras, chip-scale radar, complementary laser-based radar (i.e., lidar), along with microscopic silicon-fabricated position sensors. Tools that can sense motion, direction and velocity—from inertial changes in movement, hence the technical term, inertial measurement unit, or IMU—have been used by the military, in particular, for decades. But only in the past two decades has the IMU collapsed from coffee-cup scale to chip-scale, and only in the past decade gained both sufficient precision and affordability.

Practical, tetherless robots also needed a revolution in power, both in the ability to store onboard power and the power of actuators to effect movements and manipulations with precision and, well, power. The first commercial lithium battery, a game-changer, didn’t appear until the early 1990s, and it took another decade or two for that ecosystem to achieve the requisite maturity. Similarly, actuators—in effect, robots’ muscles—followed another, again independent trajectory of (fortuitous) advances in size and more power. Superior designs and new materials—not least the 1984 invention of rare-earth neodymium super- magnets—has engendered a roughly 50-fold gain in the power-to-weight ratio for tiny electric motors over the past several decades.^xi

The challenge that has eluded engineers for years is a mechanical and materials science one: the ability to mimic animal or human muscles. When it comes to biomimicry, the challenge has always been to find a way to use available electrical, pneumatic or polymer actuators to attempt to approach the combination of capabilities exhibited by muscles, the biological actuators: high energy conversion efficiency, a large range of motion, a strong power-to-weight ratio, durability and, ideally, self-repair.

In a 1983 paper titled “The Muscle as an Engine,” American physicist and polymath Edwin Jaynes presciently mapped out the mechanisms and the possibilities that were then not possible.^xii Jaynes observed that ultra-efficient conversion of chemical into mechanical energy would ultimately require emulating how muscles operated—that is, “that the moving parts receiving the primary energy be of molecular size.” He speculated that “far from being impossible,” that in time the design of “useful anti- Carnot molecular engines (artificial muscles) might become about as systematic and well understood as the design of drugs and antibiotics.”

Today we’re beginning to realize Jayne’s vision with the profound, if ignored, revolution in materials sciences. The technical literature is replete with successful designs of “artificial muscles,” some engineered at the molecular level and in some cases with self-healing capability.^xiii It has been a happy coincidence that materials sciences have enabled not only light-weight, durable construction of a frame (the skeleton), but also the design of actuators that have sufficient power- to-weight ratios.

As a key indicator of progress with biomimicry, robots are now, for the first time, able to move—even if most are still pre-commercial—at the same speed as the animals they mimic. Measured in terms of body-lengths-per-second (blps), Boston Dynamics, for example, has demonstrated a robo-Cheetah that approached the 16 blps speed of a biological cheetah. But just as aircraft can do things birds cannot, robots will be able to do the biologically impossible, such as converting in real time from, say, a rolling machine to a walking machine in order to adapt to terrain.

A couple of decades ago, it would have required a room-sized computer to process, in real time, all the data generated by all those actuators and sensors. Of course, not only has compute power increased to allow on-board capability, but high-bandwidth local wireless networks have enabled remote access to even more powerful computation when needed.

Standard engineering progression will soon take us from Spot to the Atlas-class robots for commercial use as technology improves and costs come down. It’s the trajectory seen after every irruption. The emergence of general-purpose robots will echo the pattern of the rise of the general-purpose transportation machine, the automobile. In the world of cyber-physical machines, the timespans between invention and commercial products are remarkably similar across categories and modern history.

Disrupting the Status Quo

It was in 1901 that one of the first cars was offered for sale signaling that commercial viability was possible. It was a Packard with a then-revolutionary steering wheel, instead of a tiller-like control (the design used since the 15-year earlier first invention of a car). And more critically, the Packard demonstrated the impressive feat of reliably completing a five-day, 300- mile drive. It sold for $1,500, which was then about 120 percent of an average annual wage. We note that Spot’s selling price is about 120 percent of today’s average annual wage.

In late 2021, DARPA held its “subterranean challenge” in which teams competed using wheeled, tracked or walking robots that competed to (successfully) perform mining-related tasks in a network of caverns.^xiv One of the contestants, for example, demonstrated the ability for its robots to survey and build out a detailed subterranean map in just one hour, a task that normally entails 100 person-hours of human surveyors to achieve the same precision.

While industrial applications for mobile robots are starting mainly with survey and safety work, a proliferation of vendors has pre-commercial machines capable of working alongside, sometimes replacing, humans in heavy-lift tasks. The warehouse “logistics” markets have become a hotbed for both development and deployment of robots, many to undertake the same kinds of lifting tasks needed across industries. Last year, Boston Dynamics, to note one example, introduced a box-handling robot that can finally match the 800 box-per-hour rate at which humans unload a truck.^xv It can move boxes up to 50 pounds and only needs to take a break every 16 hours (to recharge).

In the coming decade, far more robots are expected to be hired by warehouse operators than in all other applications combined. Within five years, overall spending on automation in warehouses is forecast to be more than double last year’s $16 billion, compared to a 60 percent spending increase over the past five years.^xvi Given the close alignment in tasks and performance metrics, all that commercialization is bound to accelerate robot capabilities for the adjacent industrial market. The population of the total robot workforce in industries and services is expected to increase 400 percent by 2030.^xvii Odds are good that’s an underestimate.

The automaton is a class of machine that holds as much promise for disruption (and for fortunes) as did the advent of the automobile.

The value of an anthropomorphic robot, outside of entertainment, arises from the fact that the utility of such a machine increases the more easily it can operate in the environs that humans normally occupy—as  opposed to specialized environments, such as warehouses or factory-floors designed for most automatons so far. It’s not only about having machines that amplify human capabilities, but also doing so by accommodating humans, rather than forcing humans to accommodate machines.

The automaton is a class of machine that holds as much promise for disruption (and for fortunes) as did the advent of the automobile. And it is a class of machine that, more than any other, has excited the dystopian anxieties of doomsayers, particularly those predicting the destruction of all work as we know it.

The anxieties and complaints are similar in character to those voiced by early critics of the automobile.

Many bemoaned that road infrastructures changed the landscape, that cars disrupted social norms, that they took jobs from ranchers and horse handlers, etc.^xviii Today we find a similar industry of pundits who compete to more loudly decry the consequences of robotification. In his masterpiece Pneumatica, Hero wrote, circa 50 AD, that while some people back then thought his automatons could “supply the most pressing wants of human life,” for others they engendered “alarm.”^xix

Fundamentally, the labor-productivity boost that robotification will bring promises to echo precisely what happened a century ago with the mechanization of industry. More businesses, more services, new kinds of jobs replacing old ones, and more wealth and well-being.

In thinking about our near future, Steffi Paepcke, a senior designer on the robot team at Toyota’s research institute, perceptively observed the modern relevance of the apocryphal quote attributed to Henry Ford: “If the inventors of the automobile had asked people riding horses what they wanted, they would have answered that they just wanted a faster horse. It can be difficult to imagine a future that’s vastly different from the status quo.”^xx

Amen. 

References

ⁱ https://www.barrypopik.com/index.php/new_york_city/entry/demography_ is_destiny

ⁱⁱ https://www.zippia.com/tradesman-jobs/demographics/

ⁱⁱⁱ https://mit-serc.pubpub.org/pub/puzzle-of-missing-robots/release/1

^iv https://mit-serc.pubpub.org/pub/puzzle-of-missing-robots/release/1

^v Woodcroft, Bennet. Pneumatica: The Pneumatics of Hero of Alexandria. New York, NY: Oia Press, 2015.

^vi Mayor, Adrienne. Gods and Robots: Myths, Machines, and Ancient Dreams of Technology. Lawrenceville: Princeton University Press, 2018.

^vii Nocks, Lisa. “500 Years of Humanoid Robots Automata Have Been Around Longer Than You Think.” IEEE Spectrum 54, no. 10 (2017): 18–19. https://doi. org/10.1109/mspec.2017.8048830

^viii Asimov, Isaac. “Visit to the World’s Fair of 2014,” New York Times, August 16, 1964. https://archive.nytimes.com/www.nytimes.com/ books/97/03/23/lifetimes/asi-v-fair.html

^ix “Atlas.” Boston Dynamics. Accessed April 12, 2021. https://www. bostondynamics.com/atlas

^x Mogg, Trevor. “Spot the Robot Dog Is Amazing, and Look How Far It’s Come.” Digital Trends, June 17, 2020. https://www.digitaltrends.com/news/ spot-the-robot-dog-is-amazing-but-look-how-far-its-come/

^xi https://www.mdpi.com/journal/actuators

^xii Jaynes, E.T. Rep. The Muscle as an Engine. Cambridge, 1983.

^xiii Bourzac, Katherine. “A Super-Stretchy Self-Healing Artificial Muscle.” IEEE Spectrum, April 18, 2016. https://spectrum.ieee.org/tech-talk/robotics/robotics- hardware/a-superstretch-selfhealing-artificial-muscle

^xiv https://spectrum.ieee.org/darpa-subterranean-challenge-2657170650

^xv https://spectrum.ieee.org/warehouse-robot

^xvi https://www.statista.com/statistics/1094202/global-warehouse-automation- market-size/

^xvii https://www.electronicproducts.com/mobile-robotics-from-amrs-to- quadrupeds/

^xviii Ladd, Brian. Autophobia: Love and Hate in the Automotive Age. Chicago, IL: Univ. of Chicago Press, 2011.

^xix Crawford, James. “The Life and Death of the Library of Alexandria.” Literary Hub, March 13, 2017. https://lithub.com/the-life-and-death-of-the- library-of-alexandria/

^xx McBurnett, Marie. “Designing Robots for Ikigai.” StackPath, October 1, 2020. https://www.machinedesign.com/markets/robotics/article/21143565/ designing-robots-for-ikigai