Advances in robotics are accelerating at an unprecedented pace thanks to machine learning algorithms allowing for endless experiments. Tiger Woods may have logged 10,000 hours practicing golf but OpenAI trained a robotic hand to dexterously manipulate a cube running simulations that effectively ran for two centuries.
But there’s another major force impacting robotics: Amazon, of course.
The e-commerce titan is on an ambitious mission to optimize its fulfillment operations around the globe, in turn spawning an entire ecosystem of industrial-robot startups, scientists and engineering students racing to create mechanical grippers with human-like dexterity. Amazon is redefining the warehouse supply chain realm with its fleet of mobile robots, already being deployed, and its pace-setting research into robots that can pick up and place, and which are still in development.
While it’s true that the influential industrial robot maker Re-Think Robotics went out of business during this mini boom, a slew of other companies, such as Soft Robotics, 6 River Systems, Righthand Robotics and Vecna Robotics, have all gained market share and staked claims as new darlings of the online order fulfillment robotics space. RightHand, which in December announced a $23 million B round, has drawn attention for its pick-and-place robotic system. Vecna makes self-driving warehouse "tuggers" that can autonomously transport light- and heavy-material payloads.
This community has sprung up, by way of Boston, Silicon Valley and Pittsburgh, around Amazon and e-commerce generally at a time when warehouse workers are growing more scarce and logistical challenges are getting harder – customers expect their goods shipped fast and inexpensively, if not for free. For Amazon, it’s all hands (all varietals) on deck.
Since introducing robotics in 2012, Amazon has added tens of thousands of its robots to its fulfillment centers while at the same time adding more than 300,000 full-time workers globally, the company said. The Amazon Robotics "drive units" can speedily locate and lift storage pods from vast rows of them. More than 100,000 individual DUs, semi-resembling Roomba vacuums, are deployed in at least 25 of Amazon’s 175 fulfillment centers worldwide, snagging pods and taking them to stations for either picking or stowing via homo sapiens.
Repetitive simulation training done in hundred-year batches will most certainly be needed by fulfillment warehouse robot pioneers because it remains a given that machines can’t touch humans, not when it comes to sifting "thousands, sometimes millions, of different products, all with varying weights, sizes, and shapes," as IT veteran Pablo Valerio has said regarding Amazon's uniquely daunting workflow challenge.
Six years ago, when Amazon acquired North Reading, Mass.-based robot maker Kiva Systems, the $775 million deal triggered a flurry of e-commerce-mindful robotics startups, according to CB Insights. Amazon's decision to repurpose Kiva into Amazon Robotics – and keep its activities strictly proprietary – opened the door for other industrial robotics companies to fill the void.
Meanwhile, between 2015-2017, Amazon lit a fuse in the robotics space by annually staging four-day grasping contests in Japan in which teams of researchers from around the world built autonomous bots capable of object perception and motion/grasp/task planning so as to be able to mimic the duties of an average fulfillment center sorter reaching for usual items from a prototypical pod in a typical Amazon warehouse. Amazon collected research, industry members came together to throw down and new milestones were met. One of Amazon’s new patents is for a multi-tooled robotic hand manipulator capable of handling both light and heavy loads, CB Insights said.
The Amazon Robotics Challenge (first prize $80,000, in-roads to Amazon team, priceless) has since discontinued. Why? Perhaps the point had been made that if you wow Amazon the end game is that they get a taste of your IP and not everyone exists to hitch their cart to Amazon’s self-driving train, to paraphrase what one Boston-area robotics analyst surmised.
But make no mistake Amazon’s push to automate picking stations – to do more in the field of dexterous manipulation than anyone thought possible – continues unabated. Amazon is rumored to be working on its own human-dexterity-replicating pick-and-place system, according to a TechCrunch report.
Brad Porter, VP & DE, Amazon Robotics, made a point to emphasize that for all of the advancements that Amazon has achieved on its own, and with all that it spurred on in industrial robotics at large, the fact remains that human employees are firmly baked into the company’s future plans.
“We regularly look at our operations and evaluate how we can bring technology to create new solutions for employees,” Porter said. “When it comes to using robotic manipulation for item picking – while we’re encouraged by the work in the research community – the simple fact is the current state of the art is not capable of handling the diversity of Amazon’s product selection.”
Humans are, and have an added benefit: they notice things.
“If there’s a jug of laundry detergent that is leaking, the human, not the robot, can make sure we don’t ship that leaking jug to a customer,” Porter said.
It’s not a secret that many of the jobs at fulfillment centers can be less than fulfilling. Stress levels at Amazon FCs has received some negative attention, even a “South Park” skewering. This past fall, employees in Germany and Spain walked off the job in protest.
Amazon knows and must care if only in the context of being efficient i.e. humans are still a big part of the equation. Its workforce is already a mix of humans and machines working side by side and that trend will continue to evolve. The company has heavily investment in robotics research and development to support this envisioned hybrid, or “centaur,” workplace of the future; Amazon’s spending on R&D is pretty much unrivaled – Amazon, in addition to revolutionizing the fulfillment supply chain, has eyes on numerous other lofty prizes, such as data analytics dominance, that is, being able to know with even more certitude what makes us tick so as to sell us more stuff – although the exact spend is not publicly specified in corporate filings. The R&D outlay totaled as much as $22.6 billion in 2017, Bloomberg has estimated. That’s $6 billion more than the second biggest R&D spender, Google. It’s roughly the size of Iceland’s GDP.
Another clue underscoring a mega-R&D budget is that recent gigantic balance sheet offset maneuver (probably worth hundreds of millions of dollars) connected with unused backdated federal R&D tax credits that Amazon seems to have just cashed in, accounting-wise, according to reports.
Robotics is a major part of that spend. Among the company’s highest paid employees are engineers experimenting with new tech to underpin mobile bots, industrial bots, drones and Jetsons-esque collaborative robots, all mostly technologies still just a twinkle in Jeff Bezos’ eye.
Here is an illustrative Amazon job posting on LinkedIn, a position supervising a lab at the company’s R&D Center outside Seattle: “Amazon is seeking a proven technology leader to lead development and testing of advanced automation technologies at Amazon's R&D Center. The R&D Tech Program Manager supports Robotics Engineering in the development and testing of Robotics/Mechatronics systems, Packaging Automation, Autonomous Vehicles and Sortation systems.”
While hundreds of employees work at Amazon Robotics, the entire robotics unit is even larger and spread around Boston, Seattle and Berlin.
Grasping advancements might be shutting off the lights at certain Japanese warehouses (an all-robot workforce can work in the dark) but it bears repeating amidst the robotics surge that Amazon can’t hire people fast enough. Last year's 14% increase to employee headcount (up to 647,500) could have been even higher. During one much-publicized “Jobs Day,” Amazon set a goal of harvesting 50,000 job applications, the company received only 20,000, according to CB Insights.
“The use of robotics, automation and technology in our fulfillment centers is enhancing our workplace and making jobs more efficient,” Amazon’s Porter said. “We are excited to continue increasing the technology we use at our sites while growing our global workforce,” he added. “We need advanced technology and automation to meet customer demand—it’s just that simple.”
“Amazon has definitely been a major factor in both the need for automation and the rapid advancement of the automation provider market,” said Matthew Cherewka, Business Development and Solutions Design Manager at Vecna Robotics in Cambridge, Mass. “With the extreme labor shortages hitting the supply chain, manufacturing, and industrial industries, if companies want to remain competitive, they have to invest in these technologies.”
Collaborative robots, or “cobots,” are among the fastest growing sub-species of robots, according to the Robotics Industry Association.
“I’m certain we will be seeing a lot more robots within the warehouse, and I suspect many of them will be collaborative, such as picking arms and self-driving tuggers,” Cherewka said.
“With Amazon, it’s not just warehouse cobots or mobile robots or bots that can lift heavy palettes, it’s all of the above and more,” said Jeff Burnstein, RIA president. “They’ll keep innovating with one thing in mind, to get products to their customers as fast as possible.”
Because of the vast array of stuff that Amazon handles, versus a more narrow lift for other smaller more specialized warehouse operations, Amazon needs to bang its own drum and march to that idiosyncratic beat, even as all sorts of other related picking and grasping breakthroughs are happening in various pockets of industrial settings.
"Warehouse robots rely on lightweight pincers or suction cups," said Ken Goldberg, William S. Floyd Jr. Distinguished Chair in Engineering, the University of California, Berkeley. "These work relatively well for boxes and new software using advances in AI is now emerging that allows them to reliably grasp more complex objects like toys, clothes, and tools."
“Amazon has everything from washing machines to penlight batteries in its warehouses,” said Nell Watson, an Irish engineer who helped pioneer machine-learning-enabled body imaging. “It's challenging to build bots that can cope with a range of situations, or reach right to the bottom of a container.”
Watson and others interviewed for this article pointed out that the company is acutely cognizant of its workplace issues and will look to technology to both get things shipped faster and also function as place where people actually want to work. The recent South Park episode, “Unfulfilled,” about the arrival of an Amazon FC, lampooned the stressful conditions that led European workers to walk out in protest. It’s worth noting here that the 2017 winner of the Amazon picking contest was an Australian sliding mechanism named “Cartman.”
“The stress of being on a constant algorithmic treadmill makes one unable to even realistically take a bathroom break, for fear of being axed,” said one European robotics engineer familiar with the issues raised by workers in Germany. “Human beings are not designed to sustain that level of performance, and this will inevitably lead to burnout.”
Amazon does offer a helping hand to workers who are interested in breaking free from bindom. It has more than 500 hourly associates currently enrolled in robotics related courses including engineering and computer science, as well as nearly 100 associates who have graduated so far in these areas, Porter said.
Among the big challenges for companies like Amazon, alongside keeping workers mentally healthier and creating a bot that pick well, is the question of how design processes and facilities to take advantage of new robotics solutions.
"Traditionally, industrial automation involves a highly-tailored solution to a highly-constrained problem, Vecna's Cherewka. "You identify a specific task in a specific process, spec out exactly what you need a robot to do, and then someone painstakingly perfects it to that, and just that. Even the slightest variability can throw the system out of whack, and any updates require a highly-trained specialist."
However, he said, new advancements are creating robots that can adapt to varying conditions and improve their performance over time. On the picking side, robots can pick a variety of objects and use machine learning to identify new, similar objects they haven’t necessarily seen before to therefore expand their capabilities, Cherewka said. Similarly, in the mobile robot space, "you might have a self-driving fork vehicle that focuses primarily on put-away, but can get pulled out to do an occasional crossdock or other opportunistic task." And all of these new systems are constantly capturing data and are connected to each other, which can be used to improve general product performance over time. Which means organizations can’t just invest in a few robots and call it a day.
"Robots are only a small part of the solution," Cherewka said. "The true impact of these technologies is their impact on the material handling workflow at large. You are no longer automating a specific task and constraining the inputs and outputs – since this new era of automation is flexible and can adapt with you, you now need to actively evaluate and manage both upstream and downstream of that automated process as well. Having technologies that can coordinate the combined efforts of both humans and robots across your entire operations is much more important than the automation equipment itself."
"At the heart of this transformation is technology that can optimize the inherent strengths of both humans and machines and can combine these strengths with real-time data so that their operations can react dynamically to the market, and the solution providers who enable that."
