Converge in Robotics Business Review: Startups Powering New Era of Industrial Robotics

This article by Converge’s James Falkoff originally appeared in Robotics Business Review.

Over the last decade, annual installations of industrial robots have increased more than threefold. Yet, while this figure certainly signals the growing importance of automation to the manufacturing sector, startups and investors see something more to this market than the headline number suggests. This is because the latent demand for manufacturing automation dramatically outstrips what robotics suppliers have so far been able to deliver. In recent years, entrepreneurs and investors have spotted an opportunity to address this mismatch and transform the role of automation by leveraging recent technological advances and novel approaches. The startup and venture capital ethos has permeated the field of industrial robotics as innovators have begun to tackle the technological, economic, and human capital constraints that have historically restricted automation to a small set of manufacturers and manufacturing processes. If successful, these endeavors will dramatically broaden industrial automation’s accessibility and applicability.   

The overall level of automation in manufacturing remains low, illustrated by the fact that 345 million people are employed on factory floors worldwide, yet the installed base of industrial robots is only 2.7 million units. In an attempt to meaningfully move the needle on this statistic, innovators are driving progress in three key areas: high-velocity programming paradigms, sensor-driven autonomy, and mobility. This article aims to survey the new generation of startups pursuing such advances. Their efforts could spark a democratization of manufacturing automation and change the way a sector responsible for one sixth of the global economy operates.

Making Robotics Scalable

The declining cost of hardware is one of the factors making more widespread robotics adoption possible. Low-cost alternatives to traditional robotic hardware OEMs are emerging, many of them from China. And there has been a revolution in the affordability of sensors for robotic perception, like the Intel RealSense depth sensor which is two orders of magnitude cheaper than prior alternatives. However, expansion of the market will be gated so long as one fact remains true: the cost of programming robots is much higher than that of the hardware itself.   

Much of the bottleneck to achieving automation in manufacturing relates to limitations in the current programming model of industrial robotics. Programming is done in languages proprietary to each robotic hardware OEM – languages “straight from the 80s” as one industry executive put it to me. There are a limited number of specialists who are proficient in these languages, and because of the time it takes to program a robot and the rarity of the expertise involved, application development typically costs three times as much as the hardware for a given installation. Ben Gibbs, CEO of Ready Robotics, a startup tackling this issue, told me, “To get explosive growth in this industry, we need a more scalable paradigm for application development.”

Programming Bottleneck

To address the skill barrier, cost, and velocity of robot programming, startups are exploring a variety of models for programming a robot without code, or at the very least without programming languages from the 80s. One approach, taken by Ready Robotics as well as other startups such as ArtiMinds and drag&bot, is to mirror a trend in the broader software development world toward no-code and low-code development platforms as an alternative to coding software applications from scratch. These platforms identify common building blocks and abstract them from code into higher-level representations that can be manipulated and interconnected using graphical, drag-and-drop interfaces. While not a solution for the most complex applications, no-code platforms have brought the development of simple applications within reach of professionals without a lot of software development training while allowing experienced developers to create applications at higher velocity. Ready Robotics, ArtiMinds, and drag&bot translate the core principles of the no-code/low-code movement from the world of web and mobile apps to the world of robotics, where the end application is automation of a repetitive task like unloading parts from a machine or assembling components. Once an application is developed in a no-code platform’s visual interface, the platform ultimately gives instructions to the robot using those programming languages “straight from the 80s,” but it shields the application developer from ever having to work in such languages.

Programming By Demonstration

Another paradigm for programming a robot without writing code is the use of demonstration systems. These systems, employed by startups like Southie Autonomy Works and Wandelbots, rely on input from handheld devices that “teach” the robot what to do as they are guided by a human instructor. A human demonstrates the motions involved in, for example, picking up items and placing them in a box, and the demonstration system translates those motions into the robot’s native code.

Other Approaches

Realtime Robotics offers yet another way to decrease the programming burden associated with industrial robots. The startup eliminates the need for laborious line-by-line programming of motion plans by calculating those plans in real-time based on the robot’s current position and a specified goal position. The real-time nature of the calculation allows the robot to account for dynamic obstacles and facilitate complex multi-robot workcells where the robots must work in tandem to avoid collisions. This is a computationally intensive task made possible by the company’s development of novel hardware purpose-built for the problem of motion planning.  

Some startups want to avoid the notion of having to program a robot at all. In lieu of a programming process, Rapid Robotics has developed templates for dozens of common tasks that require only some configuration via an iPad app before being ready for use in production. Jordan Kretchmer, the company’s CEO, told me, “Our realization was that there was no cost-effective solution for automating the simplest of tasks on the factory floor. What we have done is try to put deployment of common applications on rails as much as possible.” To maximize simplicity, the company sells a pre-integrated solution featuring a low-cost robotic arm and sensors. Yet while Rapid Robotics sells a robotics solution focused on simplicity, until recently its inner workings would have seemed downright futuristic because of the role played by advanced sensing and computer vision in how it automates tasks. This illustrates how new programming paradigms are intersecting with another key trend in industrial robotics: autonomy.    

Introducing Autonomy

Despite their linguistic similarity, the terms automation and autonomy signal very different capabilities. Robotic automation has been going on for decades as robots have been used to perform repetitive tasks in high-volume manufacturing contexts, largely in the automotive and electronics industries. Yet most manufacturing contexts feature some element of unpredictability. Parts may be in different orientations as they come down the assembly line, or the product to be produced may change frequently, as is the case in a variety of high-mix, low-volume manufacturing environments, and which will increasingly be the case due to an overall industry trend toward higher levels of product customization. An autonomous system is one that can find a solution to unpredictable problems without external intervention. A critical focus area for industrial robotics startups is developing the autonomy to handle the dynamic nature of most manufacturing contexts.

The Pursuit of Autonomy

The pursuit of autonomy is in some sense about trying to mimic the brain-eye-hand coordination that allows humans to do dynamic tasks. The eye and the hand are mimicked by collecting sensor data, specifically 2D and 3D vision data and haptic data (touch). Improvements in the affordability and power of off-the-shelf sensing hardware have been major catalysts for advancements in autonomy, though startups continue to bring their own innovations in sensing, particularly in haptics. RIOS, one of a new breed of startups taking an all-inclusive factory automation-as-a-service approach, saw haptics as an essential ingredient when it set out to build a vertically integrated hardware-software stack in support of autonomy. The company developed a proprietary haptic intelligence platform powered by next-generation tactile sensors and smart end-effectors that facilitates objectives like the ability to handle delicate or deformable objects and to detect and correct slippage. Forcen is a startup that has developed a paper-thin force sensing film to support similar goals around robot dexterity.

Machine Learning

The software “brain” that turns sensor data into decisions completes the brain-eye-hand analogy. Various branches of artificial intelligence are employed to translate robots’ perception into the actions needed to complete a task. The much-touted approach of deep learning is one tool to accomplish this, but not always the best tool for the job, which might instead be an alternative approach to machine learning or an algorithm from the separate branch of AI known as automated planning and scheduling. Startups like RIOS and Vicarious leverage AI algorithms to solve difficult problems such as grasping unfamiliar objects in random orientations or adapting to different conveyor belt speeds. These companies are addressing the things that have historically precluded automation, like unstructured environments and large SKU counts. Blending autonomy with no-code programming, Micropsi Industries combines AI with the aforementioned approach of demonstration systems, allowing humans to “show” a robot how to deal with variance.

Pain Points

Processes that require hard-to-find skilled labor are common targets for startups building autonomous robotics offerings as they look to address the most acute pain points of manufacturers. Painting, sanding, and welding fall into this category, and startups pursuing automation of these processes include Graymatter Robotics, Omnirobotic, Path Robotics, and Scalable Robotics. An aerospace manufacturer with whom I spoke who is using one of these startups’ offerings said, “Historically robotics has not been a fit for environments like ours, where we have hundreds of product designs but only produce a small number of each design. We are now able to rethink the relevance of automation to these high-mix, low-volume environments.”

Manipulation and Mobility

Implicitly, all the examples given so far have centered on one category of manufacturing robotics, which I would classify under the umbrella term manipulation. Manipulation involves the use of robotic arms, the bases of which are stationary, but which have freedom to move into any position and orientation within reach of the base, giving them precision and flexibility. Most startups are focused on making robotic arms smarter, easier to program, or both. Depending on the application, arms are outfitted with a variety of end effectors, which are like the “hands” that allow a robot to interact with its environment. End effectors are an area of innovation unto themselves, with startups like Soft Robotics developing new, highly dexterous hardware for the task of grasping.

While manipulation is the classical paradigm for manufacturing robotics, it is complemented by the emerging field of mobility. Mobility involves the use of wheeled robots referred to as AMRs (autonomous mobile robots) to move materials between stages of the manufacturing process. AMRs are useful in many industries and have seen particularly strong adoption in warehousing due to tremendous growth in the need for e-commerce order fulfillment. They are not at the same level of adoption in manufacturing but are making inroads. Startups focused on AMRs in a manufacturing context include Arculus, OTTO Motors, R-Go Robotics, and Waypoint Robotics.

As their name suggests, AMRs have the ability to navigate autonomously using technology similar to a self-driving car. I spoke with Amir Bousani, CEO of R-Go Robotics, which provides an artificial perception solution for AMRs, about the state of innovation activity around mobility. He told me, “Mobile robots today lack true, intelligent autonomy. They can’t “see” or maneuver in the sophisticated ways that humans do. The frontier of innovation is in enabling human-level understanding of complex and dynamic environments. For example, enabling a robot to recognize an approaching forklift and take a path that avoids a collision while still moving efficiently toward its destination.”This dynamic environmental intelligence can give AMRs far greater flexibility than more traditional automation approaches like conveyor belts because it facilitates quick adaptation to new routes and factory floor layouts. With this flexibility, some startups advocate for new approaches to organizing production. Rather than a linear assembly line, Arculus proposes a modular layout in which AMRs shuttle materials through a grid of loosely coupled stations, scheduling routes in real time based on capacity and other considerations. Such an approach can better adapt to changing production requirements and demand for product customization.

Cutting edge work is being done to merge the categories of manipulation and mobility. One may soon see AMRs outfitted with robotic arms so that manipulation capabilities can be relocated on demand. This would unlock a whole new level of flexible manufacturing.

Rethinking Supply Chains

The advancements in manufacturing automation being pursued by the startups discussed in this article have the potential to increase the efficiency and adaptability of a crucial sector of the economy. But they are also particularly timely because they come at a moment when manufacturers are rethinking their supply chains and, in doing so, encountering obstacles that automation can address. Manufacturers heavily dependent on China were already thinking about diversifying their supply chains after Sino-US trade tensions flared up and resulted in a series of escalating import tariffs over the course of 2018-2019. Then COVID-19 hit, and disruptions to the operations of ports, warehouses, and other potential bottlenecks in the movement of goods illuminated the risks embedded in long global supply chains like never before.

As a consequence of these events, supply chain resiliency has shot up on the agenda, including efforts to bring manufacturing back to local markets where it has been heavily outsourced like the US. But bringing manufacturing back to the US involves challenges that may only be addressable through improvements in automation. While labor is only one part of the cost equation, the labor cost differential across global manufacturing centers cannot be ignored. Manufacturing labor cost is about $5 an hour in China (and cheaper in other markets such as Mexico and Vietnam), compared to $27 in the US. The benefits of onshoring will always be weighed against the competitive pressures that favor locating manufacturing in low-cost geographies.

Even more of a challenge than the cost of labor, however, is its availability, as mature markets like the US, Japan, and Europe face a manufacturing labor shortage. The shortage is a structural one, meaning it is not resolved by cyclical events like the current economic downturn. In the quarterly ‘Manufacturers’ Outlook Survey’ conducted by the National Association of Manufacturers, the inability to attract and retain talent was the top concern in 11 of the past 13 quarters, surpassing concerns about the weakened economy and other challenges. Deloitte estimates that 2.4 million US manufacturing job openings could go unfilled in the decade to 2028. The culprits are many but interrelated. One potential culprit is a negative perception of manufacturing jobs among young people, whom the sector needs to attract to replace the many retirees in a workforce that skews older than the overall US labor force (which is itself aging). Another culprit is a mismatch between where it is economical to locate a factory and the dense urban centers where people are increasingly clustering. And some would point to an underinvestment in training, which has culminated in a skills shortage in the sector.

Automation has the potential to make American manufacturing viable for more firms by addressing the issues of cost competitiveness and labor availability. Rapid Robotics’ CEO, Jordan Kretchmer, told me he sees this on the ground every day with customers. “Automation,” he said, “is allowing the US manufacturers we work with to take on new business that they couldn’t have before.” As manufacturers mull new strategies in the wake of the trade war and the pandemic, automation will figure heavily in their decision processes.

A Potent Economic Force

With implications for how much it costs to produce goods, how quickly manufacturers can adapt to changing demand, and where manufacturing is located, the wave of innovation occurring in industrial robotics holds the promise of being a potent economic force. There is now a critical mass of startups propelling this trend through activities across the key themes of high-velocity programming paradigms, sensor-driven autonomy, and mobility. The next few years will be exciting ones for the industry as new technologies and approaches become more widely adopted. And they will serve as a good demonstration of how digital innovations increasingly have consequences for even those industries rooted in our physical world.