In today’s robotics circles, there's a consensus that we’re at an inflection point — a forthcoming boom in human productivity in the real-world.

The anticipation in the field isn’t just sparked by phenomenal technical strides — such as a 2,000-fold increase in computational power, plummeting costs of lidar sensors and other hardware, or the emergence of foundation models pioneering new ways of learning with increasing accuracy and robust safety controls. Nor is it because of the fast-growing ecosystem of developer tools empowering small teams to tackle complex tasks with human-like precision once infeasible. These technical leaps have to meet market-wide shifts to fully realize their potential. 

What’s truly unprecedented now is the sheer gravitational pull in market demand for industrial innovation. The American labor shortage has created an insatiable appetite for advanced autonomous systems to increase efficiency across our economy’s most critical industries, like supply chain, manufacturing, and transportation. What I see from the frontlines is a major shift in perception. 

C-suites in physical industries are realizing that relying solely on human labor, alone, can no longer meet customer demand. Over the years, early robotics adopters have experienced a consistent increase in return on investments, attributed to lower bill of material costs, reduced operational expenses, and the capability to deploy robotics in increasingly more complex, high-value applications.

Despite these advances, most industries are predominantly analog: more than 80% of warehouses today have no automation, and only 141 robots per 10,000 manufacturing employees are deployed today. Taken together, all these developments raise crucial questions:

- Why hasn’t robotics proliferated across our most critical industries in dire need of labor?

- What hurdles must be overcome for the widespread adoption of robotics?

- How will robotics affect GDP, and how will its proliferation reshape our society?

My goal with this article is to explore the critical factors shaping the robotics field and track real-time progress. I recently had discussions with many industry pioneers whom I’ve had the privilege to know in our portfolio, including Vikas Enti, Co-Founder and CEO of Reframe Systems and former Amazon Robotics leader who oversaw the integration of Kiva robots at Amazon’s fulfillment centers; Adrian Macneil, Co-Founder and CEO of Foxglove, previous head of infrastructure at Cruise, and Vijay Badrinarayanan, head of the AI org at Wayve, as well as industry veteran Kevin Peterson, former head of perception for Waymo.

The insights reinforce my long-held belief that the robotics revolution is just beginning. The critical factors — mature technology, favorable costs, rising demand, and a broad talent pool — are finally aligning. Previous peaks have always lacked at least one key element necessary for proliferation. Now, the next generation of robotics startups will grow output in ways that was previously not possible, defining a new golden age of economic growth and prosperity.

A Brief Retrospective

Progress, they say, unfolds slowly, and then suddenly all at once. It’s hard to fully appreciate what opportunities lie ahead without first tracing our roots to see how far we’ve come and where we stand today.

Pre-2000’s: The Rise of Modern Robotics

In 2004, as Mark Zuckerberg was creating Facebook in his dorm, Kevin Peterson was giving autonomous cars a way to perceive their surroundings in the real-world. During the DARPA Grand Challenge, his team tackled everything from soldering microcontroller boards to scraping together money for a laser range finder. They installed a two-kilowatt, 100-pound computer in the car, but “it was so buggy, it was almost laughable,” Kevin recalls. These obstacles lit a fire within the scientific research community, prompting researchers to slowly develop new capabilities that could accelerate developer velocity. We began to see organizations like Willow Garage initiate the influential Robot Open Source (ROS) project, addressing widespread inefficiencies in robotics. Through projects like these the ecosystem began to make slow, but steady progress.

The modern robotics era, reminiscent of the early homebrew personal computing movement, began with the ingenuity of hackers, hobbyists, and academics. Initiatives like the DARPA Grand Challenges catalyzed Autonomous Vehicle (AV) technology, echoing the competitions that drove early computer development. In the same vein, Willow Garage played a role analogous to communal hackerspaces, fostering collaboration that spurred significant robotics innovations. Early commercial successes like Kiva Systems' warehouse robots and iRobot's domestic cleaners mirrored the breakthroughs of personal computing firms, showcasing the industry-transforming potential of robotics.

These developments underscore a recurring pattern: groundbreaking technologies frequently begin with enthusiasts in makeshift labs, innovating in their free time.

At this point, robotics wasn't mature or affordable enough at the right price points, so the only applications were primarily for research. Since these early days, the steady march of technical innovation has resulted in dramatic improvement in performance and reduced costs throughout the robotics stack. One of my personal favorites is the mindblowing advancement in lidar technology:

2010’s: Commercial Robotics

For a long time, robotics systems were highly rigid, designed for narrow, single-purpose tasks, mostly used in controlled lab settings. Jokingly referred to as the philosophy degree of engineering sciences, they were intellectually intriguing, but lacking practical application. These systems didn't offer a viable economic case for most dynamic industries. This began to change when Kiva Systems introduced robotics into Amazon’s fulfillment centers.

When I first met Vikas Enti, who led Kiva's integration at Amazon, he clearly understood a crucial insight that eludes many founders: customers value operational efficiency, reliability, and revenue over technical sophistication. Profitability is paramount.

Kiva systems exemplifies the profound understanding of customer needs, engineering a solution with zero technical dogma. Its acquisition by Amazon in 2012 was a roaring triumph. It generated cost savings of over 20% to Amazon’s bottomline — equating to billions of dollars annually. This success spurred a new generation of entrepreneurs. A wave of robotics startups soon followed, which aimed to mitigate research risk and leverage technology to deliver substantial results within the boundaries of what’s achievable.

As I wrote in 2017, these successful companies defined the first blueprints for how robotics companies can enter the market, laying the foundation for what’s to come.

Why Robotics is About to Accelerate

Tech barriers are plummeting, experienced talent is pouring in, and market demand is rising — factors all converging to fuel a productivity boom.

The Tech Trends

While developers have created countless web 2.0 apps, and digital services dominated headlines, the world of robotics has silently built its own impressive tech foundation on these giants’ shoulders. No longer tethered to basic Programmable Logic Controllers (PLCs) or rules-based decision making, today’s robotics tech stack has advanced far beyond what we envisioned just five years ago.

Here’s a snapshot of the tech stack I’m following, poised to empower small team with greater engineering speed and velocity:

- Simulation: Breakthroughs in the gaming industry have advanced our ability to mimic real-world physics with remarkable precision. These simulation technologies allow teams to experiment with a wide-range of robots, capabilities, and environments, helping them quickly identify the best architectures for any challenge. There’s also remarkable progress in neural rendering, a subfield enabling infinite immersion. Notably, large generative world models and neural rendering techniques like NERFs and Gaussian Splatting stand out as trends to closely monitor.

Wayve's "Neural Replay" technique, for example, leveraging NeRFs and Gaussian Splatting, reproduces scenarios for in-depth analysis, enabling testing of rare "what-if" situations to pinpoint error causes. This technology forms the foundation of Wayve's Ghost Gym, a cloud-based log replay tool utilizing cost-effective T4 GPUs for concurring model simulations, showcasing scalability and efficiency.

Bridging the simulation-to-reality gap and establishing a robust evaluation infrastructure are crucial for deploying robotics at scale. As simulations improve, so will our ability to measure and evaluate safety protocols in real-world scenarios. Ultimately, more robust simulation raises the value of engineering hours and accelerates organizational learning rates.

- Foundation Models: Self-supervised learning (SSL) models and Large Language Models (LLMs) are revolutionizing robotics. SSL allows physical robots to learn directly from vast amounts of diverse, unlabeled data, reducing dependence on subjective, human-defined rules (like what constitutes a “chair?”) and detailed annotations (which is impractical for every scenario). LLMs and multimodal capabilities merge computer vision and language, naturally grouping related items and enhancing the understanding of our surrounding world. Each AI era — ImageNet for vision and language for Transformers — is defined by its toughest data challenges. Now, it’s increasingly clear that Embodied AI — driven by video complexity, real-world integration, and safety — will emerge as the next frontier.

The system's beauty lies in its virtuous flywheel: it becomes smarter and more efficient with each quality data point, requiring less data for its next task. With ongoing advancements in foundation models like GAIA-1, LINGO-2, SayCan, RFM-1, RT-X, PaLM-E, and EUREKA, the field is well-positioned to propel progress in planning and control even further.

- User Interface: Novel user-interfaces such as Wayve's LINGO and AI-powered next-gen CAD tools enable model introspection through natural language, improving explainability and decision-making in robotics. Human-Computer Interaction offers substantial potential to shape the coexistence of robots and humans in everyday life.

- Rapid Prototyping HW: Mechanical design iteration speed has soared with the adoption of rapid prototyping tools like 3D printers, alongside affordable, precise machining and forming resources. In the early days of development, iteration speed is everything. Today’s tools enable mechanical and electrical engineering teams to quickly design, build, and test prototype capabilities — lifting the slope of overall developer velocity in the robotic field.

- Compute: NVIDIA's computing advancements have evolved through multiple generations at a mindblowing pace. Their latest THOR SUPERCHIP, is set to deliver over 2,000 TOPS, marking a 2,000x performance leap in just 5 years.

- Lidar/Sensing: Advances in lidar and sensing technology enable smaller teams to accelerate development. Vikas notes, “IMU sensors, once priced over $1,000 for navigation and motion tracking, now rival the tech in everyday devices like iPhones.” Practically every component in modern robotics has experienced dramatic improvement in performance and cost reduction over the past decade.

- Developer Tools and Firmware: Emerging libraries such as TensorFlow, OpenAI Gym, Applied Intuition, and Deepmind's MuJoCo serve as the foundation of modern robotic software infrastructure. Platforms like Foxglove leverage robot action, event, and failure logs to enhance debugging, testing, and predictive modeling, accelerating development cycles.

- Cloud Computing: Low-latency, cloud-based tools now enable more efficient data management and streamlined resolution of remote issues. This eliminates the need for costly on-site technician visits and reduces reliance on cumbersome on-premises servers. That said, there is a long journey ahead to realize the long-term vision of making it possible to spin up embodied agents as easily as spinning up an S3 bucket.

- Batteries: Remarkable recent improvements in energy density and cost reduction have also played a pivotal role in pushing progress forward.

I foresee robotics mirroring the innovation surge of the Web 2.0 and smartphone revolution: decades of prototyping followed by sudden, exponential growth. Just as tools like Ruby on Rails, AWS, and Docker overcame the "cold start" problem, or when iOS and Unity empowered developers to accelerate mobile development, similar advancements in robotics are poised to drive hypergrowth.

Total shipments of smartphones (left); total websites created (right)

The robotics developer ecosystem is ripe for expansion with the establishment of foundational building blocks. Looking ahead, the field urgently requires:

- Evaluation: Development of a unified, industry framework for infrastructure, measurement, and evaluation to enable the deployment of enterprise-ready systems.

- Actuators: Advancing the actuator ecosystem to promote the broad use of mobile robotics across industries and form factors. We need actuators with “modular architectures” that can be quickly configured to address a wide array of robotics shapes and sizes without customization and long lead times common in the actuator industry. Specifically designed for mobile robotics, these actuators should prioritize high torque density, thermal efficiency, and torque sensing to scale effectively and benefit from economies of scale.

- Tactile Sensing: Progress in next-gen tactile sensing is vital for achieving versatile manipulation capabilities.

- Datasets accessibility and curation:  Improved access to comprehensive real-world production controls datasets is essential to streamline development and deployment.Expecting embodied AI models trained on research data alone to achieve production-level performance without including production data in their training distribution is a mistake. Furthermore, the process for ML engineers to efficiently assess gaps in the training data distribution remains more art than science.

We need more robust tools to help engineers rapidly evaluate model performance, identify gaps in training data, and easily source and curate supplemental production training data in order to improve model performance in the real-world.

- Next-gen Comms Protocols: On-bot communication is a bottleneck for innovation. General purpose robotics demand more sensors transmitting data at higher frequencies with larger payloads, driving the necessity for advanced communication protocols beyond CAN.

With each passing year, as these elements grow stronger, the field's infrastructure will become more resilient. Practitioners will rapidly develop and deploy new applications with greater speed and efficiency, fueling the cycle of innovation.

The Talent Pool

Fortunately, a new tide is rising in the workforce. Fatigued with the treadmill of building longtail SaaS products for optimizing ad clicks or payment processing, a growing number of engineers and builders in Silicon Valley are eager to solve more challenging and impactful problems. Adrian Macneil embodies this shift. Tasked with spearheading the infrastructure team at Cruise in 2016, he brought extensive experience in developing large engineering teams at several fintech startups.

Since then, Adrian has been instrumental in standardizing continuous integration processes, which are essential for modern autonomous systems. Leaders like him, with experience from companies like Cruise, Waymo, Kiva Systems, Tesla, and Rivian, are now vital to driving innovation in physical industries.

Traditional robotics once required a PhD from CMU or Stanford to make a significant impact. Today, thanks to increased investment and — crucially — the shift to ML as a core technical solution, any trained engineer working on ML production is relevant to robotics development, resulting in an order of magnitude boost in talent availability. Now, we are also tapping experienced talent from industries that have solved complex engineering challenges and deployed mission-critical systems at scale, such as semiconductors, cloud computing, EVs, consumer hardware, and large-scale AI. In sectors like manufacturing and construction, job postings demanding digital skills—such as programming and data collection—have tripled over the last five years, outpacing even the tech sector’s growth. Many of Eclipse’s portfolio companies are seeing talent conversion rates as competitive as those of leading tech companies like Tesla.

Lightcast and Eclipse People Report

Early talent is on a steady rise, partly due to programs like First Robotics, which nurture young roboticists. Over the past decade, degrees from leading robotics universities have surged by 13x. Unlike Vikas and Kevin's university days, every new graduate now enters the field with some basic ROS experience.

This convergence of emerging talent, big tech transplants, and seasoned domain experts in physical industries is poised to ignite the next industrial revolution. However, there remains a shortage of skilled engineers in disciplines such as mechanical, electrical, embedded software, controls, etc. My hope is that young talent will rediscover their passion for these engineering disciplines, which are crucial to this robotics revolution.

The Markets: Impact on GDP

Multiple factors are fueling market growth. Firstly, labor shortages are not going away. Companies across vital economic sectors such as construction, logistics, and manufacturing just can’t find enough workers. This trend is exacerbated by the aftermath of COVID-19, disruptions in the supply chain, and an aging population.

Turnover Rate = (Number of separations / average annual employment)*100 from Awardco calculations.

Finding workers is just the start; the real struggle lies in keeping them. Currently, 49% of all warehouse workers are leaving their jobs annually, but that’s just the average. I recently spoke with a top-tier third-party provider in the logistics sector who told me they are experiencing a staggering 30% turnover per month.

Turnover expenses can quickly escalate, amounting to millions of dollars for companies with tens or hundreds of thousands of employees. For instance, at a $15 per hour wage, turnover costs can soar up to $15,000 per employee, equivalent to half of a $30,000 annual salary. We are seeing average hourly wages across various industries rising sharply, driven by a competitive labor market that shows no signs of easing.

In response, companies are embracing innovation like never before, and investors are taking notice. Since 2019, over $90 billion has been invested in new robotics startups, with the largest categories in logistics, construction, medical, manufacturing, and agriculture sectors.

Second, when you look at total installations, robotics has surged in electronics, overtaking the traditionally dominant automotive sector. This is interesting because electronics products are typically more complex. Think about a standard smartphone — each can have hundreds to thousands of parts, and people frequently want new versions. Traditional automation has been good at single-purpose tasks that remain static for a long time, like traditional cars. But the rise of robotics in electronics signals that automation is evolving to tackle more complex and flexible tasks. Now, robotics systems that are increasingly software-defined, and therefore re-configurable, are making way for innovative companies like Bright Machines to tackle a wider array of applications without compromising on performance, stability, and safety.

International Federation of Robotics

Robotics companies that leverage human-level AI systems to address productivity and labor challenges have a massive opportunity to drive tangible economic value.

What excites me is that robotics companies prove to be very good businesses: They often secure high-value contracts with substantial long-term value and generous profit margins. With contracts starting at seven-figures — land-and-expanding to eight or nine — it's no surprise that many software entrepreneurs are salivating at the opportunity to build modern robotics businesses. What’s more, robotics offers notable cash management advantages, including high up-front payments from customers and established debt instruments for working capital. Plus, these solutions demonstrate high stickiness and are challenging to replace, resulting in low customer churn. Finally, given how deeply embedded robotics systems need to be in the business processes, the potential for added value through continuous improvements and new features is paramount.

Take Third Wave Automation, which employs autonomous forklifts for pallet manipulation and various tasks. The business case is clear: with roughly 300,000 new forklifts sold annually in North America and an installation base of 2.2 million, the potential for cost savings is significant. At $50,000 per truck per year across two shifts, the company offers substantial savings compared to manual operators, who command an average salary of $45,000 per shift, incur significant training costs, and whose errors can result in substantial workers’ compensation liabilities. Even with just a 1% market penetration, this translates to approximately $1.1 billion in recurring revenue, with a robust margin of around 75%.

When you factor in these cost savings alongside the higher throughput enabled by superior performance compared to humans, the case for customers to adopt these solutions becomes compelling. Such clear-cut opportunities are abundant across various physical industries, and I am confident we will witness the emergence of numerous massive businesses over the next decade. 

But reaching that point requires relentless effort. Throughout robotics' early history, we at Eclipse, alongside our portfolio leaders like Vikas, Kevin, Adrian, and Vijay, have observed that best-in-class businesses have advanced based on at least three core principles:

- Build in the customer environment: Early integration of domain experts is essential for achieving high-quality output and seamless integration. Bringing prototypes to customers after the fact can impede scalability.

- It’s not all-or-nothing: Prioritize the factors that bring value to customers most, such as throughput, safety, system learning, and reducing human intervention, over striving for complete autonomy.

- There are no shortcuts: While product and technical breakthroughs are crucial, success hinges on thorough and persistent execution when transitioning from prototype to large-scale production. It demands a level of dedication akin to "sleeping on the factory floor" — relentlessly testing and troubleshooting — to ensure top-notch quality, reliability, and scalability.

The Path To U.S. Economic Prosperity

Labor is a fundamental variable of the GDP equation, but the U.S. economy increasingly grapples with constraints imposed by labor supply challenges. Today our labor scarcity is crippling our nation’s most critical industrial sectors. Historically, strategies like offshoring once bolstered our productivity. Now, we face the monumental task of reshoring essential manufacturing capabilities in an era marked by labor shortage, supply chain disruptions, trade disputes, and geopolitical tensions — a shift that transcends productivity and touches the very core of our national security. 

To achieve this transition we would need an insurmountable figure of an additional 3.5 million workers. Even if every unemployed American were hired, 2.4 million positions would remain unfilled.

Now, imagine the impact of fully leveraging embodied AI to elevate real-world labor productivity to levels previously seen only in the digital sphere. In such a scenario, the "labor" variable in the GDP equation wouldn't just increase—it would become virtually boundless. Envision a scenario where there is dramatic decrease in the cost structure associated with building homes, transporting goods and people, producing food, and providing healthcare, driven by an essentially unlimited labor supply. Nations such as China and Japan have already acknowledged the potential of this shift, leading industrial automation.

We — both the startup community and the broader industrial sectors — have to step up and rise to this occasion. We need to accelerate our progress toward the rapidly approaching tipping point of embodied AI.

Of course, progress demands vigilance. Concerns about job displacement and disruptions for existing workers are understandable in the face of technological leaps. History has taught us that the transition toward automation, while disruptive, often leads to greater societal wealth — consider how the invention of trains catalyzed urbanization. Or how the decline of livery stables paved the way for interstate commerce.

Within these challenges lie opportunities to navigate the transformative tide. Embracing robotics, training our current workforce to use novel technology, and modernizing our most critical industries with less dependence on global supply chains amidst labor shortages is the pathway to prosperity.

Autonomy holds the promise of elevating the arc of economic growth. Now is the time to seize this opportunity, and fast. Let's continue to build for the well-being of our nation, our economy, and our society as a whole.

With so much at stake, Adrian puts it best: “This is the opportunity of our generation.”

Follow Eclipse on LinkedIn or sign up for our newsletter for the latest on the Industrial Evolution.