Fueling the next golden age of industrial progress isn’t a moonshot dream, but an urgent directive. 

The United States has needed to upgrade its critical infrastructure for decades. We need a robust domestic manufacturing system that ensures we can build anything, any time, for any industry: A modern power grid, comprising multiple energy sources, that can fuel the next wave of digital transformation. Adequate, quality housing for a growing population, and critical infrastructure like roads and bridges built to withstand and adapt to inevitable change. 

Building this future requires real, tangible materials and raw physical strength — deployed at a scale impossible to achieve with the tools, supply chain, and human capital we have today.  

Currently, the country is facing a deficit of nearly 500,000 skilled laborers in construction alone. This is projected to grow as more than 40% of construction workers will reach retirement age over the next decade, extensive training is required, and few young people want to work in physical labor. Already, we can’t even meet our most fundamental goals under expanded government initiatives to tackle longstanding challenges such as building more affordable housing and making much-needed upgrades to critical infrastructure such as roads, bridges, and energy grids — let alone take on the major work of reviving our domestic manufacturing capabilities, building data centers to facilitate the explosion of AI, reinvent our aerospace and defense ecosystem, and begin exploring how we can reduce our reliance on foreign sources of critical minerals. 

This is why physical AI isn’t a science fiction fantasy, but a capability we must dedicate every resource towards in order to bring it into widespread, accessible reality. If we hope to reinvent our most foundational industries, we need to equip an ever-dwindling supply of workers with the tools to do more with less. Technology that combines robotics, software, and real-world actuation can be a powerful human amplifier — and become a true accelerator for industrial progress.  

This clear need was a key focus area of Eclipse when we began conversations with autonomous vehicle technology veterans Boris Sofman and Kevin Peterson, both of whom were looking to apply their extensive expertise and hands-on experience from Waymo and other pioneering AI and robotics organizations to a new sector. We knew immediately we wanted to build with them, and committed early on to finding an idea with the perfect intersection of founder authenticity, a massive opportunity, our conviction, and experience. Boris and Kevin wanted to understand how autonomous technology could be applied to industries reliant on heavy machinery and human labor, and together we spent months in the field talking with people who build the critical infrastructure we all rely on every day. It was undeniable that organizations across construction, mining, agriculture, and others universally are feeling the pressure of huge unmet labor needs amid the national urgency to build more advanced infrastructure — and better and faster than ever before.

Recognizing that the maturation (and embrace) of physical AI aligned with the demand from these industries for sophisticated technology, Boris and Kevin decided to dedicate themselves to developing autonomous technology for heavy machinery. In 2024, they co-founded Bedrock Robotics with Tom Eliaz, and Ajay Gummalla, in partnership with Eclipse Venture Equity. You can read more about our partnership on the formation of the company here.

Led by a team of co-founders who were instrumental in advancing autonomous vehicle technology into what it is today, Bedrock is developing a generalizable autonomy platform for heavy machinery. Taking a “retrofit-first” approach, Bedrock’s platform is designed to be added to machinery manufactured within the last decade — allowing it to tap into an installed base worth hundreds of billions of dollars. The company was influenced in its earliest days through partnership with Eclipse, positioning us as key thought partners and trusted partners to lead the seed round. Bedrock has raised $80M to date, with the Series A led by 8VC.

“We hear a ton about what AI can do in the cloud, what it can do in your documents, what it can do in your spreadsheets. But at the end of the day, we build the physical world with our hands and machines,” says Bedrock Chief Technology Officer Kevin Peterson, who has spent his entire career focused on robotics and autonomy and most recently led Perception at Waymo Via, the company’s autonomous trucking division. “We need AI powering autonomous technology out in the wild. Embedding intelligence into these machines will unlock tremendous growth by helping humans build faster, safer, and more precisely.”

The company is initially targeting construction — a $1.8 trillion market — and will eventually expand to industries including agriculture (a nearly $600B market in the U.S.), mining (worth trillions of dollars worldwide), and others that utilize heavy machinery. 

“Construction is one of the lifebloods of this country, and the demand is increasing every day as we reindustrialize. But labor is just not there,” says Sofman, who has spent two decades in robotics and previously served as director of engineering at Waymo. “Autonomy is the answer that not only allows us to keep up with existing demands, but to fuel growth that is necessary for our country to thrive.”

Despite this long-running challenge and clear need for modern solutions, construction has historically been overlooked by technologists. While the industry has had a few upgrades in the form of vertical software for logistical processes such as accounting and project management, it hasn’t been the beneficiary of any physical innovations that mitigate labor shortages and drive efficiency. 

“Over the past hundred years, most sectors of the economy have increased productivity dramatically, but construction has not,” says Peterson, who also worked as an autonomy architect at machinery manufacturer Caterpillar. “Bringing in autonomy gives us the ability to not only get much more done, but to actually improve safety and precision because you effectively have more ‘workers’ on the ground.”

Along with the pervasive labor shortage and demand for technology, construction is also an obvious first choice given the increasing demand and consistent configuration of equipment used, which has allowed Bedrock to develop a customizable platform using proprietary software and a bespoke combination of existing technology including sensors, LIDAR, computer vision, machine learning controls, and robotics that can be outfitted onto any type of machinery. These components feed data into a computer placed behind the operator’s seat, which is also ingesting data from the humans currently operating the machine as the system is being trained. Eventually, the machine will be able to generate its own commands. 

“We’re bringing technologies that have been driven forward by advances in autonomy and machine learning in other industries, and that allows us to take these existing machines and turn them into these incredibly capable, autonomous tools that can solve all types of problems,” says Sofman. “This normal equipment that already exists can become autonomously capable, keep going around the clock, and do amazing work.”

Initially, Bedrock is starting with excavators. As one of the most commonly used and complex pieces of machinery, excavators handle tasks that account for a significant portion of operator hours — translating to a sizable impact on organizations working with a limited labor pool.

The company has moved quickly. A little more than a year after its formation, a Bedrock-outfitted excavator (named Fred, as in Fred Flintstone) performed its first “autonomous dump,” independently digging, lifting, and unloading a bucketful of dirt into a truck bed. 

“It’s kind of like watching a kid take their first steps,” said Sofman when watching the 100,000-lb Bedrock-equipped excavator autonomously perform the basic, yet highly complex task that fewer and fewer people each year are qualified to do. “And once it has the simple things down, these foundational models allow it to just snowball into other actions.”

Mastering the excavator — which have three actuators (boom, stick, and bucket), eight different axes of movement, and perform multiple tasks including digging, trenching, earthmoving, and more — allows the Bedrock platform to learn a wide array of the most common construction jobs in various environments, while generating data that will inform training of additional machines, says Gummalla, who leads hardware for Bedrock. This approach has enabled Bedrock to quickly advance the platform.

Gummalla, who spent more than two decades developing strategic roadmaps for novel technology rollout, previously led system architecture and programs across Waymo's many vehicle programs and was the hardware lead for trucking. He said the decision to start with the excavator was based on past experience developing autonomous technology for complex environments. Members of the Bedrock founding team were there for the first successful navigation of an autonomous vehicle on city streets — but they were also there for the bumpy road it took to get there.

“Over the course of our careers, one of the key lessons we learned is to tackle the hardest problem first, because once you tackle that hard problem, then most of the other things will just fall into place," says Gummalla. “And excavators are really hard.”

Solving the hardest problems first has proved fruitful, as even the Bedrock team has been surprised at how quickly they’ve been able to move, says Gummalla.

“It was only four or five months from the day we hired the first employee to the day the first machine was up and running, with all sensors integrated and data ingested. Then it was another two months to actually demonstrate the machine doing a full task autonomously by machine learning model.”

The most important part is that the industry is ready, says Sofman, noting the eagerness of construction organizations to partner with Bedrock. Since its inception, Bedrock has worked closely with organizations across the country like Sundt Construction, Zachary Construction, and Champion Site Prep. The system is currently running on more than 15 machines in customer construction sites in Arizona, Texas, and Arkansas, and is on track to double in scale in the coming months. 

“The need is universal and the excitement is there,” says Sofman. “These organizations are really motivated by the opportunity and they see the impact autonomy is having in other industries. The timing is right to partner together and really transform this industry.”

While autonomous vehicles now seem almost common in certain locations around the world, it’s only now that the technology can be applied to other industries, says Peterson. 

“Even just a few years ago, it would have been impossible to bring autonomy into construction,” says Peterson. “There are so many things that are coming together right now that enable us to build what we are building: Everything from compute, hardware, sensors, the models we can pull off the shelf, and the ability to control machines. It’s all new.”

Perhaps even more important is the receptiveness of humans to work with autonomous technology, says Bedrock Co-Founder Tom Eliaz, who leads cloud, infrastructure, and app technologies for the company. In this crucial training phase, the information flow has to be rich enough for machines to understand a wide range of human intentions while at the same time providing people with the visibility and insights that allow them to leverage machines to work more efficiently, safely, and precisely.

“When we talk about autonomy, we’re actually talking about bringing a bunch of different fields together in the form of interfaces between humans and machines,” says Eliaz, who previously served as an executive at Segment and was instrumental in scaling the company prior to its $3.2B acquisition by Twilio, where he then led various engineering and product organizations.

“We need to bridge what’s happening on the ground and in the operator’s seat with what the machine is learning and doing," says Eliaz. "It’s not just acting as a stand-in for labor we don’t have, but actually providing a holistic view of the entire construction job, helping with everything from current project management and scheduling to informing how to plan future projects.”

In that regard, making machines more human is the core of Bedrock’s mission. Technology that augments, enhances, and significantly extends the work of skilled humans is imperative to reshoring critical physical industries in the U.S. like manufacturing, mining, and supply chain logistics. This means building autonomous machines we can trust not only to work alongside us, but to keep the progress going when our labor pool and business day hours are exhausted.  

“We need to be able to be productive around the clock,” says Peterson. “At Bedrock, I think we're starting the fire on this incredible transition towards autonomy that we're going to see majorly impact our industrial base over the next 50 years.”

More on today's news from Forbes.

Follow Eclipse Ventures on LinkedIn or sign up for Eclipse’s Newsletter for the latest on the Industrial Evolution.