Long before artificial intelligence (AI) agents, physical AI and humanoid bots, there were plain old robots with clunky metallic arms of limited use.
Factory floors and industrial settings cluttered with forklifts and conveyer belts for decades are expected to eventually yield to an onslaught of highly-specialized AI facilities with software-defined, multi-functional robots and drones that will shape the next wave of warehouse and supply chain innovation. The day is coming…
But just not yet.
Getting there is taking longer than expected because only a tiny slice of manufacturers have the resources — money, people, know-how — to purchase and integrate the equipment. Those with the capital tend to be Big Tech, automakers, Fortune 100 companies, and risk-taking ventures.
The first wave of useful commercially available, general-purpose humanoid robots aren’t ready, so for now low-level tasks are being performed by rudimentary, purpose-built robots like high-tech dishwaters, according to Tessa Lau, CEO of Dusty Robotics, which builds robots for the construction industry.
“Spatial temporal data is what’s missing from AI today. They do not incorporate space and time,” Lau said in an interview. “It is coming. Physical Intelligence, NVIDIA, Google Gemini Robotics, Amazon/Kiva Robotics, which is working on a home robot/Amazon Echo Look, are doing it.”
“There is an expertise gap at the edge,” Blaize CEO Dinakar Munagala added in an interview. “But for someone who gets it right, it is at a massive, massive advantage.”
“Physical AI can redefine and remake manufacturing and industrial spaces, especially as LLMs (large language models) advance,” Munagala said. “It is creating real, tangible ROI in raising top lines or lowering costs.”
If anyone is to blaze a trail, it is likely to be those who already have dabbled in robotics, such as automakers, delivery services DHL and Amazon.com Inc., and increasingly agricultural giants like Deere & Co.
An onslaught of tech partnerships such as NVIDIA Corp.-General Motors Co. and Broadcom Inc.-Audi signal a key shift toward software-defined automation, replacing factory-floor hardware with flexible, edge-hosted virtual infrastructure to improve efficiency, uptime and sustainability. In the case of Audi, it will transform the German company from a factory model based on thousands of distributed industrial PCs to one driven by virtual systems hosted on edge infrastructure.
From the floor of an Audi plant in China, UBTech Robotics’ Walker S1 humanoid robot performs quality inspection tasks, including detecting air-conditioning system leaks that pose respiratory risks to humans.
At the same time, robotics startups like Robust AI are helping warehouses streamline operations on their path to a next-generation robotic supply chain market that is expected to reach $44.6 billion by 2035 from $11.9 billion today.
Mytra, which provides robotics in warehouse automation for Albertsons Companies Inc. and other retailers, is simply trying to more easily “move things from point A to point B in a three-dimensional space,” says Chris Walti, its CEO and co-founder. The former Tesla Inc. employee previously led creation of the car maker’s Optimus bot and oversaw production of the Model 3.
“Industrial is bleeding edge [these days], yet robotics are not within most factories with the exception of the auto industry, which needed to move heavy objects since the 1960s,” Walti said in an interview.
Leading the technology charge are companies like the Jeff Bezos-backed Physical Intelligence, which is developing AI robotic maids; NVIDIA, whose physical AI push broadens the scope of its platforms for building AI apps embedded in robots, autonomous vehicles and other physical forms; Google DeepMind’s Gemini Robotics, who this month introduced two new AI models, based on Gemini 2.0, that lay the foundation for a new generation of “helpful robots”; and Amazon’s Kiva Robotics, a key cog in the company’s decade-long effort to become the leader in warehouse robotics.
“The Physical AI revolution is in full swing, with Nvidia expanding its Omniverse platform across manufacturing, simulating humanoid robots for factory automation and logistics optimization,” Vitaly Golomb, managing partner of Mavka Capital, said in an email. “The company’s partnership with GM to integrate AI into factory robotics and autonomous vehicle development showcases the transformative potential of AI in traditional industries.”
A Quick Robotics History Lesson
Automation – in the new form of AI, machine learning (ML), and robotics – will manage the labor supply more efficiently than at any time in human history, roiling multiple industries and eliminating some 83 million jobs in the next two years, according to a recent World Economic Forum study. (The same study revealed that 69 million new jobs will be created under the same circumstances.)
Self-driving autonomous vehicles will overrun transportation jobs – about one in four Americans make a living driving something – creating disruption not seen since the industrial revolution.
Roots of robotics can arguably be traced to Ford Motor Co.’s seminal Model T assembly line in the early 20th Century, and evolved through inventions like the Programmable Universal Machine for Assembly (PUMA) robots from Unimation for automotive manufacturing. GM deployed the first industrial robot at a New Jersey assembly plant in 1961.
“What surprises me is how much history has been forgotten,” says Dr. Kevin Dowling, a longtime entrepreneur, roboticist and inventor with ties to Carnegie Mellon University who assisted in mobility testing of Amazon’s Astro robot. “Speech recognition and facial recognition predated AI.”
Amazon’s work at fulfillment centers in the early 2000s ushered in a new age of warehousing, taking on highly repetitive tasks to store, manage and move inventory. In late 2023, the company launched a new robotic system called Sequoia to help fulfill customer orders for holiday shopping. The new technology started at fulfillment centers in Houston.
“Warehousing and manufacturing is about dynamically moving things around. Assembling components or shipping out products. Collaborative robotics,” Rodney Brooks, legendary roboticist of nearly 50 years and Roomba co-inventor, said in an interview.
“The world has a large diversity of situations to work within,” said Brooks, who is now chief technology officer at Robust AI. He points to the concept of simultaneous localization and mapping (SLAM), the ability of a robot to actively generate the map of an area while also keeping track of its own location, as a key breakthrough in automation.
Autonomous vehicles, meanwhile, have greatly assisted in the transportation and distribution of goods.
“The DARPA Grand Challenge (2004) changed everything in industrial automation,” says Andrew Singletary, CEO and co-founder of 3Laws, a robotics software company specializing in safety software for autonomous vehicles, aircraft and mobile robots. “It was a turning point in the difference between automation with assembly-like tasks and autonomy with intelligent decision-making machines).”
“Now, it is cheaper to do autonomy with AMRs (autonomous mobile robots) than automation tools like forklifts and conveyer belts,” Singletary said.
Microsoft Corp.’s Sanctuary AI, recently revealed in a blog, promises to usher in a new era of autonomous labor via Azure. As frontline labor shortages intensify, manufacturers can explore deploying advanced general-purpose robots with dexterity-driven physical AI to automate repetitive, complex, and unsafe tasks to enhance operational efficiency, according to Microsoft.
Still, a next wave of commercially available, general-purpose humanoid robots aren’t here yet. For now, what’s out there only perform low-level tasks. For example, a high-tech dishwater is a rudimentary, purpose-built robot that looks nothing like a robot.
Harvesting robots for such tasks as tomato packing and strawberry picking are expensive and hard to program.
At Carnegie Mellon University’s robotics labs a variety of bots built and programmed to do industrial work shoulder-to-shoulder with humans included a sophisticated robotic hand, DeltaHand, that can manipulate a syringe or pick grapes; and SonicBoom, a fruit picking device that relies as much on sound as vision to pluck apples from trees.
It will take some time, but fruit-picking bots and even more sophisticated models are on the way.
