Robots can't talk to each other yet -- that's a big AI bottleneck

Gettyimages: Andriy Onufrienko

Physical AI is increasingly moving into factories, warehouses and other operational settings.

Gartner identifies it as one of its top strategic technology trends for 2026.

Yet the tendency to treat it as a single tech category could prevent companies from realizing its full value. Instead, the research firm argued that businesses should shift their attention from individual devices to the systems that coordinate them -- a strategy that is currently limited.

"Robot vendors are keen to push themselves as complete solutions and often demonstrate their products as standalone objects which can be seamlessly integrated into existing workflows. This is, clearly, not the case," said Bill Ray, an analyst at Gartner, told TechTarget.

According to Ray, current orchestration platforms are limited in functionality and primarily focused on industries where multiple robot fleets already interact, such as warehouses.

Related:China leads dexterous hand supply for humanoids

"We do see a patchwork approach being used, but there's an unserved opportunity for orchestration, which will grow as the number of robots and other physical AI increases," he added.

Operating beyond individual robots

Orchestration involves a software layer that connects robot fleet management systems to one another and to broader building and enterprise infrastructure.

Robot vendors will continue to provide fleet management software, Ray said, with the orchestration layer coordinating tasks among robots rather than communicating directly with each machine.

He pointed to Accenture's Physical AI Orchestrator as an example of the emerging approach, with more providers expected to enter the market with orchestration platforms tailored to specific industries.

Yet Jan Liphardt, a Stanford University associate professor of bioengineering who is also helping develop robotics software at OpenMind, argued that making robots work together reliably remains a substantial challenge.

"Interoperability of computers, humans, and robots is a big gap," he said. "In industrial deployments, making sure that different robots work together reliably can be very difficult, expensive and slow."

In some cases, getting robots from two different suppliers to perform coordinated tasks can take months or even a year, he added. This creates an opportunity for technologies that simplify integration, although he cautioned that the industry may not yet be ready for orchestration to become a central concern across all deployments.

"It’s hard enough to get one robot to work well, and then it’s even more difficult to get different types of robots to work together well, and then it’s even more difficult still to get different robots from different manufacturers to work together well," he said.

Related:Lack of training data stifling humanoid bot development

There is significant interest in protocols to address these challenges, Liphardt added, but he said the problem is still somewhat ahead of the market.

"It is likely to be exceptionally important in the medium and long-term, such as the middle of next year,” he added.

Interoperability is not the only barrier

Not all analysts see interoperability as the main obstacle to adopting physical AI.

Mark Beccue, an analyst at Omdia, a division of Informa TechTarget, argues that businesses are more likely to be held back by the limitations of existing infrastructure and a lack of relevant expertise than by the absence of interoperability standards.

"I don’t see interoperability standards being something driving adoption. Leveraging open source tools, skills and models, as championed by Nvidia, will help accelerate adoption," he said.

"The biggest issues are whether legacy systems are flexible enough to integrate very new technology and approaches,” he continued. "The second biggest would be mismatched expertise -- does the organization have the knowledge or expertise on staff required for some vendor systems?"

Related:Nvidia-backed Skild AI teaches robots new tasks from a single video

That is, even with orchestration in place, businesses still need the underlying infrastructure, skills and operational processes to deploy physical AI successfully.

Another, more immediate, challenge is integrating autonomous systems into existing workplaces. Deploying robots and autonomous vehicles may require changes to site layouts and workflows, particularly when machines must operate safely alongside people.

"If robots and autonomous vehicles need to be separated from people, then it may require reorganization of the site, working out how the existing workflow can be maintained, while changing how people move around the site," Ray said. "The cost of such changes is very easy to underestimate."

Businesses must also decide how to divide work among people and machines, rather than assuming that every task should be automated.

In other words, integrating the technology into existing operations matters as much as the technology itself, and making the wrong choice could leave businesses behind.

"The orchestration layer will be enormously sticky -- changing vendors will be very hard, and with every additional integration it will get harder, so it’s important to make the right decision early," Ray said.

Looking ahead

As physical AI deployments expand, understanding how robots, enterprise systems and human workflows fit together will leave companies in a better place to scale their tech.

"The first step, and this applies to any technical innovation consideration, is to develop a comprehensive business case," Beccue said. "Ask what problem are we trying to solve? What are the one-time and ongoing costs, ROI -- net present value, internal rate of return or payback period calculations."

Liphardt took a more immediate approach, arguing that businesses should be experimenting with physical AI now, even if questions around interoperability, data formats and vendor lock-in remain unresolved.

"My recommendation right now, in late 2026, is: whatever you're making, whatever you're moving, whatever you're shipping or gluing or sewing or assembling, try out different solutions, and try them out by yesterday," he said. "At this stage you should have multiple pilots deployed in your factories, so at the very least you're fully informed about a technology that is advancing extremely quickly."

About the Author