A Unitree humanoid robot at the China Humanoid Robot Conference in Seoul in March.
Robots are back in the national spotlight, and this time, the conversation is about more than what they can do. It’s about who controls them and what data they collect.
The restrictions on foreign-produced robots by the Federal Communications Commission (FCC) underscore the growing scrutiny of these machines as connected infrastructure. As robots take on more work in factories, warehouses, and other high-stakes environments, questions about how they use data and make decisions are becoming harder to ignore.
That scrutiny is likely to accelerate the move toward on-premises AI architectures and smaller, specialized models that can run closer to where robots operate. The more control companies want over who can access a robot and its data, the more incentive they have to keep sensitive processing close to the machine.
Companies deploying physical AI today need to decide which AI workloads belong on the robot, what can remain in the cloud, and whether the underlying technology stack can meet rising expectations for security, safety, and supply-chain transparency.
Robot security is becoming an architecture question
In July, the FCC added foreign-produced advanced robotic devices to a list of technologies it deems to pose unacceptable national security or safety risks. That designation prevents new devices from receiving the FCC authorization generally required to be imported, marketed, or sold in the U.S.
The restrictions apply broadly to advanced mobile robots, including humanoids and quadrupeds, that can collect detailed information about the environments around them. Officials have warned that foreign-made robots could give foreign actors a way to conduct surveillance, collect sensitive data, or remotely interfere with machines operating inside American facilities.
But hardware origin is only one part of the security equation. If a robot is continuously collecting information about a facility and sending it to an external cloud service for processing, companies also have to consider where that data goes and who can access it.
Editor’s Note: The Robot Report is hosting a free webinar on October 27 called “The FCC Robotics Ruling: What Automation Buyers Need to Know.” Join leaders from Vecna Robotics, Locus Robotics and the Association for Advancing Automation (A3) for a discussion about what the FCC action means for the robotics industry and the organizations investing in automation.
Moving intelligence closer to the robot
Keeping more inference on the robot or inside a facility was already attractive because it improves latency and reliability. The recent regulatory change could make local processing a higher priority for a broader range of AI workloads, especially those involving internal operational data.
For core control functions such as motion control and safety monitoring, local processing is non-negotiable. The bigger change is likely to happen higher in the stack, where robots interpret instructions, analyze their surroundings, respond to anomalies, and make task-level decisions.
Those workloads can be computationally intensive, and the cloud offers more processing power than most edge hardware. But relying on external cloud resources can introduce network dependence and require operational data to leave the facility. Companies will likely need to run more of that intelligence locally within tighter compute limits.
Explore our database of 200+ humanoid robotics companies from around the world.
Why SLMs could gain ground
The more organizations move higher-level intelligence on-device, the less practical it becomes to rely on cloud-dependent large language models (LLMs) for every task. That strengthens the case for small language models (SLMs) designed for industrial environments.
A robot operating in a defined environment doesn’t always need a model built to handle almost any question. It needs one that understands the context of the facility and can perform the tasks required there reliably.
That doesn’t mean the robot has to be locked into a single job. With the right context and access to local data, an SLM can support a wider range of applications within the same facility. For example, a robotic arm working on one production line could be moved elsewhere in the factory and adapted to perform a similar task in a different production environment.
SLMs can be fine-tuned on a factory’s own data and run directly on edge hardware with lower compute requirements than larger general-purpose models. For many industrial use cases, the combination of specialization and flexibility across changing local deployments can make them a stronger fit for the job.
Register now and help us celebrate 20 years of RoboBusiness!More local control means more integration work
These changes raise the bar for system design at U.S. companies building or integrating robots. You must ensure that controllers, sensors, safety systems, and AI hardware can work together reliably and that the full system can be secured and maintained over time.
When you buy a finished robot, many integration and certification decisions have already been made. Building one from components puts more of the engineering burden on your organization: How will components be integrated? How will safety be validated? How will firmware be maintained, and where will critical parts come from?
The right suppliers can help answer those questions with tested components, traceable sourcing, and deployment support.
U.S. businesses need suppliers that can show how their components have been tested together, document where critical parts come from, and provide the safety and governance support needed after deployment. When those pieces have already been validated to work together, builders can spend less time solving component-level integration problems and more time focusing on the complete system.
Building for control from the start
The most immediate effect of the FCC action is on sourcing. Its longer-term impact may be even bigger: architecture.
As robots move deeper into sensitive and mission-critical environments, companies will have to evaluate them on more than capability or performance. They need to know whether critical processing can stay local, where robot data is going, and how much confidence they can place in the underlying components.
Businesses that build those controls in from the start will be better positioned to deploy physical AI without creating problems they have to unwind later.
About the Author
Jenny Shern, general manager at NexCOBOT, is responsible for managing the company’s sales and business strategy, robotic and motion control product development and project deployment. Shern specializes in building strategic partnerships and business development in IoT automation, robotics and retail sectors. She previously worked at NEXCOM from 2005 to 2018, where she led sales teams in building global channel network and multi-national automation key account customers. In 2018, NexCOBOT spun off from NEXCOM IoT Automation Solutions Business Group, focusing on providing open robotic and machine control systems for industrial and collaborative robot applications.
Tell Us What You Think!