A humanoid robot may look like a machine built around motors, cameras, and processors. But making it useful in the real world is a far more complicated proposition. The challenge is not simply teaching a robot to see or move—it is enabling dozens of systems to sense, reason, respond, and operate safely in real time.
For Allan Lagasca, Smart Industrials, Robotics Worldwide Leader, STMicroelectronics, that complexity represents both a technology challenge and a significant opportunity. He believes the next phase of robotics will depend on tighter integration between AI, semiconductors, sensors, power electronics, software and mechanical systems. Just as importantly, the industry will need to solve cost, safety, regulation, energy efficiency and supply-chain constraints before humanoids move from impressive demonstrations to high-volume deployment.
In an interaction with Express Computer, Lagasca explains why ST is moving beyond individual semiconductor components towards a system-level approach to robotics, while also outlining where India could fit into the emerging global robotics ecosystem.
The real robotics challenge is integration
One of Lagasca’s most compelling demonstrations involves something deceptively simple: a robotic hand mimicking a human hand.
The human hand generates 21 data points, while the robotic hand in the demonstration has only six motors. Translating those 21 points into six motor movements requires AI, sensing, processing, and control to work together with extreme precision.
“The translation in here requires, again, a level of understanding that you don’t miss any data because at the same time, this hand keeps moving,” Lagasca explains.
That example captures a broader problem facing robotics. An AI engineer, hardware engineer, and software engineer may each understand their individual domains, but making those technologies work as one physical system is a different challenge.
For ST, this is becoming central to its robotics strategy. Rather than treating the humanoid robot as a single market, the company is looking at the robot as a collection of modules—hands, joints, feet, cameras, sensors, and processing systems—that can ultimately come together into a complete machine.
From components to a system-level conversation
This shift also reflects how customer expectations are changing. Lagasca argues that customers are increasingly less interested in semiconductor specifications in isolation. They want technology companies to understand the problem they are trying to solve.
He illustrates this with a customer requirement for a six-motor solution that had to fit on a 6 cm × 6 cm board. A conventional product-centric approach resulted in a much larger solution. The new system-level approach changes the conversation: instead of asking which individual components can be supplied, ST engineers can work backwards from the customer’s physical and functional requirements.
That thinking extends to AI. ST has demonstrated robotic hands combining motors, cameras and AI, including a 16-servo hand with a camera and AI capability designed to recognise objects.
The objective is not simply to put more chips into a robot. It is to make those chips work together as an integrated system.
Edge AI: Every millisecond matters
Nowhere is that integration more important than AI inference.
Lagasca sees a distributed intelligence model emerging in robotics. A large GPU can coordinate the overall system, but not every decision needs to travel back to the central processor.
Consider a robot picking up a bottle. Identifying the bottle, measuring distance, and adjusting grip pressure are tasks that can happen locally. If the robot has to send every piece of information to a central GPU and wait for instructions, latency becomes a problem.
“If you want fast reaction, complex part, move those things to the edge,” Lagasca says.
This is where edge AI becomes more than a computing architecture. It becomes a requirement for physical intelligence.
A delay of a few milliseconds can translate into a physical error. When multiple joints have to coordinate simultaneously, even a small timing or positional mismatch can cause the robot to miss its intended action. Local inference reduces that latency while also limiting unnecessary data movement.
Privacy provides another reason to keep certain intelligence at the edge. In ST’s camera-based demonstration, inference happens primarily on the camera itself, allowing visual information to be processed locally rather than unnecessarily transmitted or stored.
The economics have to work
Technology alone, however, will not create a robotics market. The economics have to make sense.
Lagasca points to today’s high cost of humanoid systems. He cites examples ranging from around $80,000 for a relatively inexpensive humanoid robot he encountered in China to $100,000–200,000, depending on the company and configuration. His own laboratory has a robot costing about $6,000, but with significantly more limited capabilities.
Battery life is another constraint. A humanoid robot consumes energy even when it is simply standing, making power efficiency a fundamental design issue.
ST’s response includes integrating multiple functions into fewer components. In one robotic-hand example, six MOSFETs and three drivers are combined into a single package, reducing the number of components that need to be mounted and potentially improving manufacturing yield and cost.
At the semiconductor level, Lagasca says ST can address approximately $600 of a robot’s bill of materials through more than 500 components, although the figure can rise towards $1,000 or more depending on the robot’s degree of freedom, number of sensors, cameras and AI requirements.
Safety may determine how quickly robots enter society
The bigger barrier may not be technology at all. It may be trust.
Humanoid robots are moving into environments designed for humans, which makes functional safety, cybersecurity, and regulation critical. Existing industrial and automotive safety standards provide foundations, but Lagasca argues that humanoids create new challenges because they are mobile and interact dynamically with people.
A robot can comply with a formal safety standard and still create risk if a sensor fails or an AI system makes the wrong decision. “Safety standards are not the final solution,” he points out.
Cybersecurity is equally consequential. A connected fleet of thousands of robots presents an entirely different attack surface from a conventional industrial machine. ST is therefore incorporating cybersecurity requirements into its product strategy, including compliance with Europe’s Cyber Resilience Act.
The path to adoption, therefore, is likely to be gradual. Demonstrations that prove endurance, agility, precision, and reliability are part of the industry’s effort to build confidence.
“Today, if you see a thousand robots in the street walking with you, you won’t feel comfortable,” Lagasca observes. But repeated exposure to useful, safe robotic applications can gradually change that perception.
India’s opportunity may begin before the humanoid
For India, Lagasca sees an opportunity that does not necessarily require building a complete humanoid robot from day one.
The near- and medium-term opportunity, in his view, lies in building the robotics supply chain—precision joints, hands, modules, components and supporting technologies—as well as developing the vast datasets needed to train robotic AI.
Yet he is candid about the current gap. India remains at an early stage, with capabilities in areas such as prosthetics and individual robotic modules, but without a full-fledged domestic humanoid robot maker at the time of the interview. Precision manufacturing and the know-how required for sophisticated robotic joints are among the gaps that need to be addressed.
That creates a potential ecosystem play for Indian startups and manufacturers—and for global semiconductor companies looking to build partnerships around them.
Lagasca says partnership-building is part of his mandate at ST, connecting startups and manufacturers with a broader global ecosystem.
A market that is still being designed
Perhaps the most revealing aspect of Lagasca’s perspective is that the robotics industry itself has not yet settled on its final architecture.
Will humanoids have two hands or more? How many degrees of freedom will become standard? Where will intelligence reside? How much processing will happen locally? What safety standards will ultimately govern human-robot interaction?
There are still no definitive answers.
For ST, that uncertainty is precisely why it needs to be close to the market now.
“We have to be there now. Because if we come later, we’ll be late,” Lagasca says.
That may ultimately be the defining semiconductor opportunity in robotics: not merely supplying components for machines that already exist, but helping shape the architecture of machines still being imagined.