By Marco Cassis, President, Analog, Power & Discrete, MEMS and Sensors Group, Head of STMicroelectronics’ Strategy, System Research and Applications, Innovation Office.

From factory floors and operating rooms to self-driving cars and smart homes, artificial intelligence is taking on tasks that require it to understand and respond to the real world in real time. This is the era of Physical AI: systems that don’t just process data, but perceive, decide, and act.

For AI to operate safely and reliably in dynamic environments, it needs a continuous, accurate feed of physical-world information from temperature to motion, pressure, proximity and orientation, interpreted fast enough to matter. This sensing layer is the foundation on which every downstream decision rests.

STMicroelectronics sits at the heart of this challenge. With a technology portfolio spanning MEMS and optical sensing, microcontrollers, edge AI processing, and wide bandgap semiconductors for power delivery, ST provides the full signal chain from first sensor contact to final actuator command.

The ability to filter, classify, and act on physical data before it ever reaches a central processor is critical. Intelligent sensing is becoming a defining enabler of Physical AI, and ST’s approach is shaping what machines can know, and do, about the world around them.

Why Physical AI needs more than data
Most AI innovation of the past decade has been driven by algorithms trained in data centers and deployed via the cloud. But as AI migrates into cars, factories, hospitals, and homes, its performance is constrained by the quality, fidelity, and timeliness of its inputs. Sensing is at the heart of making good decisions. But the dominant sensing architecture today is still fundamentally passive: sensors stream raw data to centralized processors, which derive context and issue commands. This creates three structural problems that become increasingly untenable as AI takes on safety-critical roles.

Latency is the most immediate. Autonomous vehicles, industrial robots, and human-machine interfaces cannot wait for a round trip to a remote processor. Decisions must be made in milliseconds, at the point of perception. Power and bandwidth are next. Continuous transmission of raw sensor data is expensive in energy and network resources. Physical AI requires millions of devices, many running on battery to scale.

Safety and reliability are the most consequential. Noisy signals, dropped data, or misinterpreted data in systems that operate near people pose risks. The closer AI gets to the physical world, the higher the cost of a perceptual failure. The shift underway is therefore not about deploying more sensors but about using more intelligent ones.

ST’s answer: sensors that sense and think
STMicroelectronics is addressing this challenge by moving intelligence as close as possible to where data originates. Rather than treating sensors as passive data sources, ST’s intelligent sensing solutions extract context locally, filter and interpret raw signals at the edge, and transmit only the information that matters. The result is lower latency, reduced power consumption, lighter network load, and stronger privacy, because most data never needs to leave the device.

ST pursues this through two complementary and increasingly converging pillars. Firstly, Edge AI combines ST’s sensors with STM32 microcontrollers and dedicated AI accelerators to enable analytics and inference directly on the device. This delivers fast, deterministic responses without dependence on cloud connectivity, a critical capability in automotive systems, industrial controls, and smart infrastructure where network availability cannot be assumed and response times are measured in microseconds.

Secondly, in-sensor AI pushes intelligence further still, inside the sensor itself. Through embedded cores such as the Intelligent Sensor Processing Unit (ISPU) and on-sensor machine learning engines, ST devices can process raw sensing data in real time, execute ML algorithms at ultralow power, and output high-level information using a classified gesture, a detected anomaly, a recognized motion pattern, rather than a raw data stream. This matters enormously in wearables, robotics, and personal electronics, where responsiveness and battery life are fundamental requirements. Together, these pillars define a new sensing paradigm with sensors that understand what they perceive.

ST’s full-stack advantage
What distinguishes ST’s position in intelligent sensing is not any single technology, but the coherence of the stack behind it. ST brings together leadership in MEMS motion, pressure, and environmental sensing; advanced imaging capabilities including 2D, 3D, and Time-of-Flight; STM32 microcontrollers with neural processing units; and the software tools to integrate them. Crucially, ST controls this stack from research and design through to manufacturing, including 300 mm fabs in Europe.

This integration means ST can engineer the interfaces between sensing, processing, and actuation as a coherent system, rather than optimizing each layer in isolation. For Physical AI architectures, where the interaction between sensing and decision-making is the performance-critical variable, this is a genuine differentiator. ST’s technology and product breadth positions the company to serve both consumer and high-value automotive and industrial applications simultaneously.

From consumer to mission-critical
ST has long been a trusted sensing partner for consumer electronics, enabling the motion awareness, environmental responsiveness, and biometrics that have become standard in smartphones, wearables, and gaming devices. That consumer heritage built on volume, miniaturization, and power efficiency, provides a strong foundation for the more demanding environments ST is increasingly serving.

Intelligent sensing is becoming foundational infrastructure for the most significant structural shifts of the coming decade: the energy transition, remote and continuous healthcare, smart cities, next-generation consumer devices, and the continued automation of industry. Across all of them, the sensor is no longer a passive input device. It is an intelligent agent at the frontier between the digital and physical worlds.

In automotive, ST’s intelligent sensing is enabling driver and in-cabin monitoring, advanced driver assistance systems (ADAS), and autonomous driving. In industrial settings, it underpins collaborative robotics, predictive maintenance, and condition monitoring on factory floors. In smart infrastructure, it supports continuous, low-power monitoring of buildings, energy systems, and urban environments.

In imaging, ST is concentrating on high-growth segments where precise, robust perception is essential for future mobility, automation, and smart environments.

What intelligent sensing enables: the capability map
Across every domain where Physical AI is taking hold, ST’s intelligent sensing portfolio is unlocking a consistent set of strategic capabilities:

Immediate responsiveness. Robots, vehicles, and autonomous devices can react to people and changing conditions in real time, enabling interactions that feel natural rather than mechanical.

Operational safety. Continuous, high-fidelity perception and motion awareness are prerequisites for systems that share space with humans whether on a factory floor, a public road, or in a domestic environment.

Edge independence. Local processing reduces dependence on network connectivity, increases system resilience, and cuts the energy and bandwidth cost of large-scale deployment.

System simplicity. Higher integration and embedded intelligence reduce external component counts, lowering both system complexity and bill-of-materials cost.

Predictive maintenance and uptime. Low-power continuous monitoring enables condition-based maintenance strategies that extend asset life and reduce unplanned downtime.

Privacy and trustworthiness. Sensitive data processed on-device never traverses a network. Combined with in-house manufacturing and a robust supply chain, this supports secure, auditable deployment at scale.

The humanoid robot test case
No application illustrates the demands of Physical AI more vividly than humanoid robotics. A humanoid robot must sense its environment continuously and accurately, understand human presence and intention, maintain dynamic balance, navigate unpredictable spaces, and manipulate objects, often in close proximity to people, in real time, without fail. It is, in effect, a walking proof of concept for every challenge Physical AI must solve.

ST estimates that a typical humanoid robot contains approximately $600 of addressable semiconductor content. ST is already working with multiple humanoid robot developers worldwide, providing the sensing and edge computing capabilities that allow these systems to perceive, move, and interact safely in human environments. As the humanoid market scales from research platforms to deployed systems, the quality of sensing will be among the most important variables. That awareness is what Physical AI requires. And sensing is where it begins.

Conclusion
Physical AI is already being designed into products that will define the next generation of mobility, manufacturing, healthcare, and infrastructure. By embedding intelligence at the point of perception ST is helping customers build machines that are not merely connected to the world, but genuinely aware of it.