ARTIFICIAL INTELLIGENCE
Trends pushing 3D sensing from measurement to machine perception
By Dr. Stephan Ihmels, head of segment marketing and system architect for industrial automation and robotics, STMicroelectronics
T
he next wave of physical AI will be constrained as much by sensing as by compute. A robot, drone or autonomous machine can only act on the world it can perceive, and the quality of that perception determines how confidently it can move, inspect, pick, assist or avoid hazards. That makes 3D sensing a strategic part of the AI stack. The timing is significant as the International Federation of Robotics recently reported that factories installed a record 603,000 industrial robots in 2025, taking the global operational stock above five million. For engineers, those changes are already reshaping what they expect from 3D sensing
Richer 3D perception meets application-specific performance
Time-of-flight sensing has traditionally been used to measure distance, but newer applications are asking for a much fuller picture of the surrounding environment. In robotics, that can mean understanding where several objects are in three-dimensional space so a machine can navigate, inspect or handle goods. In assisted living, depth information can help identify whether someone is standing or lying on the floor without capturing a conventional image. The requirements can look very different from one application to another. A 3D sensor monitoring the fill level of a large silo may only need to take a measurement every hour, while an autonomous warehouse robot needs to refresh its view continuously as it moves around equipment and people. In a care environment, wider coverage may matter more because blind spots can reduce the usefulness of the system.
That makes application context increasingly important when engineers choose and configure 3D sensing. With 3D sensing now delivering a more detailed view of the environment, more attention also has to be paid to where that data is processed and how quickly it can be turned into something useful.
Edge AI makes power and local processing inseparable
This changes what engineers need from the sensor itself. More processing can happen close to where the data is generated,
reducing the need to send every raw measurement to a central processor and easing pressure on both power and compute. That makes local processing much more useful in practice. A motion sensor can identify a relevant event and wake the rest of the system only when action is needed, while a 3D sensor can pre-process ranging information into a depth map before passing it onwards. The goal is to do more with the data before it reaches the main processor, without adding unnecessary power or processing demands elsewhere.
There is another side to this - the AI is only as useful as the data it receives. If an AI application is trying to recognise a person’s posture or classify an object, more accurate sensor output gives the model a stronger starting point. Once several sensors are feeding the same decision, the challenge becomes getting those inputs to work together properly.
Sensor fusion turns perception into a system problem
Physical AI systems need to understand both their surroundings and their own movement, which means bringing together several types of sensor data. In an autonomous warehouse robot, for example, depth sensing might identify an obstacle or pallet while motion sensors track how the robot itself is moving. Combining those inputs gives the system a more complete picture of what is happening
14 OCTOBER 2026 | ELECTRONICS FOR ENGINEERS
around it and how it should respond. That also makes the software behind the sensing more important. A depth map on its own only shows what the sensor can see - the system still has to interpret that information and turn it into something useful, whether that means avoiding an obstacle, recognising an unsafe posture or changing course. As more sensor inputs contribute to the same decision, the challenge increasingly lies in making sure they work together reliably and quickly enough for the application.
This is also why simulation is becoming more closely tied to sensor development. Testing how sensing, perception and decision- making interact in a virtual environment gives engineers a way to understand how the complete system may behave before deployment, including how different sensor inputs respond under changing conditions. Done well, that can help close the gap between a prototype that works in development and one that performs reliably in the real world.
Where sensing goes next
The next breakthrough in 3D sensing will come from systems that can understand their environment, process the right information locally and act on it fast enough to matter. Better range and resolution will keep raising the ceiling, while the real measure of progress will be how reliably those capabilities translate into useful action in the physical world.
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42
