Nico Peng
← Back to the Physical AI index

Physical AI · Enabling market

Sensing

The purchased systems that measure the scene, the machine's movement and its contact with the world.

AS OF 2026-09-13DATED · NON-EXHAUSTIVE

Research reviewed 2026-09-13. Available means documented access or production, not assured spare capacity for every specification. Tight refers to the scope identified in the node, not every robot component. Supply unclear means the reviewed evidence does not settle availability at the selected boundary. Expansion describes how additional usable supply can be delivered, not a promised lead time. Pricing power describes our assessment of the ability to defend price or terms; it is not inferred from technical importance alone.

Useful intelligence needs usable measurements

Seeing an object is not feeling a contact

A machine needs observations that support its task. Images describe appearance; ranging systems measure spatial relationships; navigation instruments estimate movement and position; force sensors measure mechanical interaction. The useful commonality is a supplier's ability to deliver a reliable measurement, not the assertion that all machines need the same sensor set.

This map uses supplied sensing systems as its main boundary. An image sensor, lens, housing, calibration and acquisition software can be separate purchases, but treating all of them as compulsory nodes would obscure the camera or vision system the machine builder is selecting. By contrast, combining cameras with force sensing would erase a genuinely different physical quantity and supplier proposition.

The chosen systems are representative economic routes, not an exhaustive instrument catalogue. Thermal imaging, ultrasonic proximity sensing and distributed tactile arrays can matter in particular machines. Joint position feedback is examined in Motion, where it directly supports the controlled axis.

Measure useful coverage, not sensor count

Growth can come from new machines, more demanding coverage or new tasks that need a different measurement. Better algorithms may also extract more from an existing camera, while improved packaging can combine functions. A more capable robot need not contain more separate sensor units.

There is actual shipment evidence in lidar. Hesai reported 142,371 robotics lidar shipments in the second quarter of 2026. That establishes a growing commercial business at one supplier. It does not establish that every sensor category is scarce, or that shipment growth translates proportionally into revenue or margin.

Supply-chain map

Different ways to observe the scene

Vision, lidar and radar can complement or substitute for one another within a chosen architecture. No machine is assumed to buy all three.

Offering / activityEnables / feedsConditional route◇ Downstream boundary
Select a node to read its explanation
Different ways to observe the sceneVision, lidar and radar can complement or substitute for one another within a chosen architecture. No machine is assumed to buy all three.

Dashed paths mean architecture-dependent use. The shared destination is a perception function, not a common purchaser, universal supplier or certified safety system.

Vision systems

Supply unclear
Role
Supplies calibrated images or depth observations
Expansion
Camera assembly, optical calibration and application qualification
Pricing power
Integration can differentiate; hardware and software boundaries vary

Why it exists

A machine builder needs repeatable observations with suitable resolution, timing and geometry. The delivered camera or vision system brings together optics, sensing electronics and acquisition support. Stereo and time-of-flight systems can also provide depth rather than leaving all spatial reconstruction to the customer's software.

Relevant businesses

  • Basler: pre-calibrated stereo and time-of-flight cameras, with acquisition software and optional robotics application modules.
  • Teledyne: 3D camera systems including stereo and combined imaging/depth products for robotics, logistics and inspection.
  • Cognex: In-Sight vision systems and robot-guidance functions, representing a more application-integrated purchase than a bare camera.

Worth understanding

Basler's stereo visard offering illustrates the scope choice: processing and software can accompany the camera. A customer buying usable depth observations transfers more integration work than one buying a bare imager. This is why the node is a vision system rather than every optical and electronic part inside it.

The platforms are established products, but the reviewed material does not settle lead times or spare capacity across camera types. That is the specific reason for the unclear supply label. A high-resolution sensor shortage should not be inferred from a successful bin-picking demonstration, nor should a catalogue be treated as proof of unrestricted production.

Our inference is that calibration, synchronization, interfaces and supported application software can justify differentiation. The counterforce is that customers may use standard camera interfaces and retain their own perception stack. More capable image interpretation can increase the usefulness of installed cameras without adding a new hardware sale.

↑ Back to the map

Lidar systems

Available
Role
Measures spatial structure through laser ranging
Expansion
Production scaling, component sourcing and calibration yield
Pricing power
Mix and cost reduction matter; volume growth does not establish pricing power

Why it exists

Lidar supplies spatial measurements that a machine can use for perception or localization. The supplied system integrates emission, detection, scanning or other ranging architecture, processing and calibration. These functions are not assumed to be independent purchases by the robot builder.

Relevant businesses

  • Hesai: reports commercial shipments separately for ADAS and robotics lidar.
  • Ouster: supplies lidar and other sensing products into industrial, robotics and additional markets; its latest combined sensor count is not a pure lidar shipment total.

Worth understanding

The availability label rests on reported commercial shipments, not just product announcements. It means functioning merchant supply is demonstrated; it does not assert spare capacity at every specification. Hesai's second-quarter robotics shipments grew faster than total company revenue, while its company gross margin was lower year over year. Product mix and costs matter alongside units.

Ouster's August 2026 results reported more than 17,000 lidar and camera sensors shipped for revenue in the quarter, with lidar representing about 53%. Comparing that total directly with a competitor's lidar-only count would misread the supply base.

Our inference is that manufacturing scale and usable performance can widen adoption as costs fall. That may reward a well-run supplier without creating a durable shortage rent. A warehouse vehicle, road vehicle and mapping platform also need different combinations of range, field of view, robustness and integration. The category is shared; the qualified product is not necessarily interchangeable.

↑ Back to the map

Radar systems

Available
Role
Measures surrounding objects using radio-frequency sensing
Expansion
Established production platforms; custom interfaces and qualification take additional work
Pricing power
Scale and integrated processing compete with programme-level bargaining

Why it exists

Radar provides observations of surrounding objects using radio waves. The system purchase includes the sensor architecture and an agreed output, which may be rawer detections, tracked objects or more integrated functions. Its value depends on performance within the selected vehicle or machine environment.

Relevant businesses

  • AUMOVIO: distinguishes off-the-shelf industrial radar from special applications and customer-specific automotive systems.
  • Bosch (private): front and corner radar systems, including integrated electronics and processing for vehicle functions.

Worth understanding

AUMOVIO's sample programme offers low-volume access to technology based on production platforms. This supports availability at that stated boundary, not unlimited capacity for a new automotive contract. Its industrial portfolio explicitly separates standard purchases from products needing additional development.

Road and off-highway buyers can draw on related production capability, but industrial protective sensing has its own duties. Inxpect's radar-based protective products illustrate that separate application. A perception radar should not be assumed to satisfy a protective-system requirement merely because it detects people.

Our inference is that integration can move value between the sensor and central computing platform. Bosch describes on-sensor processing, while AUMOVIO's satellite radar architecture shifts more processing centrally. Sensor count alone therefore misses changes in purchased content. Large programme buyers and alternative architectures limit any automatic claim of strong pricing power.

↑ Back to the map

Supply-chain map

Knowing movement and contact

Navigation instruments and force sensors retain different product boundaries. One locates a moving platform; the other measures its mechanical interaction.

Offering / activityEnables / feedsConditional route◇ Downstream boundary
Select a node to read its explanation
Knowing movement and contactNavigation instruments and force sensors retain different product boundaries. One locates a moving platform; the other measures its mechanical interaction.

Navigation can use additional visual or ranging information. Satellite assistance is not universal, particularly indoors. A wrist force sensor is not a substitute for a distributed tactile skin.

Navigation instruments

Supply unclear
Role
Estimates platform position, velocity and orientation
Expansion
Inertial calibration, receiver integration and application qualification
Pricing power
Accuracy under difficult conditions and integration can differentiate

Why it exists

A moving machine must estimate its own motion as well as observe surrounding objects. An inertial navigation instrument combines measured rotation and acceleration with estimation software, and may integrate satellite positioning or other aids. The economically useful boundary here is the calibrated navigation solution rather than a separate node for every measurement axis.

Relevant businesses

  • Hexagon | NovAtel: CPT7 combines GNSS and inertial navigation in one enclosure, including an externally sourced Honeywell IMU.
  • SBG Systems (private): Quanta Micro integrates inertial sensing, GNSS and navigation capability in an OEM package.

Worth understanding

NovAtel's CPT7 makes the purchasing test concrete: a buyer can obtain a combined solution while upstream inertial hardware remains a distinct supplier's contribution. We do not merge all those suppliers into one company, but the machine builder need not separately buy and integrate each measurement function.

NovAtel describes GNSS and inertial measurements as complementary. That does not make satellite reception mandatory for indoor robots, or inertial estimates immune to drift. Camera, lidar, wheel and other observations may participate in a different localization architecture.

Our inference is that calibration quality, difficult operating conditions and supported interfaces can support a premium. The reviewed evidence establishes integration choices but not a node-wide order queue or realized price trend. The supply label remains unclear at that narrower commercial question. Counting every robot as a new high-accuracy GNSS/INS sale would substantially overstate exposure.

↑ Back to the map

Force / torque sensing

Supply unclear
Role
Measures mechanical interaction during a task
Expansion
Transducer manufacture, calibration and mechanical/interface adaptation
Pricing power
Calibrated performance and application integration can support differentiation

Why it exists

A contact task needs to know what force is being applied, not just where the tool appears to be. A force/torque sensing system provides calibrated measurements for control or analysis. Its transducer, electronics and calibration form a coherent supplied product even when the electronics are mounted separately.

Relevant businesses

  • ATI, a Novanta business: multi-axis force/torque sensing with different mounting, interface and calibration options.
  • OnRobot (private): HEX six-axis force/torque sensing with software for applications such as assembly and surface finishing.

Worth understanding

ATI's robot-specific integration options and OnRobot's supplied software show why buyers may prefer a supported measurement system over an isolated transducer. Fewer engineering hours can be valuable even without a physical shortage. However, the reviewed catalogues do not establish industry spare capacity or common delivery times.

A wrist force sensor measures a different thing from the contact distribution across a tactile finger. Nor is buying one compulsory: some machines use estimates from their existing actuators or operate tasks that do not require fine contact control. The opportunity depends on applications in which the extra measurement improves accepted output enough to justify its cost.

Our inference is that calibration, overload tolerance, compact packaging and software integration can create differentiation. The key test is whether a supplier carries those capabilities across customers without proportionate custom engineering. Broader participation in several robot families is evidence of relevance, not proof that all of those programmes share a production line or margin.

↑ Back to the map

The valuable measurement is the one the task needs

Sensing contains established merchant businesses and meaningful differences in how value is captured. Shipped lidar volume and production-based radar offerings provide firmer availability evidence than a catalogue alone. Other nodes still need direct lead-time and capacity work before receiving a stronger supply judgment.

For each supplier, separate hardware content, calibration and application software. Then ask whether a better sensor expands the task's economics, whether existing hardware can achieve the same result through software, and which integration work the customer is actually paying the supplier to own.

Related reading

Control & runtime · Motion · Compute & memory · Road autonomy · Aerial systems

This dated map selects economically useful boundaries. Company participation does not establish a trade relationship between the named suppliers, and a product announcement is not evidence of accepted machine output.