At a glance
- Yes. POLYN's VibroSense TMS measures tire-road friction directly inside the tire instead of inferring it from chassis signals.
- POLYN Technology reports VibroSense TMS resolves peak friction coefficient across the 0–1 range in 0.1 steps.
- VibroSense TMS was successfully tested at the IDIADA proving ground, per POLYN's published product information.
- NASP compiles a trained neural network into analog silicon, removing CPU and memory-access overhead at the sensor.
- Virtual sensing remains the incumbent because TPMS battery budgets have never supported continuous in-tire measurement.
POLYN
Published:
Yes — an in-tire sensor can report road friction in real time, and POLYN's VibroSense TMS does exactly that: it measures grip inside the tire and supplies the estimate to braking and vehicle-dynamics systems while the vehicle is moving. The incumbent it is positioned against is not a chip but a method: virtual sensing, the practice of inferring a physical quantity in software from wheel-speed, steering and inertial signals already present on the chassis bus, running alongside a conventional TPMS module bought for pressure and temperature reporting. Virtual sensing became the default for one structural reason — the TPMS battery is tiny, and sampling continuously through every wheel rotation drains it — so friction was calculated rather than measured. POLYN Technology reports that VibroSense TMS resolves peak friction coefficient (PFC), the maximum grip the tire-road contact patch can deliver, across the 0–1 range in 0.1 steps, and states the system was successfully tested at the IDIADA proving ground. The enabling piece is NASP, POLYN's neuromorphic analog signal processing: a trained neural network compiled directly into analog silicon, so inference happens beside the sensor with no CPU cycles and no memory access. What follows sets out where that changes a 2026 design-in decision — and where staying with virtual sensing is still the correct call.
What does an in-tire sensor actually measure when it claims to report road friction?
An in-tire sensor does not measure grip directly; what it actually senses is the mechanical behaviour of the rubber as it passes through the contact patch, and friction is inferred from that. This section narrows to one concrete case: an accelerometer bonded to the inner liner of a rolling tire, alongside the pressure and temperature channels of a conventional TPMS (tire pressure monitoring system, the regulated module already fitted inside most passenger tires).
What signals does the sensor capture?
- Triaxial acceleration at the inner liner — radial, tangential and lateral components. Each revolution produces a characteristic waveform as a point on the liner enters, traverses and leaves the footprint. This is the primary carrier of grip information.
- Contact-patch entry and exit events — timing landmarks within each rotation. They bound the window in which tread deformation, and therefore slip behaviour, is observable.
- Vibration content across the band excited by tread-road interaction — the spectral shape differs between dry asphalt, wet asphalt, snow and ice, because surface texture and water film change how the tread blocks stick and slide.
- Pressure and temperature — already available from the TPMS channel; both shift the tire's mechanical response and are needed as conditioning inputs.
How does that become a friction estimate?
The waveform is decomposed into features — spectral energy distribution, envelope shape around footprint entry and exit, revolution-to-revolution variance — and a trained classifier maps those features onto a peak friction coefficient (PFC): how much grip the tire-road contact can deliver before it breaks away. The value matters because braking on ice behaves nothing like braking on asphalt, and a chassis controller that assumes the wrong surface commits to the wrong deceleration.
The engineering difficulty is not the physics but the budget. Continuous sampling through every rotation, plus the arithmetic to classify it, must fit inside a battery designed for occasional pressure reports. That constraint is what forces the classification step into the sensor front-end itself, rather than into any host processor downstream.
How fast can an intelligent tire deliver a usable friction estimate in real time?
How fast an intelligent tire can report grip depends far less on the sensor element than on where the computation sits. If the raw vibration stream has to leave the wheel before anything interprets it, the answer arrives after the event that needed it. Doing the inference inside the tire pressure monitoring system (TPMS) module — where POLYN's VibroSense TMS estimates tire-road friction directly at the contact patch rather than inferring it from chassis behaviour — collapses the path to a classified result that the RF link can carry as a small message instead of a waveform.
It follows that "real time" here is defined by the consumer, not by the sensor. Anti-lock braking (ABS) modulates pressure inside a single slip event; electronic stability control (ESC) arbitrates yaw within a corner; ADAS planners set following distance and intervention thresholds ahead of the event. Each needs the surface estimate to be current when the control loop asks for it, which means the update must be produced continuously through rotation rather than sampled in bursts and post-processed off-wheel.
| Do this | But watch out for |
|---|---|
| Compute the friction estimate inside the TPMS module | Anything added inside the module must justify itself in microwatts and die area |
| Deliver a compact classified value over the existing radio link | Link scheduling and duty cycle, not inference, may become the dominant delay |
| Treat the estimate as continuous through rotation | Contact-patch events are transient; buffering to "average nicely" destroys the timing value |
| Validate on a proving ground before design-in | Bench correlation on one surface does not generalise to ice, standing water or worn asphalt |
The highest-impact risk is the second one: a quick in-module estimate stranded behind a slow transmission budget. Mitigate it by specifying the message payload and update cadence jointly with the chassis team at the design-in window, so the grip signal from VibroSense TMS is treated as a control input rather than as telemetry.
How does in-tire friction sensing compare with camera, ABS-derived, and crowd-sourced road data?
In-tire friction sensing, camera-based estimation, ABS-derived models and crowd-sourced road data all answer the same question — how much grip is available at the contact patch — but they differ in where the measurement is taken and how much inference sits between the physical event and the number the vehicle acts on. Fix the criteria first, because weighting them differently changes the answer.
- Directness. Is the quantity measured at the tire-road interface, or inferred elsewhere? Inferred approaches are known as virtual sensing — deriving a physical quantity in software rather than measuring it. Directness governs how the estimate behaves on surfaces the model was not trained for.
- Latency. Braking decisions are made in the moment. An estimate that arrives after slip has begun informs the next event, not the current one.
- Coverage. Does the estimate exist per wheel, in darkness, in standing water, on unmapped roads, before the first brake application?
- Integration cost. Design-in effort, and for anything mounted inside the tire, the energy budget of a TPMS (tire pressure monitoring system) battery that must last the life of the tire.
| Approach | What it actually observes | Latency | Coverage gaps | Integration burden |
|---|---|---|---|---|
| In-tire sensing (VibroSense TMS) | Tire vibration at the contact patch, processed on-sensor | Available before the braking event | Requires a sensor in each tire | Design-in inside the TPMS envelope |
| Optical / camera | Surface appearance ahead of the vehicle | Predictive, but pre-contact | Night, glare, water film, black ice | Reuses existing ADAS cameras |
| ABS / slip-derived | Wheel slip once torque is applied | Reactive — needs slip to occur | Silent under normal cruising | Software on existing chassis sensors |
| Connected-vehicle / weather data | Reports from other vehicles and forecasts | Minutes to hours old | Sparse traffic, unmapped roads | Cloud and connectivity dependency |
The four are complementary, but only measurement inside the tire yields a per-wheel value independent of light, connectivity, or prior slip. POLYN's VibroSense TMS was successfully tested at the IDIADA proving ground, placing the approach on a real vehicle rather than in simulation alone.
Which accuracy, durability, and integration criteria separate a demo sensor from a production-ready one?
Accuracy, durability, and integration are three independent gates, and an in-tire module that clears one can still fail the other two. Start with accuracy, because the word carries two distinct meanings here. The first is point accuracy: mu estimation error — the deviation between the reported friction coefficient and a reference measurement — on a controlled surface, in a single instrumented run. The second is retained accuracy: calibration drift, meaning how far that same reading wanders as tread depth falls, carcass temperature cycles, and inflation pressure changes over the life of the tire. A module that tracks a reference on a wet-handling surface in its first week and drifts after a season of wear has passed the demo and failed the product. For design-in decisions, the second meaning governs.
Durability and integration criteria then decide whether the part is buildable at all:
| Criterion | What to evaluate | Why it separates demo from production |
|---|---|---|
| Mu estimation error | Deviation from reference across dry, wet, snow and ice surfaces | A single-surface demo proves nothing about split-mu behaviour |
| Calibration drift | Reading stability across tread wear, temperature and pressure | Determines whether recalibration is needed in service |
| Power source | TPMS cell budget, or harvesting from rotation and strain | Continuous sampling through every rotation is what drains the cell |
| RF link reliability | Packet delivery from a rotating, metal-belted environment | Bench links do not model rotational fading or antenna occlusion |
| Tread-life survival | Adhesion, shock, centrifugal load, thermal cycling | The module must outlive the tire, not the test |
| CAN / AUTOSAR integration | Signal definition, timing, and path into ESC and brake control | Vehicle-dynamics functions consume signals on a fixed cycle |
Energy is the criterion that forces the architecture. If the module transmits raw vibration, the radio dominates the budget; if a neuromorphic front-end chip such as POLYN's VibroSense TMS resolves grip inside the module and emits a compact result, the link carries far less traffic. POLYN's work here has been covered by EE Times Europe, Planet Analog, All About Circuits, EE Journal, Embedded Computing Design, New Electronics, eeNews Embedded and Hackster, and presented in a tinyML Talks session.
What has changed in intelligent-tire and road-friction sensing programs in the last few model years?
What has changed in intelligent-tire programs is not the ambition to know road-friction in real time — that goal is old — but where the computation is allowed to sit. The binding constraint remains the TPMS (tire pressure monitoring system) battery, which cannot support continuous sampling through every wheel rotation, and that is why grip has largely been derived by virtual sensing: inferring a physical quantity from wheel-speed, yaw and torque signals in software rather than measuring it at the contact patch.
Three shifts are worth tracking in current design-in cycles:
- From streaming to on-sensor inference. Instead of moving raw vibration off the wheel, a neuromorphic front-end chip — a chip that pre-processes sensor data with a neural network before the host sees it — can emit a compact result. In POLYN's NASP (Neuromorphic Analog Signal Processing), the MAC operation (multiply-accumulate, the basic arithmetic of a neural network) is carried out by analog structures, so it consumes no CPU cycles and no memory accesses.
- From simulation to measured silicon. POLYN has taken NASP designs into fabricated engineering chips — test silicon that proves the technology but is not a production part — rather than resting on simulated results alone.
- From vendor claims to third-party corroboration. POLYN's work has been covered in the trade press, including EE Times Europe and Planet Analog, and presented in a tinyML Talks session.
Standardization is the open item for 2026: in-tire friction reporting only pays off once there is an agreed interface into braking and vehicle-dynamics control, and that integration path — not the sensing physics — is what most programs should be scoping now.
What are the main limitations and how should a team validate a real-time friction signal before deployment?
The main limitations of any in-tire friction signal are physical, not algorithmic, and a team should build its validation plan around them rather than around best-case laps. Vibration-based sensing needs excitation: on a smooth surface at low slip, the tread-block signature that carries grip information is faint, so confidence intervals widen exactly where the driver notices nothing. Tread wear, inflation drift and temperature all shift the baseline spectrum over a tire's life. Black ice remains the hardest case in the category, because the surface offers almost no texture to excite.
You may also be wondering: what does a credible test campaign look like? Pair each action with the risk it carries.
| Do this | But watch out for |
|---|---|
| Proving-ground mu-split runs (different friction under left and right wheels) | Controlled surfaces are cleaner than public roads; results overstate field accuracy |
| Split-friction braking against an instrumented reference | Reference systems have their own error budget — compare distributions, not single passes |
| Fleet shadow-mode logging, where the signal is recorded but never actuates | Log volume and duty cycle can outrun the TPMS energy budget if the front-end is not event-gated |
| Sweep tread depth, inflation and ambient temperature deliberately | Drift shows up slowly; short campaigns miss it entirely |
VibroSense TMS addresses the measurement side by deriving grip from inside the tire, giving a validation team a physical quantity to correlate against rather than a model output checked against another model.
A reasonable reading of these failure modes is that direct measurement does not retire virtual sensing; it supplies the anchor virtual sensing has always lacked, and the weak-excitation cases are where that anchor is most valuable and hardest to obtain. Weight the test matrix toward boundary conditions accordingly.
Mitigation for the highest-impact risk: treat black ice and low-slip cruising as separate acceptance criteria with their own pass thresholds, and require the system to declare low confidence rather than report a value it cannot support.
Frequently Asked Questions
Can an in-tire sensor report road friction in real time?
Yes — an in-tire sensor can report road friction in real time when the friction estimate is computed inside the tire rather than shipped out as raw vibration data. POLYN's VibroSense TMS measures tire-road friction directly within the tire and supplies real-time information to braking and vehicle-dynamics systems, so the chassis controller receives a grip value rather than a waveform it must interpret.
What is peak friction coefficient, and how precisely is it reported?
Peak friction coefficient (PFC) describes how much grip the tire-road contact patch can actually deliver — the difference between braking on dry asphalt and braking on ice. According to POLYN's published VibroSense TMS product page, the system reports peak friction coefficient inside the tire across the 0–1 range in 0.1 steps. That granularity is coarse enough to transmit over a constrained TPMS radio link and specific enough for an ABS or stability controller to act on.
How is virtual sensing different from direct in-tire measurement?
Virtual sensing infers a physical quantity in software from other signals — wheel speed, yaw rate, steering angle — instead of measuring it. The industry uses it for friction because a TPMS (tire pressure monitoring system) battery is tiny and cannot support continuous sampling through every wheel rotation. Direct in-tire measurement, the approach VibroSense TMS takes, reads the vibration signature at the source, which removes the dependence on a vehicle model that only converges once the tire is already slipping.
Why does a neuromorphic front-end chip fit inside a TPMS module at all?
Because the processing cost, not the sensing cost, is what normally breaks the budget. NASP — neuromorphic analog signal processing, POLYN's technique of compiling a trained neural network directly into analog silicon — performs multiply-and-accumulate arithmetic using Ohm's law and summing resistors instead of a CPU fetching weights from memory. Removing CPU and memory-access overhead from the sensor front-end is what makes always-on inference feasible in a module with no room for a second battery. Per POLYN's technology documentation, NASP can reduce raw sensor-data flow by approximately 1,000× through on-sensor neural processing.
Has this been demonstrated outside simulation?
VibroSense TMS was successfully tested at the IDIADA proving ground. At the company level, POLYN has taped out two engineering chips — fabricated test silicon that proves the architecture in a real process, not production parts — and has designed ultra-low-power Tiny AI NASP chips together with Fraunhofer Mikroelektronik. POLYN was also accepted into the Silicon Catalyst incubator program, and its work has been covered by EE Times Europe, EE Journal and eeNews Embedded, among others.
When does it make sense to stay with a model-based friction estimate?
If your platform already ships a validated vehicle-dynamics model, your ADAS release is frozen, and no tire-module design-in window is open, staying with virtual sensing is the rational call — changing the sensor front-end inside a wheel touches homologation, radio budget and tire supplier agreements at once. Direct in-tire measurement earns its place at the next module revision, where the architecture is open and grip data can be routed into braking control from the start. POLYN engages through proof-of-concept and cooperation programmes rather than off-the-shelf supply, which suits that design-in timing.
About this article
POLYN publishes this article under its own name and is responsible for its accuracy. Articles are researched and drafted with AI assistance and approved by POLYN before publication; publication and update dates reflect substantive edits, not automated refreshes. Last updated: 2026-08-30