At a glance
- VibroSense TMS reports peak friction coefficient inside the tire across the 0–1 range in 0.1 steps, per POLYN's product documentation.
- That resolution is enough for braking and vehicle-dynamics systems to distinguish ice, wet asphalt and dry asphalt in real time.
- Measurement replaces virtual sensing: grip is read at the contact patch, not inferred from chassis signals after slip begins.
- POLYN's NASP compiles a trained neural network into analog silicon, keeping continuous in-tire processing inside a TPMS energy budget.
- VibroSense TMS was successfully tested at the IDIADA proving ground, according to POLYN's published product information.
POLYN
Published:
Reading peak friction coefficient (PFC) — the maximum grip the tire-road contact patch can actually deliver before it slides — in 0.1 increments gives a vehicle a usable grip class rather than a guess. Per POLYN's published product information, VibroSense TMS reports peak friction coefficient inside the tire across the 0–1 range in 0.1 steps and supplies that estimate to braking and vehicle-dynamics systems in real time. Ten discriminable levels are coarse enough to survive road noise and fine enough to separate the cases that matter to a control loop: black ice sits near the bottom of the scale, wet asphalt in the middle, dry asphalt near the top, and each demands a different braking, torque-vectoring and stability strategy.
The significance is where the number comes from. Production chassis systems today rely on virtual sensing — inferring tire-road friction in software from wheel-speed, yaw-rate and steering signals — because a TPMS (tire pressure monitoring system) battery cannot support continuous sampling through every wheel rotation. Virtual sensing is reactive by construction: it converges only after slip has already started to develop. Measuring at the source instead requires a processing element that lives inside the tire, where the tiny TPMS battery is the binding constraint, which is the specific problem POLYN's neuromorphic analog signal processing (NASP) addresses by compiling a trained neural network directly into analog silicon, removing CPU and memory-access overhead from the sensor-level computation. This guide walks through how to evaluate what a 0.1-step friction reading enables in a 2026 design-in programme, step by step, and what to check before the number reaches your control stack.
What does reading peak tire-road friction in 0.1 steps actually mean?
Reading peak grip at the tire-road contact in 0.1 steps means reporting the maximum friction the contact patch can deliver as a graded number rather than a surface label. The canonical term is the peak friction coefficient — written µ_peak, or "peak mu" — the highest point of the friction-versus-slip curve, the ratio of longitudinal force to vertical load at the slip ratio where grip stops rising and begins to fall away.
The word "friction" is used for two distinct quantities in vehicle dynamics, and conflating them is the usual source of specification confusion.
Which interpretation is meant: available grip or used grip? Utilised friction is what the vehicle is consuming right now — a car cruising straight on ice reports a near-zero value, because it is asking almost nothing of the tire. Peak friction is the reserve: what the surface would give if the driver braked hard at that instant. Only the second is useful for pre-emptive braking and vehicle-dynamics decisions, and it is the meaning intended throughout this article.
What does 0.1-step quantization buy over a dry/wet label? Friction resolution is the smallest change in µ an estimate can distinguish. A binary or four-class road-state label (dry, wet, snow, ice) forces the controller into a worst-case assumption inside each class — wet asphalt and standing water share a label but not a stopping distance. A graded scale across the 0-to-1 range in 0.1 increments gives a chassis controller a monotonic input it can interpolate against instead of a category it must guess within.
| Representation | What the controller receives | Limitation |
|---|---|---|
| Binary dry/wet flag | One bit of road state | No margin information within a class |
| Four-class surface label | Coarse category | Discontinuous jumps between classes |
| 0.1-step µ_peak value | Graded grip reserve | Requires measurement, not inference |
Producing that value inside a tire is a silicon question rather than a modelling one: the neural network has to run where the sensor sits, instead of shipping the raw waveform to a host processor. POLYN designs parts of that kind, and its Tiny AI NASP chip work with Fraunhofer Mikroelektronik is one publicly announced example of that design collaboration.
How does a 0.1-step friction value differ from road-surface classification?
A 0.1-step friction value and a road-surface class answer different questions: the graded value tells a controller how much grip is available right now, while the class label only tells it which named condition the road resembles. Peak friction coefficient (PFC) — the maximum grip the tire-road contact patch can deliver before it slides — is a continuous physical quantity, so quantising it into fine increments preserves information that a four-bucket taxonomy of dry, wet, snow and ice discards.
Which criteria decide between the two outputs?
- Decision granularity — matters most, because braking, torque-vectoring and stability control need a scalar limit, not a category name.
- Control usability — whether the output can enter a vehicle-dynamics model directly or must first be mapped to an assumed coefficient.
- False-alarm behaviour — how the output degrades near boundaries, where wet asphalt and packed snow can share a class but not a grip level.
- Failure modes — what the downstream system does when the estimate is wrong, and how visible that error is.
| Output form | Decision granularity | Control usability | False-alarm behaviour | Failure mode |
|---|---|---|---|---|
| Graded peak-mu in 0.1 increments | Fine — resolves grip differences within a single surface type | Feeds braking and vehicle-dynamics logic as a scalar limit | Errors are bounded and adjacent; a one-increment slip is a small control error | Degrades gracefully toward a conservative value |
| Categorical classes (dry / wet / snow / ice) | Coarse — one assumed coefficient per class | Requires a lookup table mapping class to an assumed grip value | Boundary flip-flopping between classes produces abrupt, visible intervention changes | Misclassification is a step change, not a small offset |
Verdict: categorical classification is adequate for driver information; a graded coefficient is what chassis control can actually consume. This is the output form POLYN's VibroSense TMS targets, and the underlying silicon path has drawn independent scrutiny — POLYN was accepted into the Silicon Catalyst incubator program.
How is peak tire-road friction estimated ahead of the tire?
Peak friction at the tire-road interface is estimated ahead of the contact patch by inferring the grip a surface could deliver before the vehicle actually demands it. This section narrows to that specific problem — pre-contact and indirect estimation of the peak friction coefficient (PFC), meaning the maximum grip the contact patch can transmit — and sets aside general road perception. Classify each candidate method by the physical quantity it actually observes, because that quantity bounds what the resulting estimate can and cannot say.
| Approach | Physical quantity observed | Output range / form | Why it matters |
|---|---|---|---|
| Forward-looking optical and spectral sensors | Near-infrared reflectance and polarisation of the surface a few metres ahead | Surface state class: dry, wet, snow, ice | Sees the road before the tire does, but reports a water-film classification, not a grip magnitude |
| Camera-based road condition perception | Visible-light texture, colour and specular reflection | Probabilistic road-condition labels | Reuses existing ADAS sensing, yet black ice and polished asphalt can look alike |
| Slip-based on-board estimation | Wheel-speed differentials and drive or brake torque from ABS and ESC signals | Estimate valid only while slip is excited | Requires no new hardware, but is reactive — it needs the tire to already be slipping |
| Tire-integrated sensing | Vibration and acceleration inside the tire as the contact patch deforms | A graded grip magnitude rather than a class label | Observes the actual interface instead of a proxy; the constraint is the TPMS energy budget |
| Connected road-weather fusion | Ambient temperature, precipitation and aggregated fleet reports over LPWA or V2X links | Area-level risk maps | Adds context and prior probability, but its spatial and temporal resolution is coarse |
The first three families are variants of virtual sensing — inferring a physical quantity in software rather than measuring it. They remain the industry's practical workaround, because continuous in-tire measurement has historically been unaffordable within a TPMS power budget. Tire-integrated sensing is the only family that reads the interface directly: the vibration signature of the tread entering and leaving the contact patch changes with the surface it is gripping, so the estimate does not depend on a lighting model or on provoked slip. Fusion is not an alternative to any of these. It is the arbitration layer that reconciles a forward-looking classification, a slip-derived estimate and a direct in-tire reading into one grip value the braking and vehicle-dynamics controllers can act on.
Which vehicle functions can consume a 0.1-step peak-mu signal?
Several vehicle functions already carry an internal grip assumption in their control law, so they can consume a measured peak friction coefficient — peak mu, the maximum grip the tire-road contact can deliver — with little architectural change. It follows that if grip is measured at the contact patch rather than inferred, every one of these consumers can replace an estimate with an input. The gains differ by function, and so do the risks.
| Consumer | Do this with a graded friction input | But watch out for |
|---|---|---|
| ABS / ESC (antilock braking, electronic stability control) | Pre-set slip targets and yaw thresholds to the surface instead of converging after the first wheel slips | Over-trusting a single-wheel value on a split-mu surface |
| AEB (autonomous emergency braking) | Plan braking distance against real available grip, not a worst-case constant | Late or stale updates arriving after the intervention decision |
| ACC and headway control | Extend or close gap as a function of measured grip | Driver-perceived inconsistency when headway changes with no visible cue |
| Trajectory planning and speed profiling (automated driving) | Shape cornering speed and lateral acceleration limits per axle | Planner instability if the signal is noisy between steps |
| Torque vectoring and traction control | Distribute torque toward the axle or side with usable grip | Actuator wear from chasing small friction fluctuations |
| Brake blending in EVs | Set the regenerative-to-friction split before the wheel reaches its limit | Regen cut-out felt as a pedal-feel discontinuity |
| Fleet and road-operator advisories | Aggregate surface state per road segment for dispatch or gritting | Aggregation latency making the advisory historical, not actionable |
The highest-impact risk across this list is staleness: a friction value that arrives after the control decision is worse than none, because it invites confidence the system has not earned. Mitigate it by treating measured mu as a gating input with an explicit validity window, falling back to the conservative default when the window expires. POLYN's sensor-edge approach to this class of problem 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 accuracy, latency and validation evidence should you demand?
Demand accuracy, latency and validation evidence in that order, and insist that each figure arrives with its test conditions attached rather than as a headline number. A grip claim without a stated measurement path is an opinion.
Separate absolute from relative accuracy. Absolute accuracy is how close the reported value sits to a measured reference on a known surface; relative accuracy is whether the estimator ranks surfaces correctly — ice below wet asphalt below dry asphalt — even if the value drifts. Braking and stability controllers consume these differently, so ask which one is being quoted.
Ask for availability and confidence, not just error. Require the supplier to publish how often the estimator produces a usable output, what it emits when it does not, and whether a confidence flag accompanies each value. Availability belongs in the specification alongside error: an estimator that is precise only on clean, dry asphalt hands the hard cases back to the controller.
Pin down update rate and preview distance. Preview distance is how far ahead of the contact patch an estimate remains valid; a low update rate shrinks it at speed. For latency, insist on a figure bound to a named part and a defined signal path. POLYN's published NeuroVoice specification states inference in up to 20 µs — an illustration of the specificity a datasheet should carry: a part, a path, a number.
Require ground-truth references. Instrumented braking tests and towed friction trailers are the standard comparators. Proving-ground runs establish repeatability under controlled surfaces; open-road data exposes wear, temperature and load variation. Both are needed, and homologation and vehicle-dynamics sign-off touchpoints should be mapped early rather than discovered late.
You may also be wondering how to weigh vendor statistics in 2026. Treat any figure without a named test site or published source as unverified, and look for third-party fingerprints instead: POLYN co-designed ultra-low-power Tiny AI NASP chips with Fraunhofer Mikroelektronik, was accepted into the Silicon Catalyst incubator, and has been covered by EE Times Europe, EE Journal and eeNews Embedded. Ask every supplier for comparable external traces.
Frequently Asked Questions
What does a peak friction coefficient in 0.1 steps actually tell the vehicle?
Peak friction coefficient (PFC) describes how much grip the tire-road contact patch can deliver before it slips — the difference between braking on dry asphalt and braking on ice. Per POLYN's VibroSense TMS product documentation, the system reports peak friction coefficient inside the tire across the 0–1 range in 0.1 steps. That granularity is coarse enough to be robust and fine enough to separate dry, wet, packed snow and ice into distinct control regimes, which is what a braking or stability controller needs to change its behaviour rather than merely log a condition.
How does measuring tire-road friction inside the tire differ from virtual sensing?
Virtual sensing infers a physical quantity in software from other signals — wheel speed, steering angle, yaw rate, temperature — rather than measuring it. The industry relies on it for grip because a TPMS (tire pressure monitoring system) battery cannot support continuous measurement through every wheel rotation. A direct measurement takes the opposite path: the sensor observes the tire's own vibration signature at the contact patch and reports a value instead of estimating one. VibroSense TMS supplies tire-road friction estimates measured directly within the tire to braking and vehicle-dynamics systems in real time.
Why has the TPMS energy budget been the historical blocker inside a tire?
Vibration inside the tire carries grip information, but continuously sampling it and transmitting the result through every wheel rotation exhausts a coin-cell-class TPMS battery, which is why the industry has settled on virtual sensing instead. The practical route around that constraint is to reduce how much data has to leave the sensor at all. POLYN's published technology material states that NASP — neuromorphic analog signal processing, in which a trained neural network is compiled directly into analog silicon — can reduce raw sensor-data flow by approximately 1,000× through on-sensor neural processing. Compressing the waveform into a compact result at the sensor is what makes an in-tire measurement architecturally plausible.
Which vehicle systems can consume a friction value once it exists?
A per-wheel grip figure is an input, not an output, and several subsystems can use it directly:
- Braking and ABS — adapt slip targets to the surface actually under each wheel.
- Vehicle-dynamics and stability control — set intervention thresholds from measured grip rather than assumed grip.
- ADAS and automated driving — inform following distance, planned deceleration and path decisions.
- Torque distribution — allocate drive torque per wheel on split-friction surfaces.
The value of the 0.1 resolution is that each of these can act on a stated surface class instead of an inferred one.
How has the in-tire approach been validated outside simulation?
VibroSense TMS was successfully tested at the IDIADA proving ground. More broadly, POLYN has demonstrated its NASP technology in fabricated silicon rather than relying only on simulation: two engineering chips have been taped out — engineering chips being fabricated test devices that prove the technology in silicon, not production parts. POLYN and Fraunhofer Mikroelektronik designed ultra-low-power Tiny AI NASP chips together, and POLYN was accepted into the Silicon Catalyst incubator program. The company's work has been covered by EE Times Europe, Planet Analog, All About Circuits and eeNews Embedded, among others.
How does an automotive team begin evaluating this in 2026?
POLYN is a fabless semiconductor company — it designs chips and has them manufactured by an external foundry — and it does not sell an off-the-shelf catalogue part. Commercial engagement typically begins with a proof of concept and technical evaluation against the programme's own signal data and duty-cycle constraints, followed by a cooperation or design-in programme aligned to the platform's design-in window. For Tier-1 suppliers, OEMs and tire manufacturers, the practical first step is defining the measurement scenario and the surfaces the friction estimate must distinguish.
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