Researching road condition with a phone: how Xinexis started

A bump in the road raises a surprisingly difficult engineering question: what, exactly, did the phone measure? The surface matters, but so do the tyres, suspension, speed and phone mount. A useful road-monitoring product has to preserve enough context to investigate the response.

At Xinexis, we are building Road Monitor to explore whether ordinary journeys can contribute useful evidence about road condition. Our starting point is a working iPhone recording and review prototype. We have begun collecting exploratory recordings; a validated pothole detection system and a multi-driver field study are still ahead.

Begin with the measurement chain

The phone measures its own motion after forces have travelled through the vehicle and its mount. That makes stable placement a research variable, not just an installation preference. A loose phone can introduce motion that has little to do with the surface beneath the wheels.

We consulted established road-anomaly research and platform documentation to shape the prototype. The Pothole Patrol study (opens in a new tab) examined sensor placement, labelled road events and repeated observations across vehicles. It provides useful precedent for investigating these factors. Its specialised equipment, dataset and reported performance do not establish the accuracy of our phone implementation.

Our own design separates three questions: whether the recording is usable, what motion occurred, and what caused it. Each requires different evidence.

Preserve what was recorded

Road Monitor stores motion, location, timestamps and acquisition context locally. Original recording chunks retain integrity information, so later review can check that the accepted files have not changed. Derived charts and analysis are built from those originals.

This choice matters when a method changes. We can revisit an earlier observation, examine timing or apply a revised analysis without pretending that the revised result was the original measurement. We also distinguish the motion values supplied by iOS from untouched hardware output.

Apple documents gravity and user acceleration (opens in a new tab) as separate Core Motion quantities. Our vibration view uses the gravity-removed acceleration and the gravity direction to examine signed vertical motion. That is a description of phone movement, not a direct measurement of a hole's depth.

Make uncertainty visible early

A line on a chart can look convincing even when the underlying recording contains gaps. We therefore check timing, sample continuity and location quality before presenting a measurement. An unavailable window stays unavailable. A jolt without a suitable nearby GPS fix keeps its time and strength without acquiring an invented map position.

The same principle shapes interpretation. A sharp response can be worth investigating while its cause remains unresolved. A manhole, expansion joint, speed bump or handling event may produce a noticeable signal. Amplitude alone cannot settle the classification.

Test the software, then test the road hypothesis

We use synthetic recordings with known frequencies, pulses and missing intervals to check calculations and failure handling. Native and server checks also exercise recording integrity and repeated uploads. These tests tell us whether the software behaves as intended under specified inputs.

Field validation has a different job. Our next experiments need consistent mounting, repeated passes in the same direction, recorded speed and vehicle context, and independently identified surface features. We need ordinary road sections and confusing alternatives as well as known defects. Evaluation must include roads and vehicles that were not used to tune the method.

A useful result is evidence someone can examine

For a future road-maintenance workflow, the ambition is a reviewable location supported by dated observations, usable coverage and an explanation of uncertainty. Whether that helps prioritise inspection is a question the pilot must answer with operational users.

Today, the research foundation is capture, replay and descriptive vibration analysis. We are building the evidence needed to decide what reliable inference can follow. If you work with fleet or road-maintenance data, discuss a research or pilot use case with Xinexis.