What risk signals are we missing?

The improvement in European road safety over the past decades is a major achievement that deserves recognition. Advancements in vehicle safety technology and a heightened awareness of how driver behaviour impacts outcomes have been the primary contributory factors to this progress. However progress has been uneven and, in many places, slower than hoped. Around 20,000 people are still killed on European roads each year, and an estimated 100,000 are seriously injured. While fatalities have declined over time, recorded serious injuries have fallen more slowly and unevenly — and in some countries, they have increased. The recent plateau in safety progress is prompting a renewed focus on the road environment as a potential missing variable. While researchers and authorities have long recognised that this environment likely plays a decisive role in safety, the ability to analyse this relationship has been fundamentally constrained by a lack of systematic, network-wide data. This has left a critical hypothesis regarding road risk largely unexamined, not due to a lack of interest, but due to a lack of evidence at scale.

Road infrastructure — from signage and roadside features to the condition of the road itself — is recognised as a core component of road safety alongside road user behaviour (e.g. speeding, alertness, intoxication) and vehicle safety technology (from seatbelts to advanced driver assist and autonomous features). In practice, however, safety work related to infrastructure tends to prioritise factors that are easier to define, regulate and enforce, such as speed limits and junction design. Many infrastructure-related risks are harder to assess consistently at network scale, which creates a practical gap in day-to-day safety prioritisation.

One such risk is road surface condition. While it is widely acknowledged to affect vehicle dynamics — including grip, braking performance and stability — it has largely been handled through maintenance processes rather than treated as an explicit input to safety analysis. That is partly because surface condition has been difficult to observe and compare systematically across entire road networks, and therefore difficult to link consistently with crash, fatality and other related safety data.

What is changing is not the fundamentals of road safety, but the tools available to make road surface condition visible at scale. Systems such as Univrses' 3DAI™ use AI to transform visual data from everyday vehicle fleets into consistent, network-wide road intelligence. This makes it possible to observe and compare road surface condition across networks more continuously, allowing it to be treated as a practical, safety-relevant input rather than a background maintenance concern.

What new research shows

Several empirical studies have analysed how measurable road surface characteristics relate to crash outcomes, using large-scale datasets that combine traffic, pavement and accident records. This allows surface condition to be examined retrospectively as a network-level safety variable.

Crash risk rises sharply on deteriorated surfaces

One recent quantitative study analysed detailed crash and pavement data from the US state of Iowa, linking geolocated accident records to measured surface condition across thousands of individual road sections. Each section was classified by pavement roughness and skid resistance, and crash frequency was calculated as crashes per 100 million vehicle miles travelled.

The differences were pronounced. Road sections classified as being in poor pavement condition recorded 466 crashes per 100 million vehicle miles travelled, compared with 134 crashes on sections in good condition. When skid resistance was examined separately, the contrast was even stronger: sections with poor measured friction recorded 571 crashes per 100 million vehicle miles travelled, compared with 118 crashes on sections with good friction.

The study is observational, but the pattern is consistent. Across both roughness and friction measures, worsening surface condition was associated with markedly higher crash frequencies, particularly on lower- and medium-speed roads that make up a large share of everyday traffic.

Consistent signals at network scale

These findings are reinforced by a larger, multi-state analysis conducted by the US Federal Highway Administration. Drawing on pavement friction measurements and crash data from more than 55,000 road segments across five states, the study examined how changes in measured friction relate to crash frequency across different road types and locations.

Across the full dataset, higher friction values were consistently associated with lower crash rates. On average, a 10-unit increase in measured friction corresponded to an approximate 10% reduction in total crashes. At locations where surface grip is especially critical — including curves, intersections and ramp access points — the estimated reductions were substantially larger, in several cases exceeding 20–30%. The analysis also showed that friction has a stronger relationship with crash occurrence in wet conditions than in dry ones.

Taken together, the evidence shows that surface characteristics such as friction and pavement condition are measurable attributes that consistently correlate with crash risk across different road types and geographies.

AI is making surface-related risk measurable at scale

For many road authorities, surface condition has traditionally been assessed through snapshots. Roughness and friction data exist, but they are often collected through periodic inspections or isolated tests, making it difficult to maintain a consistently updated and comparable view of conditions across an evolving network.

Artificial intelligence is beginning to change that equation. By analysing large volumes of vehicle-captured data, visual and other forms, AI-based approaches can convert what vehicles "see" and "feel" into structured, comparable indicators of road surface condition. Instead of relying on occasional checks, surface-related signals can be observed more frequently over time.

3DAI™: Network-wide road intelligence

Univrses' 3DAI™ is an AI-powered road monitoring system that converts various vehicle data sources, including camera imagery from everyday vehicles, into structured, geo-referenced data about road surface condition as well as the surrounding road environment. Rather than relying on dedicated survey vehicles or infrequent inspections, the system operates through cost-effective cameras and can be deployed across existing vehicle fleets, enabling continuous coverage of large road networks.

3DAI™ provides much richer context than surface performance metrics alone. While measures such as friction or skid resistance can indicate that surface performance has changed, they do not explain why. Through 3DAI™, authorities can observe physical road conditions directly — either by reviewing the imagery themselves, or by using the system's built-in AI to analyse the imagery and identify potholes, cracks, surface wear and other visible defects.

In a road safety context, this makes it possible not only to detect changes over time, but to understand what is driving them. When surface performance indicators shift, visual road data helps distinguish whether the cause is structural deterioration requiring maintenance, temporary conditions such as water on the road, or readings that do not require action. This reduces uncertainty and supports more targeted follow-up as part of broader safety and maintenance work.

The same principle extends beyond the road surface itself. Other observable elements of the road environment — such as signage, lighting and roadside assets — can also be captured and structured through 3DAI™. This supports a more systematic view of road infrastructure conditions that influence safety, making it easier for authorities to include these elements in proactive safety analysis.

How 3DAI™ works

1. Camera-based scanning. Vehicles equipped with 3DAI™'s camera-based vision system continuously scan road surfaces as they travel.
2. AI-powered detection. Computer vision models analyse the data to detect and classify surface-related features and defects, such as cracks, potholes and surface wear.
3. On-device edge processing. Key data is processed directly on the device, reducing file sizes, lowering transfer costs, and accelerating initial analysis.
4. Cloud refinement. Detections are uploaded to the cloud, where they're further structured and validated — ready to support accurate mapping and insight generation.
5. Geo-referenced mapping. Each detection is mapped with precise GPS coordinates, allowing conditions to be mapped accurately across the road network.
6. Trend analysis. As coverage accumulates, 3DAI™ highlights change patterns — helping teams see where deterioration is progressing and where follow-up may be needed.

Trusting AI in road safety decision-making

Integrating road surface data into safety-related analysis depends on delivering precise and repeatable insights that meet the standards of other trusted infrastructure inputs. Road authorities need confidence that the information they work with is reliable enough to be considered alongside engineering assessments, crash statistics and professional judgement in safety-related analysis.

Verified accuracy against established methods

3DAI™ has been benchmarked against established reference methods for road condition measurement. Independent evaluations conducted by the Swedish National Road and Transport Research Institute (VTI) show that the system's outputs align closely with traditional measurement techniques for surface condition.

Adapting to real-world road conditions

Unlike static survey methods, 3DAI™ operates continuously across diverse environments. As vehicles equipped with the system cover different road types, surface materials and weather conditions, the models are refined to reflect real-world variation.

Structured data for prioritisation

Beyond detection accuracy, 3DAI™ delivers road condition data in a structured and repeatable format. Networks are divided into comparable segments, making changes visible over time rather than captured as isolated observations.

By aggregating insights from repeated vehicle passes, the system highlights patterns of change rather than individual defects. Delivered via API or web interface, the data integrates with existing GIS and analysis workflows, supporting prioritisation and follow-up as a complementary source of evidence.

Proven in operational use

3DAI™ is already used by multiple cities and road transport authorities in Europe to support asset management workflows. These deployments show how network-scale road condition data can be integrated into day-to-day planning, prioritisation and follow-up.

With this foundation in place, a clear next step for 3DAI™ is to relate observed road surface conditions to reported safety outcomes. In Sweden, this means working towards linking 3DAI™'s road condition data with accident records held in Trafikverket's STRADA database — enabling analysis of how surface conditions and real-world incidents align over time.

See how cities and authorities are using 3DAI™ in practice:

• England had 110 000 streetlights classified in 90 days
• Trafikverket moved from annual road surveys to continuous monitoring across Sweden's network
• Mesta and Statens Vegvesen used 3DAI™ to detect signs and surface damage across 17,459 km of Norwegian roads
• Helsingborg reduced its potholes by 70% in five months
• Tauragė, Lithuania became the first city to use 3DAI™ for bus stop monitoring

A clearer way forward

Road surface condition has long been managed primarily as a maintenance issue, rather than treated as an explicit input to road safety analysis. Recent research makes that separation harder to sustain, showing consistent statistical associations between measurable surface characteristics — including pavement condition and skid resistance — and crash frequency across multiple datasets.

The practical shift is that surface-related risk no longer has to be inferred or addressed only through periodic cycles. When surface conditions can be observed more frequently and compared across a network, it becomes a usable input to proactive risk management, particularly on rural and urban roads where most fatalities occur and where conditions vary most.

For road authorities, this does not change the fundamentals of road safety or replace proven measures. Systems such as 3DAI™ illustrate how road surface conditions can be made more visible and comparable at network scale. This strengthens the evidence base for where to investigate further, where conditions are changing, and where follow-up should be prioritised alongside established inputs such as crash history, engineering assessments and local knowledge.

The opportunity is straightforward: better visibility at network scale makes it possible to manage surface-related risk with greater consistency and precision, supporting safer roads.

‍