Why Safer Roads Are No Longer Enough: The Global Research Shifts Changing Road Safety
Engr. Kamran Abbas
Table of Contents
- Introduction
The Global Paradox: Progress Is Real, but the Target Is Still Far Away
The First Major Shift: From Preventing Crashes to Preventing Fatal and Serious Injury
Speed Is Moving From an Operational Variable to a Safety Variable
The Second Major Shift: Vulnerable Road Users Are Becoming the Design Benchmark
The Third Shift: Crash History Is No Longer Enough
Surrogate Safety Measures Are Changing What Engineers Can Observe
AI Is Moving Road Safety From Analysis Toward Prediction—But With Serious Limitations
Connected Vehicles Are Turning the Road Into an Information System
Digital Twins May Change How Engineers Test Roads Before Building Them
“Safe” Does Not Mean “Compliant”
The Implementation Gap May Now Be a Bigger Problem Than the Knowledge Gap
Road Safety Is Becoming a Network Problem Rather Than a Black-Spot Problem
The Research Frontier Is Increasingly About Interaction
What This Means for Transportation Engineers
What Should Come Before AI?
The Most Important Research Questions Are Changing
A Critical Warning: “Zero Deaths” Is Not the Same as Zero Crashes
The Next Frontier: From “Safe Road” to “Safe Transport System”
The Uncomfortable Conclusion
Research Snapshot: Where Road Safety Is Heading
Final Engineering Perspective
References and Further Reading
Road safety engineering is entering an uncomfortable new phase.
For much of the modern transportation era, the central engineering problem was relatively straightforward: identify hazardous locations, determine why crashes occur, modify the road, improve traffic control, enforce regulations, and measure whether crash numbers decline.
That approach has saved lives. It still does.
But the global evidence is forcing transportation engineers to ask a harder question:
What if a road can satisfy conventional engineering requirements and still produce an unacceptable level of death and serious injury?
The question matters because the global road-safety burden remains enormous even after decades of engineering, regulation and technological progress. The latest World Health Organization data show that approximately 1.16 million people were killed in road crashes in 2025. Road traffic injuries remain the leading cause of death among people aged 5–29, and more than half of road deaths occur among vulnerable road users, including pedestrians, cyclists and motorcyclists.
Yet there is also an important reason for optimism: WHO reports that global road deaths declined by 21% between 2011 and 2025, despite the addition of more than one billion motor vehicles to the world’s roads.
That combination creates the real research challenge.
Road safety is improving. But it is not improving fast enough, evenly enough, or systematically enough.
The emerging response is not simply to design “safer roads.”
It is to redesign the entire safety system.
The global paradox: progress is real, but the target is still far away
The global road-safety story should not be presented as either success or failure.
It is both.
There has been measurable progress. But the remaining burden is so large that incremental improvement is no longer sufficient.
The global road-safety picture
| Indicator | Latest global evidence | Why it matters |
|---|---|---|
| Road deaths annually | ~1.16 million | The absolute burden remains extraordinarily high |
| Change in deaths, 2011–2025 | −21% | Demonstrates that large-scale improvement is possible |
| Road deaths among vulnerable road users | >50% | Safety cannot be evaluated primarily from the driver’s perspective |
| Road deaths in low- and middle-income countries | ~92% | Road-safety capacity and infrastructure remain highly unequal |
| Leading cause of death, age 5–29 | Road traffic injuries | Road safety is also a major public-health problem |
| Non-fatal injuries annually | 20–50 million | Fatality statistics substantially understate the human burden |
| Global 2030 ambition | 50% reduction in deaths and serious injuries | Current progress requires substantially greater action |
Sources: WHO Global Status Report on Road Safety 2023 and WHO’s July 2026 road-safety update.
The numbers reveal something important.
The question is no longer whether road engineering works.
It clearly does.
The more difficult question is:
Why does proven knowledge fail to produce sufficiently safe outcomes when applied across entire road networks?
That question moves road safety from a purely technical discipline toward a systems-engineering, governance, behavioural and public-health problem.
The first major shift: from preventing crashes to preventing fatal and serious injury
The most important conceptual transformation is the rise of the Safe System approach.
Traditional road-safety thinking often placed considerable emphasis on preventing crashes. If drivers complied with traffic rules, road geometry was appropriate, signs were visible and traffic control operated correctly, the system was assumed to be reasonably safe.
The Safe System approach begins from a different premise:
Humans make mistakes.
Therefore, a road network that depends on perfect human behaviour is inherently fragile.
The objective becomes not simply:
How can we eliminate crashes?
but:
How can we ensure that foreseeable human errors do not result in death or serious injury?
The International Transport Forum describes Safe System thinking around several interconnected principles: humans make mistakes, humans are physically vulnerable, responsibility is shared, and the system should contain multiple layers of protection so that failure of one component does not produce a fatal outcome.
FHWA similarly describes the approach through safe road users, safe vehicles, safe speeds, safe roads and post-crash care.
This is not merely a change in terminology.
It changes the engineering objective.
Traditional engineering logic
Crash → investigation → countermeasure
Emerging safety logic
Exposure → risk → conflict → crash → injury severity → survivability
The engineer is therefore increasingly interested not only in whether a crash happens, but also in:
- the kinetic energy involved;
- the speed at impact;
- the mass difference between road users;
- the probability of conflict;
- the availability of escape space;
- roadside hazards;
- pedestrian exposure;
- cyclist separation;
- motorcycle interaction;
- vehicle crashworthiness;
- emergency response;
- and the ability of the human body to survive the resulting forces.
That is a much broader engineering problem.
Speed is moving from an operational variable to a safety variable
One of the clearest examples of this philosophical change is speed.
Transportation engineering has historically treated speed as an important operational parameter. Higher speeds can reduce travel time and improve network mobility.
But Safe System research asks a different question:
At what speed can a human being realistically survive the conflicts created by the road environment?
WHO states that increasing mean speed increases both crash probability and crash severity. Its current road-safety guidance reports that every 1% increase in mean speed is associated with approximately a 4% increase in fatal-crash risk and a 3% increase in serious-crash risk.
The implication is profound.
A posted speed limit is not automatically a safe operating speed.
A road can have a legally posted limit and still create an environment in which vehicles routinely travel too fast for the surrounding human activity.
This is why modern safety thinking increasingly considers:
speed limit + road geometry + roadside environment + user mix + enforcement + vehicle technology
as a combined system.
A speed limit sign alone cannot physically change the kinetic energy of a vehicle.
Road design can.
This is why contemporary Safe System guidance increasingly supports designing the physical environment so that desired speeds emerge naturally rather than relying entirely on driver compliance.
The second major shift: vulnerable road users are becoming the design benchmark
For decades, many road-performance metrics were implicitly vehicle-centric.
Traffic engineers measured:
- capacity;
- delay;
- level of service;
- travel time;
- queue length;
- vehicle throughput;
- intersection performance.
These measures remain useful. But they are insufficient. A road can move thousands of vehicles efficiently while remaining extremely dangerous for people walking or cycling across it. That is why vulnerable road users are increasingly becoming a design stress test for the transportation system.
WHO’s 2026 data indicate that more than half of global road deaths involve pedestrians, cyclists and motorcyclists. Motorcyclist deaths alone now account for nearly one-third of global road fatalities, while the number of motorcycles on the world’s roads more than tripled between 2011 and 2025.
This is particularly significant for developing and rapidly motorizing regions.
A road designed primarily around passenger-car movement can become increasingly unsafe as motorcycles, bicycles, pedestrians, e-bikes, e-scooters and other mobility modes enter the same physical space.
The engineering response is not simply to tell vulnerable users to “be careful.” It is to reduce the opportunity for high-energy conflicts.
That means considering:
- physically separated cycle facilities;
- continuous sidewalks;
- protected pedestrian crossings;
- refuge islands;
- appropriate crossing distances;
- intersection visibility;
- lighting;
- speed management;
- motorcycle-specific treatments;
- safe roadside design;
- accessible facilities for people with disabilities;
- and separation of incompatible traffic streams.
WHO’s Global Plan explicitly calls for infrastructure that protects all road users, beginning with the most vulnerable, and identifies sidewalks, safe crossings, cycle paths, motorcycle facilities, safe roadsides, median separation and appropriate speed management as components of safe infrastructure.
The critical change is therefore this:
The road is no longer judged solely by how safely a competent driver can use it. It is increasingly judged by how safely the least-protected person can use it.
The third shift: crash history is no longer enough
One of the most consequential methodological changes in road safety is the move from reactive safety analysis toward proactive and predictive analysis.
The traditional approach is intuitive:
- collect crash data;
- identify high-crash locations;
- investigate causes;
- install a countermeasure.
The problem is obvious.
A location may need to accumulate crashes before it appears dangerous.
That creates an uncomfortable ethical and statistical problem:
Why should the system wait for severe crashes before recognizing risk?
Modern safety analysis therefore increasingly combines historical crashes with other forms of evidence.
These may include:
- traffic exposure;
- roadway geometry;
- operating speed;
- intersection configuration;
- pedestrian and bicycle volumes;
- traffic conflicts;
- lighting;
- weather;
- roadside characteristics;
- vehicle composition;
- land use;
- behavioural observations;
- and other roadway attributes.
The FHWA Highway Safety Manual provides a formal framework for quantitative safety analysis, including predictive methods and Crash Modification Factors (CMFs) for estimating the expected safety effects of infrastructure treatments.
This matters because observed crash counts are not perfect measurements of underlying risk.
FHWA specifically identifies problems such as regression to the mean, random variability in crash frequency and changes in roadway characteristics as reasons why relying solely on recent crash history can be misleading.
The result is a fundamental methodological transition:
Reactive safety
Where have crashes happened?
Proactive safety
Where are dangerous interactions occurring?
Predictive safety
Where is serious risk likely to emerge—even before crashes accumulate?
That is one of the most important research shifts in modern transportation safety.
Surrogate safety measures are changing what engineers can observe
There is another important development underneath proactive safety analysis: the use of surrogate safety measures. A crash is a relatively rare event. That makes crashes statistically valuable but operationally inconvenient. An engineer cannot deliberately wait for hundreds of crashes to understand how an intersection behaves.
Instead, researchers increasingly study near-conflicts and traffic interactions.
Examples include:
- Time to Collision (TTC);
- post-encroachment time;
- deceleration rate;
- gap acceptance;
- trajectory conflicts;
- hard braking;
- lane-change interactions;
- pedestrian-vehicle conflicts.
Traffic simulation research increasingly uses such measures to assess safety without waiting for actual crashes. A 2024 review of traffic-simulation research found growing use of microscopic simulation and surrogate safety measures to evaluate risks associated with speed, signal timing, geometry and driver behaviour.
This creates a powerful possibility:
The engineer can study dangerous interactions before they become crashes.
That does not make surrogate measures equivalent to crashes. They require validation and careful interpretation. But they allow safety analysis to move closer to the actual mechanisms that generate risk.
AI is moving road safety from analysis toward prediction—but with serious limitations
Artificial intelligence is one of the most visible changes in transportation research. Machine-learning models are increasingly being investigated for:
- crash-frequency prediction;
- crash-severity prediction;
- real-time crash-risk estimation;
- traffic-conflict detection;
- image-based infrastructure inspection;
- driver-behaviour analysis;
- pedestrian detection;
- anomaly detection;
- and integration of heterogeneous transportation datasets.
A systematic review published in Accident Analysis & Prevention examined machine-learning approaches for crash occurrence, crash frequency and injury-severity prediction and identified both significant potential and important methodological gaps.
A separate 2024 review identified 95 studies using non-visual machine-learning approaches for road-traffic accident prediction across areas such as risk level, occurrence, frequency and severity. The research direction is clear.
Instead of asking only:
What happened?
AI increasingly allows researchers to ask:
What conditions are associated with elevated risk?
And potentially:
What is likely to happen next?
But this is precisely where transportation engineers should become more—not less—critical.
Prediction is not causation
Suppose an algorithm identifies an intersection as high-risk.
That does not automatically tell the engineer whether the dominant mechanism is:
- excessive approach speed;
- poor sight distance;
- inadequate pedestrian visibility;
- signal timing;
- turning conflicts;
- lane geometry;
- heavy-vehicle interaction;
- poor lighting;
- or behavioural adaptation.
An algorithm may find correlation without identifying the physical mechanism responsible.
There are also problems involving:
- biased crash datasets;
- inconsistent reporting;
- class imbalance;
- data leakage;
- geographic transferability;
- model drift;
- explainability;
- and poor performance when models are deployed outside the conditions in which they were trained.
Therefore the correct relationship is:
AI → evidence → engineering interpretation → intervention
not:
AI → automatic engineering decision
The strongest future safety systems will probably combine machine intelligence with transportation engineering judgment rather than attempt to eliminate the engineer from the process.
Connected vehicles are turning the road into an information system
Another major shift is occurring at the boundary between infrastructure and vehicles.
Historically, the road supplied a largely passive environment:
Road → signs → markings → signals → driver
Connected transportation changes this relationship.
The emerging architecture is closer to:
Road ↔ Vehicle ↔ Infrastructure ↔ User ↔ Network ↔ Data
Vehicle-to-Everything (V2X) technologies can enable vehicles and infrastructure to exchange information relevant to safety and traffic operations.
NHTSA research has examined connected-vehicle systems involving vulnerable road users and technologies such as Cellular-V2X, including their potential use in safety-related applications.
Potential applications include:
- collision warnings;
- intersection alerts;
- work-zone warnings;
- emergency-vehicle priority;
- vulnerable-road-user detection;
- signal information;
- road-hazard communication;
- cooperative perception;
- and automated safety responses.
But connected infrastructure introduces another engineering reality:
Digital safety depends on physical safety.
A connected intersection with poor geometry is not automatically a safe intersection. A vehicle receiving a warning does not eliminate the consequences of excessive speed. A pedestrian detection system cannot compensate indefinitely for a fundamentally dangerous crossing arrangement. Technology should therefore be treated as a layer of protection, not as a substitute for sound infrastructure.
Digital twins may change how engineers test roads before building them
One of the most promising research directions is the use of digital twins and high-fidelity simulation.
The basic idea is straightforward:
Create a digital representation of a real road, intersection or network and use it to test alternative conditions before implementing them physically.
For example, an engineer could simulate:
- different speed limits;
- lane arrangements;
- signal timing;
- pedestrian demand;
- weather;
- visibility;
- vehicle mix;
- traffic demand;
- geometric modifications;
- and different control strategies.
Recent research has begun combining microscopic traffic simulation, vehicle dynamics and environmental conditions within digital-twin frameworks for safety analysis. A 2025 Scientific Reports study demonstrated a digital-twin framework incorporating vehicle dynamics and environmental factors and using surrogate safety measures such as Time to Collision and Deceleration Rate to Avoid a Crash.
The significance is not that digital twins are automatically superior.
It is that they allow engineers to ask:
What happens if we change the system before we physically change the system?
That could substantially improve design-stage safety evaluation.
But digital twins have the same fundamental weakness as AI:
A sophisticated model is only as credible as its assumptions, calibration and data.
False precision is still error.
“Safe” does not mean “compliant”
Perhaps the most important distinction for practicing engineers is the difference between design compliance and safety performance.
A design can satisfy:
- geometric standards;
- sight-distance requirements;
- lane-width requirements;
- signage requirements;
- pavement-marking requirements;
- traffic-control requirements;
and still produce an undesirable safety outcome.
That does not mean standards are unnecessary.
It means standards are not the same thing as a guarantee of safety.
FHWA’s IHSDM illustrates this evolution. Rather than relying solely on policy compliance, it incorporates quantitative safety-performance analysis, including crash prediction and design-consistency assessment.
This distinction should become fundamental in modern road design:
Compliance asks whether the design meets the rule. Safety analysis asks what the design is likely to do.
The two questions overlap.
They are not identical.
The implementation gap may now be a bigger problem than the knowledge gap
This is perhaps the most uncomfortable conclusion. The global road-safety community already knows many effective interventions. We know that excessive speed increases crash severity. We know that vulnerable road users require appropriate infrastructure. We know that helmets, seat belts and child restraints save lives. We know that impaired driving is dangerous. We know that safer vehicles matter. We know that road design influences crash risk. We know that post-crash response affects outcomes.
WHO’s 2026 global update explicitly emphasizes that countries already know many of the interventions that work and calls for stronger implementation of safe infrastructure, safer speeds, safer vehicles, stronger laws and enforcement.
This suggests that the central constraint may increasingly be implementation capacity.
The problem is therefore not simply:
“What new technology should we invent?”
It is also:
“How do we consistently implement proven safety interventions across millions of kilometres of roads?”
That is a governance problem. It is a funding problem. It is a data problem. It is an institutional problem. And, in many places, it is an engineering-management problem.
Road safety is becoming a network problem rather than a black-spot problem
Another important shift is from treating isolated dangerous locations toward understanding systemic risk across road networks. A black-spot programme may identify ten intersections with unusually high crash counts. A systemic approach asks a broader question:
What characteristics are repeatedly producing dangerous conditions across the network?
For example:
- urban arterials with high pedestrian activity;
- rural curves with excessive operating speed;
- uncontrolled crossings near schools;
- motorcycle-heavy corridors;
- intersections with large turning volumes;
- roads with insufficient roadside recovery areas.
This changes the scale of intervention.
Instead of waiting for each individual location to develop a crash history, agencies can identify common risk patterns and treat many locations systematically.
The FHWA Roadway Safety Data Program explicitly includes systemic approaches, predictive methods and data-driven safety analysis as components of modern roadway safety management.
The future road-safety engineer therefore increasingly needs to think in terms of:
network exposure + risk patterns + vulnerable users + systemic countermeasures
rather than simply:
high-crash location + local fix.
The research frontier is increasingly about interaction
The next generation of road-safety research will probably not be dominated by one technology. It will be dominated by interaction between systems.
Consider a modern urban corridor. A pedestrian is detected by a roadside camera. The pedestrian’s trajectory is analysed by an AI model. The intersection controller receives the information. The connected vehicle receives a warning. The vehicle’s driver-assistance system detects the pedestrian. The traffic signal changes its timing. The road design provides a protected crossing. The vehicle is travelling at a speed compatible with pedestrian survival. Emergency services are automatically notified if a collision occurs. No single component guarantees safety. The protection comes from redundancy. That is exactly what Safe System thinking is trying to achieve. The system should remain protective even when one component fails.
The evolution of road-safety thinking
TRADITIONAL MODEL
Crash
↓
Investigate
↓
Identify black spot
↓
Engineer countermeasure
↓
Monitor crashes
↓↓↓ EVOLUTION ↓↓↓
PROACTIVE MODEL
Roadway + Traffic + Speed + Exposure + Conflicts
↓
Risk assessment
↓
Predictive safety analysis
↓
Targeted intervention
↓
Evaluation
↓↓↓ EMERGING MODEL ↓↓↓
SAFE-SYSTEM MODEL
Human error
+
Human vulnerability
+
Safe speeds
+
Safe infrastructure
+
Safe vehicles
+
Data / AI / V2X
+
Post-crash care
↓
Multiple layers of protection
↓
SURVIVABLE TRANSPORT SYSTEMThis evolution is more important than any individual technology.
What this means for transportation engineers
For practicing road and transportation engineers, the shift has practical consequences.
The engineer of the future will need to be comfortable working across several traditionally separate domains.
1. Geometry
Horizontal and vertical alignment, sight distance, intersection geometry, roadside design and access management remain fundamental.
But their safety implications must increasingly be evaluated quantitatively.
2. Traffic operations
Capacity and delay still matter.
But operational efficiency should increasingly be considered alongside speed distribution, conflict exposure and vulnerable-road-user safety.
3. Human factors
Engineers need to understand how people actually perceive, react and behave—not merely what regulations assume they will do.
4. Data analytics
Crash databases will remain important, but they will increasingly be combined with:
- trajectory data;
- speed data;
- video analytics;
- connected-vehicle data;
- infrastructure inventories;
- weather;
- land-use information;
- and exposure measurements.
5. Predictive methods
Engineers should understand safety-performance functions, crash modification factors, empirical-Bayes methods and surrogate safety measures rather than relying exclusively on historical crash counts.
6. Vulnerable-road-user design
Walking, cycling, motorcycles, accessibility and emerging micromobility cannot remain secondary considerations.
7. Technology evaluation
AI, V2X and digital twins should be evaluated according to measurable safety benefits—not novelty.
What should come before AI?
This question deserves a blunt answer.
If an intersection has:
- no usable sidewalk;
- uncontrolled pedestrian crossings;
- excessive vehicle speeds;
- poor visibility;
- inadequate lighting;
- confusing geometry;
- weak enforcement;
then deploying an AI platform may be the wrong first investment.
A sophisticated prediction system cannot compensate for a fundamentally unsafe physical environment.
The hierarchy should therefore be:
1. Proven safety principle
Identify the actual safety problem.
2. Appropriate engineering intervention
Modify the physical or operational system.
3. Enforcement and governance
Ensure the intended behaviour is supported and maintained.
4. Technology
Use AI, sensors, V2X or automation when they provide measurable additional protection.
5. Continuous evaluation
Measure whether the intervention actually changes safety performance.
In simplified form:
PROBLEM
↓
RISK MECHANISM
↓
ENGINEERING RESPONSE
↓
IMPLEMENTATION
↓
MEASUREMENT
↓
TECHNOLOGY — WHERE IT ADDS VALUE
↓
FEEDBACKThat is a much more defensible engineering philosophy than “technology first.”
The most important research questions are changing
The questions being asked in road-safety research are increasingly different from those of previous decades.
Old question
Where do crashes happen?
New question
Where does dangerous exposure occur?
Old question
What caused the last crash?
New question
What system conditions repeatedly create severe conflict?
Old question
How can we make drivers safer?
New question
How can the transport system remain safe when drivers inevitably make mistakes?
Old question
Does the design comply with the standard?
New question
What safety performance is the design expected to produce?
Old question
Can technology detect the problem?
New question
Does technology measurably reduce the risk created by the problem?
Old question
How do we reduce crashes?
New question
How do we prevent crashes from becoming fatal or seriously injurious?
That is the intellectual shift behind the modern Safe System.
A critical warning: “zero deaths” is not the same as zero crashes
The language surrounding Vision Zero and Safe System can sometimes create confusion. The objective is not to claim that crashes can literally be eliminated from every transport network. Human beings will make mistakes. Vehicles will fail. Weather will change. Road users will behave unpredictably. Infrastructure will deteriorate.
The more realistic engineering objective is to prevent those inevitable failures from producing catastrophic consequences. That is why the Safe System concept emphasizes survivability rather than perfection. The ITF explicitly describes Safe System as a forgiving strategy that accepts human error while rejecting the idea that death or serious injury should be an inevitable consequence of that error. This is an important distinction. Zero fatalities is a safety objective. It is not a claim that zero human error is achievable.
The next frontier: from “safe road” to “safe transport system”
The phrase “safe road” may itself become increasingly inadequate.
Consider two roads with identical geometry.
One operates with:
- high vehicle speeds;
- heavy motorcycle traffic;
- frequent pedestrian crossings;
- poor lighting;
- limited enforcement;
- weak emergency response.
The other operates with:
- speed-compatible design;
- protected pedestrian facilities;
- separated cycling;
- safer vehicles;
- connected warning systems;
- reliable emergency response;
- continuous safety monitoring.
The physical pavement may be similar.
The safety system is not.
This is why the future of road safety is likely to be less about individual components and more about the interactions between:
infrastructure + vehicles + users + speed + data + governance + emergency response
That is a systems-engineering problem.
The uncomfortable conclusion
The most important transformation in road safety may not be AI. It may not be autonomous vehicles. It may not be V2X. It may not even be digital twins.
The deepest transformation is philosophical:
Road safety is moving from a discipline concerned primarily with preventing crashes toward a discipline concerned with preventing unacceptable human harm.
That distinction changes how engineers evaluate roads. A road with excellent pavement is not necessarily safe. A road with good geometric compliance is not necessarily safe. A road with few historical crashes is not necessarily safe. A road with sophisticated sensors is not necessarily safe. And a road that moves traffic efficiently is certainly not automatically safe.
The more important question is whether the transport system is capable of absorbing human error without converting it into death or serious injury. That means designing for human vulnerability. It means managing speed as a function of survivability. It means protecting people who walk, cycle and ride motorcycles. It means moving beyond crash black spots toward network-wide risk. It means using predictive methods before fatalities accumulate.
It means treating AI as an analytical instrument rather than an oracle. It means using connected technology where it creates another layer of protection. And it means evaluating safety performance rather than assuming that compliance equals safety.
The latest global evidence makes the urgency clear. Road deaths have declined substantially since the beginning of the current decade of action, but 1.16 million deaths in 2025 is still an extraordinary human toll, and global progress remains uneven. WHO’s 2026 assessment also makes clear that the world is not suffering from a complete absence of known solutions; the harder challenge is accelerating implementation, strengthening institutions, improving data and ensuring that safe-system principles become routine practice.
So perhaps the most important question for the next generation of transportation engineers is no longer:
“How do we build safer roads?”
It is:
“How do we build transport systems in which the mistakes we know humans will make are no longer allowed to become fatal?”
That is a much harder engineering problem. And it is probably the one that matters most.
Research Snapshot: Where Road Safety Is Heading
| Research direction | Traditional emphasis | Emerging emphasis | Engineering consequence |
| Crash analysis | Historical crashes | Risk + exposure + conflicts | Earlier intervention |
| Design | Compliance | Expected safety performance | Quantitative safety evaluation |
| Speed | Mobility/operations | Human injury tolerance | Context-sensitive speed management |
| Vulnerable users | Secondary consideration | Primary safety criterion | Separation and safer crossings |
| Data | Crash databases | Multimodal, real-time data | Continuous monitoring |
| AI | Experimental | Prediction and detection | Engineer + machine collaboration |
| Simulation | Operations | Surrogate safety | Test interventions before construction |
| Connected systems | Vehicle-centric | Vehicle–road–user interaction | Cooperative safety |
| Digital twins | Limited use | Dynamic scenario testing | Virtual safety evaluation |
| Governance | Project-level | Network/system-level | Institutional safety management |
Final engineering perspective
The future of road safety will not be won by the country that builds the most technologically sophisticated road. It will be won by the systems that most consistently translate known safety principles into measurable reductions in fatal and serious injury risk.
Technology can accelerate that process. Data can reveal hidden risk. AI can improve prediction. Simulation can expose dangerous interactions. Connected infrastructure can add another protective layer. But the underlying engineering principle remains remarkably simple:
People are fallible. People are physically vulnerable. Therefore, the transport system must be designed to absorb human error.
That is the real meaning of the shift from “safer roads” to “safer transport systems.”
References and further reading
World Health Organization (WHO). Global Status Report on Road Safety 2023. Foundational global assessment of road-safety progress, reporting an estimated 1.19 million road traffic deaths annually and documenting progress toward the 2030 target of reducing road deaths and injuries by at least 50%.
World Health Organization (WHO). Road Traffic Injuries. Current WHO information on the global burden of road traffic injuries, including the approximately 1.16 million annual deaths reported in current WHO data, vulnerable road users, major risk factors and the Safe System approach.
World Health Organization (WHO). Road deaths fall by 21% globally but stronger action is needed to save lives — 20 July 2026. Current WHO global update reporting a 21% decline in the global rate of road traffic deaths between 2011 and 2025, while 1.16 million people were still killed in road crashes in 2025. The release also discusses Safe System implementation, vulnerable road users and the 2030 global target.
WHO — Road deaths fall by 21% globally but stronger action is needed to save lives
International Transport Forum (ITF/OECD). The Safe System Approach. Explains the Safe System principles of accommodating human error, recognizing human physical vulnerability, sharing responsibility across the system and strengthening multiple layers of protection.
Federal Highway Administration (FHWA). Highway Safety Manual. Provides a quantitative framework for data-driven highway safety management, including safety-performance models, predictive methods and Crash Modification Factors.
Federal Highway Administration (FHWA). Safe System — An Approach Toward Zero Traffic Deaths. Provides practical guidance on the Safe System approach, including safe road users, safe vehicles, safe speeds, safe roads and post-crash care.
Federal Highway Administration (FHWA). Roadway Safety Data Program. Provides resources for roadway safety data, predictive analysis, Crash Modification Factors, systemic safety analysis and the Model Inventory of Roadway Elements.
Ali, Y., Hussain, F., & Haque, M. (2024). Advances, challenges, and future research needs in machine learning-based crash prediction models: A systematic review. Accident Analysis & Prevention, 194, 107378. Reviews machine-learning approaches for crash prediction and identifies research needs involving data quality, imbalance, real-time modelling and prediction methodology.
ScienceDirect — Machine Learning-Based Crash Prediction Models
Mustapha, A., Abdul-Rani, A. M., Saad, N., & Mustapha, M. (2024). Advancements in traffic simulation for enhanced road safety: A review. Simulation Modelling Practice and Theory, 137, 103017. Reviews contemporary traffic-simulation techniques and their use in proactive and reactive surrogate safety analysis, including microsimulation, conflict prediction and sensitivity analysis.
ScienceDirect — Advancements in Traffic Simulation for Enhanced Road Safety
Xu, G. et al. (2025). Enhancing traffic safety analysis with digital twin technology: integrating vehicle dynamics and environmental factors into microscopic traffic simulation. Scientific Reports, 15, 44404. Examines digital-twin approaches that integrate microscopic traffic simulation with vehicle dynamics and environmental factors such as weather and lighting for more comprehensive traffic-safety analysis. Nature — Enhancing Traffic Safety Analysis with Digital Twin Technology
National Highway Traffic Safety Administration (NHTSA). Research on Connected Vehicle Technology. Reports on NHTSA research into connected-vehicle technology, including vehicle-to-pedestrian applications and research involving bicyclists and other vulnerable road users. NHTSA — Research on Connected Vehicle Technology
National Academies of Sciences, Engineering, and Medicine (2025). A Guide to Applying the Safe System Approach to Transportation Planning, Design, and Operations. NCHRP Research Report 1135. Provides practical guidance for applying Safe System principles across transportation policy, planning, design, operations and maintenance, law enforcement and post-crash response. National Academies — A Guide to Applying the Safe System Approach to Transportation Planning, Design, and Operations