Inside the MnDOT Study: How Lane-Keeping Systems Really Perform in Work Zones

What happens when your car’s lane-keeping system encounters a work zone? A new study from the Minnesota Department of Transportation (MnDOT), conducted by SRF Consulting Group with ViaSight as a technology partner, set out to answer that question. The findings have real implications for work zone safety across the country.

SRF recently discussed the project on the inaugural episode of their Unbox the SRF podcast. Here’s what they found, and how ViaSight’s technology played a key role.

Why This Study Matters

ADAS lane-keeping systems are now standard in most new vehicles. They use forward-facing cameras to detect lane markings and keep vehicles centered in their lane. The problem: work zones regularly disrupt the visual cues these systems depend on. Temporary markings, removed lines, cones, drums, and pavement transitions all create scenarios that ADAS wasn’t designed to handle.

MnDOT wanted to know: what happens to these systems when drivers enter a work zone? And more importantly, what can designers and field crews do about it?

The Black Box Problem

One of the biggest challenges in studying ADAS performance is that vehicle systems are a black box. A vehicle’s dashboard will tell you whether lane-keeping is engaged or disengaged, but that’s it. You get a binary signal with no insight into what the camera is actually seeing or why the system made a particular decision.

This is where ViaSight came in. As Uthej Vattipalli of SRF explained on the podcast:

“This is where our partner ViaSight comes into picture. They have their analytics that is a good proxy for how these systems are working… what are they seeing in that particular frame… We needed a good way to understand and look into that black box.”

ViaSight’s analytics provided an objective, video-based assessment of pavement marking quality and visibility from the vehicle’s perspective. While the vehicles could only report on/off status, ViaSight’s system assigned performance scores to each roadway setting, giving researchers the quantitative data they needed to understand why systems were engaging or disengaging.

Testing at MnROAD: Controlled Conditions

The study began at MnROAD, MnDOT’s pavement research facility northwest of the Twin Cities. It’s a closed-loop test track with no live traffic. Researchers set up 15 different simulated work zone scenarios and drove multiple test vehicles through them, including a Toyota Camry, Hyundai Elantra, Tesla Model Y, and a commercial semi-truck.

Three cameras captured data simultaneously: one facing the road, one on the vehicle dashboard to record system status, and one running ViaSight’s analytics to evaluate what the vehicle’s camera system was likely perceiving.

Key Findings from MnROAD

Pavement marking removal methods matter. Of the three methods tested (blackout tape, grinding, and water blasting), water blasting was the most effective at eliminating old markings from ADAS detection. Both blackout tape and grinding left residual cues that vehicles inconsistently picked up.
Cones and drums are invisible to ADAS. None of the lane-keeping systems recognized cones or drums as lane boundaries. The Tesla’s system could detect them as objects but did not use them for lateral guidance, even in autopilot mode.
Camera mounting height affects perception. Using ViaSight’s analytics at different heights, the team found that a commercial vehicle’s elevated camera position was more consistent at detecting removed markings. Still, the semi’s Level 1 system underperformed the passenger vehicles’ Level 2 systems overall.

Real-World Testing: More Variables, Bigger Surprises

The team then took the study to active work zones across Minnesota, on Highway 610 and near Hinckley, Austin, and Princeton, using 2025 model-year vehicles including a Ford Bronco, Toyota Camry, Hyundai Palisade, and Tesla Model Y.

The real world introduced complexity that MnROAD couldn’t replicate. Multiple variables stacked up in a single frame: drums, cones, missing markings, pavement color transitions, and conflicting old and new lines all at once. Isolating any single cause-and-effect relationship became significantly harder.

Standout Real-World Findings

A 2025 Ford Bronco could not detect yellow pavement markings at all, causing the vehicle to deviate from its lane whenever yellow lines were present. As one researcher noted with some irony, “the most American car in our test sample cannot detect yellow pavement marking, which is apparently only used in America.”
Missing extension lines at exits and entries caused vehicles to default to the rightmost path, potentially steering drivers onto exit ramps unintentionally.
Transition zones are the danger zones. Vehicles performed well in straight, tangent sections of work zones. But at lane closures and tapers, where workers are most likely to be positioned, the systems were unreliable and sometimes drove straight toward drums without disengaging.
Tesla’s Full Self-Driving mode was the exception. While Tesla’s basic autopilot struggled like the other vehicles, its Full Self-Driving mode navigated all work zone scenarios without issue.

What This Means for Work Zone Safety

The study’s conclusions are clear: it all comes down to pavement markings. Current ADAS lane-keeping systems are fundamentally dependent on visible lane lines, and work zones routinely compromise that visibility.

The practical takeaways for transportation agencies and work zone designers include:

Water blast old markings rather than grinding or taping over them to minimize confusing ADAS systems
Maintain extension dotted lines at transitions, exits, and entries. They’re affordable and critical for ADAS guidance.
Recognize that transition areas carry the highest risk for ADAS-related incidents, and plan worker positioning accordingly
Educate the public about the limitations of their vehicles’ lane-keeping systems in work zones

As SRF’s Jon Jackels emphasized: “We have to give information to the workers that these systems don’t react like we think they’re going to react.”

ViaSight’s Role: Seeing What Vehicles See

This study validates the core problem ViaSight was built to solve. Transportation agencies can’t rely on ADAS manufacturers to disclose how their systems perceive road infrastructure. What agencies can do is proactively assess whether their work zones are machine-readable, before vehicles and drivers encounter them.

ViaSight’s analytics gave this research team the ability to move beyond binary engaged/disengaged data and understand the quality of what ADAS cameras were seeing. That same capability is what powers our Zone Sure platform, enabling agencies and contractors to evaluate work zone compliance through the lens of machine vision, not just human eyes.

Watch the Full Discussion

The full conversation is available on SRF Consulting’s Unbox the SRF podcast.