New field experiments with a transitional autonomous vehicle (tAV) performing 78 mandatory lane changes on a public road show that lead and lag gaps consistently converge toward a narrow range just before crossing the lane boundary. The dataset also reveals that collision risk rises steadily through the maneuver, peaking at the moment of physical lane entry.
What the Researchers Built
The North Carolina Transitional Autonomous Vehicle Lane-Changing (NC-tALC) dataset captures 78 controlled mandatory lane-change trials on a public four-lane road in Apex, North Carolina. Four instrumented vehicles—a lane-changing tAV (called X), two target-lane followers (A and B), and one target-lane leader (C)—created repeatable initial traffic patterns. All vehicles operated in a transitional automated driving mode (tAV), meaning the system handled longitudinal and lateral control but a human still monitored the environment.
Researchers placed X at three initial positions relative to the gap between vehicles A and B: near A, near the center, and near B. Each condition was repeated multiple times to gather statistically meaningful behavioral data. High-resolution RTK-GNSS/INS units on every vehicle recorded position, speed, heading, and acceleration at 20 Hz with centimeter-level accuracy.
The dataset provides time-synchronized trajectories for the entire lane-change process—from the moment X decides to merge through gap acceptance, lane crossing, and post-merging stabilization. This goes far beyond the typical snapshots of gap acceptance found in prior studies.
Key Results
The most striking finding is that lead and lag gaps—the time headways between the lane changer and the target-lane leader/follower—converge toward a narrow band of 0.5–1.5 seconds around the lane-crossing moment, regardless of the initial spacing condition. This suggests a predictable behavioral pattern inherent to tAV lane changing.
Collision risk, measured by modified time-to-collision and deceleration rate to avoid crash (DRAC), increased steadily as the maneuver progressed. Risk peaked at the instant of physical lane entry (when the vehicle’s front tire crossed the lane boundary) and remained elevated even after lane change completion. The dominant threat was always the target-lane leader, not the follower.
Among the 78 trials, the lane changer selected the gap between A and B (Gap 1) in 52 cases, the gap ahead of A (Gap 0) in 16 cases, and the gap between B and C (Gap 2) in 10 cases. No trial resulted in a merge behind C (Gap 3). This distribution highlights a strong preference for the immediate target gap.
How It Works
Each vehicle carried a single INS unit integrating RTK-GNSS with accelerometers and gyroscopes, mounted at the vehicle center. Raw positions were converted from geographic coordinates to a local Cartesian frame under the WGS‑84 model, then projected onto a reference lane centerline. A 20‑point (1 second) moving average filtered out GNSS noise.
Researchers identified key lane-change timestamps—start of lateral movement, lane-crossing instant, and full lane-change completion—from the lateral trajectory profile. They then computed three time-gap variables:
- Lead gap: time headway between the lane changer and the target-lane leader (negative when the lane changer is ahead of the leader)
- Lag gap: time headway between the lane changer and the target-lane follower (negative when behind)
- LC gap: time headway between the target-lane leader and follower (the available gap)
The following table summarizes the target-gap selection results:
| Candidate Gap | Description | Number of Trials |
|---|---|---|
| Gap 0 | Ahead of A | 16 |
| Gap 1 | Between A and B | 52 |
| Gap 2 | Between B and C | 10 |
| Gap 3 | Behind C | 0 |
Surrogate safety measures (time-to-collision and DRAC) were calculated from the trajectory data at each timestamp. The analysis shows that risk does not disappear at lane-change completion—the vehicle still needs to stabilize its longitudinal position relative to the new leader, and safety margins remain compressed for several seconds afterward.
Why This Matters for Robotics
Transitional automated vehicles represent the current generation of production-ready autonomous driving systems. Understanding how these systems actually behave in real mandatory lane-change scenarios is critical for simulation validation, safety certification, and public acceptance. The NC-tALC dataset provides empirical benchmarks that can directly inform the calibration of motion planning algorithms and behavior models.
For fleet operators managing warehouse robots or automated guided vehicles, the finding that collision risk peaks at lane entry—not just at gap acceptance—is a practical insight. Safety systems should monitor the entire maneuver, not merely the decision to merge.
The convergence of lead and lag gaps suggests that tAV lane-changing behavior is more predictable than human driving, which could simplify traffic flow modeling and V2X coordination strategies. This dataset and its insights can accelerate the development of safer, more reliable autonomous driving stacks.
Limitations and Open Questions
The experiments were conducted on a straight, dry public road with no other traffic beyond the four instrumented vehicles. Results may differ on curves, in inclement weather, or with surrounding human-driven vehicles. All vehicles operated in tAV mode; findings may not generalize to fully autonomous systems (Level 4/5) or to vehicles using different sensor suites.
The sample size of 78 trials, while substantial for a controlled field study, limits statistical power for rare events or edge cases. Additionally, the lane changer’s initial position within the gap was the only controlled variable—factors like vehicle speed differential or driver attention were not independently varied. Future work should explore how these results scale to more complex multi-vehicle scenarios.
Frequently Asked Questions
What is a transitional autonomous vehicle (tAV)? A tAV is a vehicle that can handle both longitudinal and lateral control but still requires a human driver to monitor the environment and intervene when needed—similar to SAE Level 3 or 4 systems currently on the road.
How was the gap convergence measured? Researchers computed time headways between the lane changer and the target-lane leader and follower at each timestamp. They then plotted the evolution of these gaps and observed that, regardless of initial spacing, they consistently narrowed to 0.5–1.5 seconds near the lane-crossing moment.
Why did collision risk peak at lane entry rather than at gap acceptance? The lane changer’s lateral motion reduces the effective safety margins while crossing the lane boundary. Even though the gap was accepted earlier, the physical encroachment into the new lane temporarily compresses time-to-collision values before the vehicle stabilizes longitudinally.
Can I access the NC-tALC dataset for my own research? The paper states the dataset is available for research purposes—contact the authors for access. It includes high-resolution trajectories and computed safety metrics for all 78 trials.
Conclusion
The NC-tALC dataset and its analysis provide the first controlled, empirical characterization of the full mandatory lane-change process in transitional autonomous vehicles. The observed convergence of lead and lag gaps and the peak crash risk at lane entry offer concrete benchmarks for developers and safety assessors. This work underscores that lane-change safety must be evaluated over the entire maneuver, not just the gap acceptance decision.