Controlled Experiments on Lane Changing by Transitional Autonomous Vehicle: Dataset and Behavioral Insights

Abhinav Sharma; Danjue Chen; George F. List; Md Abdullah Al Hasan

Controlled Experiments on Lane Changing by Transitional Autonomous Vehicle: Dataset and Behavioral Insights

T0 review · 3 major / 6 minor · reviewed 2026-07-30 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read In controlled public-road trials, transitional automated vehicles converge lead and lag gaps by lane crossing while front collision risk peaks at physical lane entry and can outlast the maneuver.

desk verdict Useful controlled mandatory-merge trajectories and process-level patterns; the “preferred region” and SAE risk story should stay tied to one stack and fixed kinematics. read the letter →

arxiv 2607.27085 v2 pith:GCK4VTRO submitted 2026-07-29 cs.RO cs.SYeess.SY

Abhinav Sharma , Md Abdullah Al Hasan , Danjue Chen , George F. List This is my paper · ORCID

classification cs.ROcs.SYeess.SY

keywords transitionalautonomousvehiclesmandatorylanechanginglead-laggapssurrogatesafetymeasurescontrolledfieldexperimenttrajectorydatasetcollisionriskevolution

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper introduces a controlled public-road dataset of 78 mandatory lane changes by transitional automated vehicles—systems that decide and execute lane changes under driver supervision—and uses the trajectories to map behavior across the full maneuver, not only at gap acceptance. Four instrumented vehicles created repeatable initial spacings while the lane changer picked among candidate target gaps. Despite very different starting lead–lag positions, lead and lag time gaps consistently moved toward a relatively narrow band by the moment the vehicle center crossed into the target lane, with the lag gap often larger than the lead gap thereafter. Projected longitudinal collision risk, measured by an emergency-braking spacing surrogate, rose through the maneuver, peaked when the vehicle’s left edge first entered the target lane, was dominated by conflict with the target-lane leader rather than the follower, and sometimes remained after the lateral maneuver had ended. The practical stake is clear: models, simulators, and safety checks that treat automated lane change as a single acceptance instant will miss both the convergence behavior and the risk that peaks at entry and can persist afterward.

What carries the argument

The NC-tALC controlled four-vehicle mandatory-merge experiment with high-rate RTK-GNSS/INS trajectories, annotated key timestamps (activation, lane-change start, left-edge touching, crossing, end, and post-end offsets), lead/lag/LC time gaps, and the spacing-after-emergency (SAE) surrogate that scores front and rear longitudinal risk under a fixed emergency-braking scenario.

What would settle it

Repeat the same mandatory-merge protocol with other manufacturers or software versions, or with weaker assumed braking and longer reaction times: if lead–lag states no longer converge near those bounds by crossing, or if rear risk dominates or risk vanishes by maneuver end, the claimed convergence and front-risk pattern fail.

Watch

Extended reading notes

Core claim

Despite substantially different initial lead–lag conditions, the tested transitional automated vehicles’ lead and lag gaps evolve toward a relatively narrow operating region by lane-change crossing (empirical minima near about 0.54 s lead and 0.65 s lag, with lag often larger than lead thereafter). Significant projected longitudinal collision risk develops during the maneuver, typically peaks at left-edge touching (physical lane entry), is predominantly front risk versus the target-lane leader, and can persist beyond lane-change end.

Load-bearing premise

The claim that these patterns show a preferred operating region and real collision risk rests on one supervised vehicle type, one site, challenging small gaps, a subjective initial-condition grouping, and a fixed emergency-braking formula whose parameters may understate or mislocate risk.

Editorial extensions

If this is right

Lane-change models for transitional automation should track longitudinal adjustment through the full process, not only gap acceptance or a single timestamp.
Safety evaluation should score front and rear risk separately from physical lane entry through post-maneuver stabilization, because risk can peak at entry and outlast completion.
The dataset supplies empirical benchmarks for calibrating behavioral models and validating simulation of mandatory merges under controlled initial conditions.
Lane-crossing is a behavioral milestone: initially different lead–lag conditions have largely converged and rear risk is largely reduced by that point.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If front-risk tolerance is a controller preference rather than a site artifact, mixed traffic may see more leader braking or cut-in friction than human-centric merge models predict.
Regulators and OEM test suites that only check gap size at initiation or completion would miss the highest-risk window identified here.
Comparing the same protocol under ACC-only followers versus automated followers (reserved in the paper’s companion experiment) would isolate how follower automation changes the risk timeline.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The thing worth knowing is simple: they ran a real public-road, four-vehicle mandatory merge experiment (n=78) with RTK/INS at 20 Hz, forced different initial positions in small target gaps, and tracked the full maneuver—not just gap acceptance. That fills a hole relative to Waymo L4 snippets, TGSIM assisted events, and the JRC assisted-LC campaign (few system-initiated cases, data not public).

What they do well is the craft. Activation, LCS/LET/LCC/LCE definitions are explicit; lead/lag/LC gaps are cleanly defined; the descriptive pattern is credible on the sample: different starts tighten toward a common lead–lag band by centerline crossing, lag often larger than lead after that, and projected longitudinal risk (their SAE) often worst at first physical entry, mostly vs the target leader, sometimes still negative after LCE. They also say out loud that results are one supervised system, one site, challenging gaps, and that softer braking or slower reaction would look worse. That honesty helps.

Soft spots are real but proportionate. The ~0.54 s / ~0.65 s bounds and “preferred operating region” are same-sample minima plus a subjective 0.5 s grouping, then read as controller intent—mild circularity, not fraud. SAE with τ=0.5 s and 0.8 g is a fixed surrogate, not observed crash risk; front-risk dominance is interesting under those assumptions, not a general tAV law. “tAV” is a useful label here, not a settled SAE class. Code/raw release is not evidenced in-text, which matters for a dataset paper. None of that kills the trajectories or the process-level plots.

Who it’s for: people building or validating mandatory LC models, AV safety metrics, and mixed-traffic sims who need full-maneuver benchmarks rather than acceptance-instant snapshots. Math is elementary kinematics; citations look fair to the prior AV LC work. I’d send it to referees. Engage if you work this area; treat the numbers as this stack under these conditions, not universal tAV behavior.

Referee Report

3 major / 6 minor

Summary. The manuscript introduces the NC-tALC dataset from 78 controlled mandatory lane-change trials of a transitional automated vehicle (tAV) on a public road in Apex, NC, using four instrumented vehicles and 20 Hz RTK-GNSS/INS trajectories. It defines key timestamps (ACT, LCS, LET, LCC, LCE and post-LCE offsets), computes lead, lag, and LC time gaps, and applies a spacing-after-emergency (SAE) surrogate to track longitudinal risk. The main empirical claims are that, despite varied initial positions within the target gap, lead and lag gaps evolve toward a relatively narrow band by lane-change crossing (reported minima near 0.54 s lead and 0.65 s lag, with lag often larger thereafter), and that projected collision risk often rises through the maneuver, peaks near left-edge touching, is predominantly front risk versus the target-lane leader, and can persist past lane-change end. The authors position the work as one of the first controlled characterizations of the full mandatory LC process for tAVs that decide and execute the maneuver under supervision.

Significance. If the reported process-level patterns hold under the stated scope, the paper fills a genuine empirical gap: systematic public-road data on tAVs that independently decide and execute mandatory lane changes, rather than Level-4 open datasets, pooled Level 1–2 assisted events, or unreproduced assisted-LC campaigns. The controlled four-vehicle design, high-resolution trajectories, explicit timestamp definitions, and public dataset framing are concrete strengths for model calibration, simulation validation, and safety-method benchmarking. The emphasis on evolution through the full maneuver (not only gap acceptance) is useful for traffic and safety modeling. Credit is due for transparent experimental control of initial spacing classes, clear instrumentation, and explicit scope limitations in the conclusions. The contribution is primarily observational and dataset-centered rather than a new theory; its lasting value depends on careful interpretation of the preferred-region and SAE-risk claims and on dataset usability.

major comments (3)

[GAP ACCEPTANCE AND EVOLUTION] GAP ACCEPTANCE AND EVOLUTION: The conjecture of a controller “preferred lead–lag operating region” at LCC rests on minima (~0.54 s lead, ~0.65 s lag) read from the same near-leader/near-follower trials that define the groups, then applied back as shared bounds that the near-center group also meets. That is a same-sample descriptive pattern, not independent evidence of a preferred setpoint. Please either (i) reframe strictly as an empirical convergence description without “preferred region” language, or (ii) support the claim with out-of-sample checks (e.g., hold-out trials, Gap-0 vs Gap-1/2, or pre-registered bounds) and report group sizes n for near-leader / near-center / near-follower at ACT and at LCC.
[RISK ASSESSMENT] RISK ASSESSMENT, Eq. (4)–(7): Front-risk dominance and at-risk fractions (e.g., 83.9% SAE_LC<0 at LET; ~1/3 still at risk at LCE) are load-bearing for the safety narrative, yet SAE uses a single fixed emergency parameter set (τ_e=0.5 s, d=0.8 g, ℓ_v=5 m, b_0=2 m) with no sensitivity table. The text correctly notes that longer τ or weaker braking would worsen SAE, but the quantitative claims (peak at LET, persistence past LCE, front dominance percentages) could shift under plausible alternatives. Please add a compact sensitivity analysis (at least τ_e and d) showing whether peak timing, front/rear dominance, and post-LCE persistence are robust, and keep the interpretation as a kinematic surrogate under stated assumptions rather than operational collision probability.
[CONCLUSIONS AND DISCUSSIONS] CONCLUSIONS AND EXPERIMENTS: The abstract and novelty statements characterize “tAV” mandatory LC behavior, while the body correctly limits findings to one supervised stack, one site, deliberately small candidate gaps, and same-manufacturer tAV followers. That scope mismatch is load-bearing for transferability. Please align title/abstract/practical-applications wording with the single-system, controlled-challenge design (e.g., “a commercial tAV under supervised mandatory merge”), and state more clearly that Gap-3 was never selected and that small-gap forcing may inflate risk relative to naturalistic merges.

minor comments (6)

[Title page] Submission date on the title page reads “July 29, 2026” and the arXiv stamp is 29 Jul 2026; confirm this is intentional and consistent with the journal’s dating practice.
[Time gap measurement] Eq. (2)–(3): clarify sign convention and when negative lead/lag gaps are retained versus undefined; a short note that vehicle-center headway with fixed ℓ_v can bias short gaps would help reproducibility.
[Figures 5–12] Figures 5–12: axis units, sample sizes per panel/group, and the 0.5 s grouping threshold should appear in captions; several figures are hard to interpret from the text alone.
[TABLE 1 / GAP ACCEPTANCE] Table 1 is clear on gap selection, but the main text never tabulates how the 78 trials map into the three ACT lead–lag groups used in §GAP ACCEPTANCE; add a small contingency table.
Minor prose issues: spacing typos (“throughoutthe”, “lane-changeprocess”), inconsistent “tAV dataset” vs “NC-tALC”, and mixed “lead–lag” hyphenation. A copy-edit pass would help.
[INTRODUCTION] Related work: the JRC/Mattas campaign and TGSIM are well cited; if the NC-tALC companion arXiv (Sharma et al. 2026) is the data release, state access conditions and what is public versus reserved for the ACC-follower paper.

Circularity Check

1 steps flagged · score 2.0 of 10

Observational experiment; only mild same-sample bound extraction framed as a preferred operating region.

self definitional [GAP ACCEPTANCE AND EVOLUTION; lead–lag convergence / preferred-region conjecture (Figs. 5–6)]
"At LCC, the minimum lead gap is 0.54s. ... At LCC, the minimum lag gap is 0.65s. ... we conjecture that the tAV adjusts toward a preferred lead–lag operating region by LCC. This region is characterized by a lead gap exceeding the minimum lead-gap bound of approximately 0.54 and a lag gap exceeding the minimum lag-gap bound of approximately 0.65s. ... although the lower bounds are extracted from the near-leader and near-follower groups, they also describe the near-center group: one can see that by LCC, all cases in this group meet both bounds."

The ‘preferred’ lower bounds are defined as the observed sample minima at LCC in the near-leader (lead) and near-follower (lag) groups. Stating that those groups reach gaps above those bounds is true by construction of the minima. The near-center check adds some independent descriptive support, but the preferred-region claim still partly renames the same-sample floors as a controller target rather than testing a bound fixed before seeing LCC.

full rationale

This paper is a controlled field study that reports measured lead/lag trajectories and a fixed-parameter SAE surrogate on 78 trials. It does not claim a first-principles derivation that forces the outcome, nor does it fit a model on a subset and then ‘predict’ a tightly coupled quantity. The main findings—gap evolution toward a narrow band by LCC, SAE risk peaking near LET and remaining front-dominated—are descriptive summaries of the processed trajectories under stated definitions (key timestamps, time headways, Eq. 4 SAE). The only mild circularity is interpretive: minima of lead and lag at LCC are read off the near-leader and near-follower groups, then those same numerical floors are labeled a ‘preferred lead–lag operating region’ that the groups (by construction for the extracting groups) and the near-center group ‘meet.’ That is post-hoc characterization from the plotted sample, not a prediction forced by definition of the measurement chain. Self-citations (e.g., the companion dataset preprint) are bibliographic, not load-bearing uniqueness theorems. Score 2 reflects that single minor same-sample framing step; the empirical content remains independent of any circular reduction.

Assumptions & free parameters 7 free parameters · 6 assumptions · 3 invented entities

Load-bearing content is empirical. Claims rest on domain modeling choices for gaps and emergency braking, hand thresholds for event detection and grouping, and the construct of a tAV ‘preferred’ lead–lag region inferred from the same trials. No deep new physics entities; free parameters are analysis thresholds and SAE constants that shape the risk and convergence narratives.

free parameters (7)

Near-leader/near-follower grouping threshold = 0.5 s
Cases split by |lead gap|<0.5 s or |lag gap|<0.5 s at ACT; paper calls 0.5 s subjective. Group-wise convergence story depends on this cut.
SAE emergency response time τ_e = 0.5 s
Fixed in Eq. 4 risk metric; directly controls whether SAE is negative and thus at-risk fractions.
SAE emergency deceleration d (=b_l=b_f) = 0.8 g
Hand-set to 0.8 g; paper notes weaker braking would increase apparent risk. Central risk-magnitude claims depend on it.
SAE buffer b_0 and vehicle length ℓ_v = b_0=2 m, ℓ_v=5 m
Additive spacing constants in SAE; shift the zero-risk boundary.
Lateral velocity threshold for LCS/LCE = 0.05 m/s (LCE hold 0.3 s)
0.05 m/s (and 0.3 s hold for LCE) defines maneuver start/end times that anchor all evolution plots.
Trajectory/speed smoothing window = 1 s / 20 samples at 20 Hz
20-point (1 s) moving average applied to positions and speeds before analysis; affects gap and lateral-velocity timing.
Empirical LCC lead/lag lower bounds = ~0.54 s lead, ~0.65 s lag
Minima ~0.54 s lead and ~0.65 s lag read off the sample and then used to describe a preferred operating region.

assumptions (6)

domain assumption Time headway with vehicle-center longitudinal positions and fixed vehicle length is an adequate operational definition of lead, lag, and LC gaps.
Eqs. 2–3; negative ‘lead’ when X is ahead of A is retained by convention.
domain assumption Longitudinal emergency-braking SAE with equal max decelerations, constant response delay, and no lateral conflict model is a valid surrogate for potential collision risk from LET onward.
Risk Assessment, Eq. 4–7; SAE before LET is interpretive only.
domain assumption The tested supervised automated lane-change stack is representative enough to label findings as tAV mandatory-LC behavior under the stated ODD.
Introduction/Experiments define tAV; Conclusions limit generalization but abstract still frames tAV behavior.
domain assumption Lane centerlines from repeated tAV drives and midpoint lane boundaries correctly define LET/LCC geometry on the public site.
Data Processing section; centimeter RTK assumed sufficient after 1 s smoothing.
standard math Standard kinematic projection and geodesic WGS-84 local framing preserve interaction ordering for gap and SAE calculations.
Coordinate conversion and SAE closed-form stopping distances.
ad hoc to paper Convergence of disparate initial conditions to similar LCC lead–lag statistics indicates adjustment toward a preferred operating region of the controller.
Explicit conjecture in Gap Acceptance section; mechanisms proprietary and bounds not independently verified.

invented entities (3)

Transitional autonomous vehicle (tAV) category independent evidence
purpose: Name vehicles beyond common L1–2 assist that can decide and execute full lane changes under driver supervision, distinct from HDV and unsupervised L4.
Definitional framing in Introduction; maps onto commercial supervised auto lane-change features but is a paper-specific umbrella label.
Preferred lead–lag operating region at LCC
purpose: Explain cross-group gap convergence as goal-directed adjustment rather than initial-condition echo.
Inferred from same-trial minima and distribution tightening; no controller disclosure or external validation.
NC-tALC dataset / four-vehicle mandatory-merge experimental construct independent evidence
purpose: Provide repeatable public-road trajectories with controlled ds, near-zero dv, and multiple candidate gaps.
Core contribution; existence is the experiment itself, with companion citation Sharma et al. 2026.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Controlled Experiments on Lane Changing by Transitional Autonomous Vehicle: Dataset and Behavioral Insights." pith.science (2026). https://pith.science/paper/GCK4VTRO

@misc{pith2026260727085,
  author       = {Pith},
  title        = {Pith review of: Controlled Experiments on Lane Changing by Transitional Autonomous Vehicle: Dataset and Behavioral Insights},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GCK4VTRO}},
  note         = {Machine review of arXiv:2607.27085}
}

read the original abstract

This paper presents the North Carolina Transitional Autonomous Vehicle Lane-Changing (NC-tALC) dataset and uses it to characterize mandatory lane-changing behavior of transitional automated vehicles (tAVs). It quantifies the evolution of lead--lag gaps throughout the lane-change process and examines how potential collision risk develops during the maneuver. A controlled field experiment comprising 78 mandatory lane-change trials was conducted on a public roadway in Apex, North Carolina. Four instrumented vehicles created repeatable traffic conditions while varying the lane changer's initial position within the candidate target gap. High-resolution RTK-GNSS/INS trajectories were processed to identify key timestamps, calculate lead, lag, and lane-change gaps, and estimate interactions using time-gap- and speed-based surrogate safety measures. Despite substantial differences in initial conditions, lead and lag gaps consistently converged toward a relatively narrow range near lane crossing. Potential collision risk increased as the maneuver progressed, peaked near physical lane entry, and was dominated by interactions with the target-lane leader. Lane-change completion did not necessarily coincide with the disappearance of collision risk. This study provides one of the first controlled empirical characterizations of the complete mandatory lane-change process of tAVs using repeatable public-road experiments. The NC-tALC dataset supports analysis of behavioral and safety evolution throughout the maneuver rather than only at the gap-acceptance instant. The dataset and findings provide empirical benchmarks for evaluating automated lane-changing behavior, calibrating behavioral models, and validating simulation and safety assessment methods for mandatory lane-change scenarios.

Figures

Figures reproduced from arXiv: 2607.27085 by the authors.

**Figure 5.** However, at LCC, most cases satisfy 𝑙𝑎𝑔 𝑔𝑎 𝑝 ≥ 𝑙𝑒𝑎𝑑 𝑔𝑎 𝑝; as indicated by the points on or above the diagonal line. Between LCC and LCE, the lag gap continues to increase relative to the lead gap, resulting in further separation above the diagonal line in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

**Figure 6.** At LCC, the lag gap exceeds the lead gap in most cases; as shown by the points above the [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Pith tools

Reviewed July 30, 2026 · model on record in the stance chip above.