arxiv: 2605.01485 · v1 · submitted 2026-05-02 · 💻 cs.RO · physics.soc-ph

Recognition: unknown

Cut-In Gap Acceptance Toward Autonomous vs. Human-Driven Vehicles: Evidence from the Waymo Open Motion Dataset

Abdulaziz Alhuraish , Yuhang Wang , Hao Zhou

Authors on Pith no claims yet

Pith reviewed 2026-05-09 14:16 UTC · model grok-4.3

classification 💻 cs.RO physics.soc-ph

keywords autonomous vehiclesgap acceptancecut-in maneuvershuman driving behaviorlane change detectionWaymo datasettraffic safetymotion data

0 comments

The pith

Human drivers accept shorter gaps when cutting in front of autonomous vehicles than in front of human-driven vehicles.

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

The paper investigates whether human drivers treat autonomous vehicles differently by accepting smaller gaps for cut-in maneuvers. It uses thousands of real-world highway scenarios from the Waymo dataset to compare cut-ins in front of the AV against those in front of other cars. The analysis shows a statistically significant 1.99 meter reduction in median gap for AV targets, along with faster cut-in speeds and more frequent small-gap events. This difference suggests that the conservative nature of AV motion policies allows humans to behave more aggressively around them. Such findings matter for designing safer AV systems that anticipate human responses.

Core claim

Using an eight-criterion detector on the Waymo Open Motion Dataset, the authors extract 706 cut-in events targeting the AV and 3,172 targeting HDVs. They report that the median accepted gap ahead of the Waymo AV is 7.58 meters, compared to 9.57 meters for HDV targets. This 1.99 meter reduction is statistically significant and remains after speed-matched resampling. Additionally, cut-ins toward the AV occur at 37 percent higher speeds, and a larger share happen at gaps under 10 meters.

What carries the argument

The eight-criterion lane-change detector applied to 10 Hz motion data to isolate comparable cut-in events in identical traffic environments for AV and HDV targets.

If this is right

AV motion planning must incorporate expectations of more aggressive human cut-ins.
Traffic simulation models require separate gap-acceptance parameters for interactions with autonomous vehicles.
The asymmetry in behavior points to a need for AV policies that account for predictable human adaptations.
Safety assessments should consider higher rates of close-proximity maneuvers around AVs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

As more people encounter AVs, the gap acceptance difference could diminish or increase depending on learned expectations.
The finding may generalize to other maneuvers such as yielding or merging, suggesting broader studies on human-AV interaction asymmetries.
Data from additional AV operators could test whether the effect is tied to specific driving styles like conservative policies.

Load-bearing premise

The eight-criterion lane-change detector extracts comparable and unbiased cut-in events for both AV and HDV targets without systematic differences in detection accuracy or labeling between the two classes.

What would settle it

A replication study that manually labels a subset of the cut-in events from the same dataset and finds no difference in accepted gaps between AV and HDV targets would undermine the central claim.

Figures

Figures reproduced from arXiv: 2605.01485 by Abdulaziz Alhuraish, Hao Zhou, Yuhang Wang.

**Figure 1.** Figure 1: Gap at entry (dLC ) for HDV→AV (red) and HDV→HDV (blue) cut-in events. (a) Kernel density estimate; dashed vertical lines mark group medians; dotted lines indicate the 5 m and 10 m risk thresholds. (b) Empirical CDF; the AV-targeted curve is consistently left of the HDV-targeted curve, corresponding to the 1.99 m median reduction (p=5.76×10−8 , d=−0.224). TABLE I DESCRIPTIVE STATISTICS: HDV→AV VS. HDV→HDV … view at source ↗

**Figure 2.** Figure 2: Speed analysis for HDV → AV (red) and HDV → HDV (blue) events. (a) Cut-in speed vs. lead-vehicle speed scatter with group regression lines. (b) Violin plot of relative speed differential ∆v at entry; AV-targeted events exhibit a 80% larger differential (d=0.428). (c) KDE of the speed ratio vci/vlead; the AV-targeted distribution peaks above 1, indicating cut-in vehicles are faster than their lead target (d… view at source ↗

**Figure 3.** Figure 3: overlays 2-D KDE contours in the gap–TTC plane. Panel (a) shows HDV→AV mass concentrating in the lowerleft quadrant (small gap, moderate TTC). Panel (b) shows HDV→HDV events dispersed toward larger gaps and higher TTC. Panel (c) combines both groups with group medians (diamonds); the HDV → AV median falls in the moderaterisk zone, while the HDV→HDV median lies in the low-risk region. The larger post-cut-… view at source ↗

**Figure 4.** Figure 4: Severity classification by gap-at-entry threshold. view at source ↗

read the original abstract

Autonomous vehicles (AVs) are widely known to follow conservative, rule-based motion policies that surrounding drivers can learn to anticipate. A direct consequence is that human drivers may accept shorter longitudinal gaps when cutting in front of an AV than when targeting another human-driven vehicle (HDV). We test this hypothesis using the Waymo Open Motion Dataset (WOMD), which provides 25,906 real-world highway scenarios at 10 hertz. An eight-criterion lane-change detector extracts 706 HDV-to-AV and 3,172 HDV-to-HDV cut-in events from the same traffic environment. The median accepted gap in front of the Waymo AV is 7.58 meters versus 9.57 meters for HDV targets, a 1.99 meter reduction that is statistically significant (p equals 5.76 times 10 to the negative eighth power, d equals negative 0.224) and persists under speed-matched resampling. Cut-in speeds toward the AV are 37 percent higher (51.7 versus 37.7 kilometers per hour, d equals 0.502), and 68.0 percent of AV-targeted cut-ins occur below the 10 meter gap boundary versus 51.8 percent of HDV-targeted events (chi-squared equals 60.5, p is less than 10 to the negative thirteenth power). These results reveal a systematic and safety-relevant asymmetry in human gap-acceptance behavior that warrants AV-specific calibration of both motion-planning safety envelopes and traffic simulation models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper measures a 2-meter smaller median gap for cut-ins in front of Waymo AVs versus HDVs in real highway data, but the lane-change detector may systematically favor short-gap AV events due to smoother trajectories.

read the letter

The core finding is that drivers accept shorter gaps and cut in faster when targeting the Waymo AV than when targeting other human-driven vehicles, with the 1.99 m median difference and the 68% vs 52% below-10 m split both statistically clear. The speed-matched resampling keeps the gap result intact. That supplies a usable real-world number for safety envelope work and simulator calibration, drawn from the same traffic scenes rather than separate experiments. The dataset is public and the counts are decent (706 AV events, 3172 HDV), so the comparison is at least replicable in principle. The statistical reporting is direct: medians, p-values, effect sizes, and a chi-squared check on the proportion below 10 m. Those pieces are straightforward and worth having on record. The main soft spot is the eight-criterion detector. Fixed kinematic thresholds on relative position, velocity, and heading can behave differently when the target vehicle has low acceleration variance and tight lane-keeping, which the AV does. That raises the chance that short-gap AV cut-ins are over-counted relative to HDV ones without any shift in actual driver behavior. The speed match does not address trajectory smoothness or detection sensitivity, so the bias concern stays live. The abstract gives only high-level description of the criteria and no validation numbers on recall or false-positive rates by target type. If the full methods section shows the exact thresholds and any cross-checks that the event sets are comparable, the result strengthens; otherwise the asymmetry could be partly mechanical. This is the sort of paper that belongs in a reading group for people who build or test AV interaction models. It is not a full causal study, but it gives a concrete empirical anchor that modelers can cite or test against. I would send it to peer review with a request that reviewers focus on detector validation and any AV-specific motion confounds.

Referee Report

1 major / 1 minor

Summary. The paper claims that human drivers accept shorter longitudinal gaps when cutting in front of Waymo autonomous vehicles (AVs) than human-driven vehicles (HDVs), based on analysis of the Waymo Open Motion Dataset. Using an eight-criterion lane-change detector, the authors identify 706 AV-targeted and 3,172 HDV-targeted cut-in events, reporting a median gap reduction of 1.99 meters (7.58 m vs. 9.57 m, p=5.76e-8, d=-0.224) that persists in speed-matched resampling, along with higher cut-in speeds and more frequent short-gap events for AVs.

Significance. This result, if robust, has important implications for the design of AV motion planners and the calibration of traffic simulation models, highlighting a safety-relevant behavioral asymmetry where humans may take advantage of AV predictability. The manuscript is strengthened by its grounding in large-scale naturalistic data, inclusion of effect sizes, p-values, and a speed-matched robustness check. These elements provide a solid foundation for the empirical claim.

major comments (1)

[§3] §3: The eight-criterion lane-change detector applies fixed kinematic thresholds on relative position, velocity, and heading. Because AVs have lower motion variance than HDVs, these thresholds may produce higher recall for AV targets, particularly for short-gap events, biasing the median gap downward. The speed-matched resampling conditions only on speed and does not correct for differences in trajectory predictability that affect detector performance. This is load-bearing for the central claim that the event sets are comparable.

minor comments (1)

The abstract provides only a high-level description of the detector; including the specific criteria or a reference to the full definition in §3 would improve readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

Thank you for the detailed review and for highlighting the potential bias in our lane-change detection method. We provide a point-by-point response to the major comment below. We believe the core finding is robust but agree that additional checks are warranted to confirm the comparability of the event sets.

read point-by-point responses

Referee: The eight-criterion lane-change detector applies fixed kinematic thresholds on relative position, velocity, and heading. Because AVs have lower motion variance than HDVs, these thresholds may produce higher recall for AV targets, particularly for short-gap events, biasing the median gap downward. The speed-matched resampling conditions only on speed and does not correct for differences in trajectory predictability that affect detector performance. This is load-bearing for the central claim that the event sets are comparable.

Authors: We acknowledge this as a valid concern. The fixed thresholds are applied to the kinematics of the cutting-in HDV relative to the target vehicle, and lower variance in AV trajectories could indeed facilitate meeting the detection criteria more readily for certain events. While the speed-matched resampling helps control for one confounding factor, it does not fully address differences in predictability. To strengthen the analysis, we will revise the manuscript to include: (1) a comparison of the distributions of all eight kinematic features between the AV-targeted and HDV-targeted events, (2) an additional robustness check matching on both speed and the standard deviation of the target's longitudinal acceleration (as a proxy for variance), and (3) a brief discussion of this potential limitation. These additions will allow readers to better evaluate the comparability of the two event sets. revision: yes

Circularity Check

0 steps flagged

No significant circularity; direct empirical comparison of observed gaps

full rationale

The paper reports a statistical comparison of median accepted gaps (7.58 m vs. 9.57 m) extracted from 706 AV-targeted and 3,172 HDV-targeted cut-in events in the Waymo Open Motion Dataset. The eight-criterion lane-change detector applies fixed kinematic thresholds; the medians, speed statistics, and chi-squared tests are computed directly from the resulting event sets. No equations, fitted parameters, self-citations, or ansatzes are invoked that would reduce the reported difference to a definitional or self-referential input. The result is an observational finding whose validity rests on the detector's unbiased application across classes, not on any internal derivation loop.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The claim depends on the accuracy and lack of bias in the custom lane-change detector and on the assumption that the extracted events represent typical driver behavior around AVs.

axioms (1)

domain assumption The eight-criterion lane-change detector identifies cut-in events with equal validity for HDV-to-AV and HDV-to-HDV cases
Detector is the sole source of the 706 and 3,172 event counts; any differential error would directly affect the gap comparison.

pith-pipeline@v0.9.0 · 5590 in / 1302 out tokens · 101720 ms · 2026-05-09T14:16:00.764297+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

12 extracted references · 6 canonical work pages

[1]

Large scale interactive motion forecasting for autonomous driving: The Waymo Open Motion Dataset,

S. Ettinger, S. Cheng, B. Caine, C. Liu, H. Zhao, S. Pradhan, Y . Chai, B. Sapp, C. Qi, Y . Zhou, Z. Yang, A. Chouard, P. Sun, J. Ngiam, V . Vasudevan, A. McCauley, J. Shlens, and D. Anguelov, “Large scale interactive motion forecasting for autonomous driving: The Waymo Open Motion Dataset,” inProc. IEEE/CVF Int. Conf. Computer Vision (ICCV), pp. 9710–9719, 2021

2021
[2]

Examining lane change gap accep- tance, duration and impact using naturalistic driving data,

M. Yang, X. Wang, and M. Quddus, “Examining lane change gap accep- tance, duration and impact using naturalistic driving data,”Transportation Research Part C: Emerging Technologies, vol. 104, pp. 317–331, 2019. DOI: 10.1016/j.trc.2019.05.024

work page doi:10.1016/j.trc.2019.05.024 2019
[3]

A comprehensive examination of naturalistic lane changes,

S. E. Lee, E. C. B. Olsen, and W. W. Wierwille, “A comprehensive examination of naturalistic lane changes,” NHTSA Report DOT HS 809 702, U.S. Dept. of Transportation, Washington, D.C., 2004

2004
[4]

Analysis of cut-in behavior based on naturalistic driving data,

X. Wang, H. Xu, G. Ma, J. Xu, H. Xu, L. Wang, and D. Hur- witz, “Analysis of cut-in behavior based on naturalistic driving data,”Accident Analysis & Prevention, vol. 124, pp. 127–137, 2019. DOI: 10.1016/j.aap.2019.01.006

work page doi:10.1016/j.aap.2019.01.006 2019
[5]

Driving behaviour: models and challenges,

T. Toledo, “Driving behaviour: models and challenges,”Transport Reviews, vol. 27, no. 1, pp. 65–84, 2007

2007
[6]

Estimation of gap acceptance parameters within and across the population from direct roadside observation,

C. F. Daganzo, “Estimation of gap acceptance parameters within and across the population from direct roadside observation,”Transportation Research Part B, vol. 15, no. 1, pp. 1–15, 1981

1981
[7]

Behavioral adaptations of human drivers interacting with automated vehicles,

S. Soni, N. Reddy, A. Tsapi, B. van Arem, and H. Farah, “Behavioral adaptations of human drivers interacting with automated vehicles,” Transportation Research Part F: Traffic Psychology and Behaviour, vol. 86, pp. 48–64, 2022. DOI: 10.1016/j.trf.2022.01.006

work page doi:10.1016/j.trf.2022.01.006 2022
[8]

Characterizing car-following behaviors of human drivers when following automated vehicles using the real-world dataset,

X. Wen, D. He, S. Jian, and C. Huang, “Characterizing car-following behaviors of human drivers when following automated vehicles using the real-world dataset,”Accident Analysis & Prevention, vol. 172, p. 106689,
[9]

DOI: 10.1016/j.aap.2022.106689

work page doi:10.1016/j.aap.2022.106689 2022
[10]

Influence of autonomous vehicles on car-following behavior of human drivers,

Y . Rahmati, M. Khajeh Hosseini, A. Talebpour, B. Swain, and C. Nelson, “Influence of autonomous vehicles on car-following behavior of human drivers,”Transportation Research Record, vol. 2673, no. 12, pp. 367–379,
[11]

DOI: 10.1177/0361198119862628

work page doi:10.1177/0361198119862628
[12]

Pedestrians, autonomous vehicles, and cities,

A. Millard-Ball, “Pedestrians, autonomous vehicles, and cities,”Journal of Planning Education and Research, vol. 38, no. 1, pp. 6–12, 2018. DOI: 10.1177/0739456X16675674

work page doi:10.1177/0739456x16675674 2018