DrivingResearch.

A merging interaction model explains human drivers' behaviour from input signals to decisions

Abbink, David A. · 2023 · arXiv (Cornell University)

archive: archived pipeline: cataloged verified

Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of creating safe and socially acceptable interactions between automated vehicles (AVs) and human drivers, specifically focusing on highway merging scenarios. Current driver models typically isolate specific aspects of driving behavior—such as high-level decisions, safety margins, or low-level control inputs—and often rely on game-theoretic assumptions that treat drivers as rational, non-communicating agents. The authors argue that these limitations hinder the development of interaction-aware AVs because real-world merging is a reciprocal process involving implicit communication and joint decision-making. To bridge this gap, the study introduces a unified computational model based on the Communication-Enabled Interaction (CEI) framework, which captures behavior across all three levels: individual control inputs, joint safety margins, and high-level merging decisions. The researchers developed a model where drivers maintain a deterministic plan for their own trajectory and form a probabilistic belief about the other driver’s intentions based on implicit communication via vehicle kinematics (position, velocity, and acceleration). This belief, combined with the driver’s own plan, generates a perceived risk calculated as the probability of collision. If this risk exceeds a personalized threshold, the driver updates their plan to mitigate the danger. The model incorporates dynamic risk thresholds adjusted by incentive functions reflecting traffic norms. To validate the model, the authors used empirical data from a coupled, top-down view driving simulator experiment involving nine pairs of human participants. The experiment utilized a simplified merging scenario with 11 conditions varying in initial relative velocity and projected headway, resulting in 990 total trials. The model parameters were fitted to the human data using grid search and linear regression. The results demonstrate that the CEI-based model accurately replicates human behavior across all three behavioral levels. Qualitatively and quantitatively, the model matched the intermittent, piece-wise constant acceleration patterns observed in human drivers, including the timing of control inputs and the magnitude of velocity deviations. The model successfully captured individual differences in risk tolerance, explaining why some drivers actively mitigated risk while others remained passive. Regarding joint behavior, the model produced safety margins comparable to human drivers (mean gap of 4.8 m vs. 4.5 m) and correctly predicted the high-level outcomes of who merges first based on kinematic advantages. The model also replicated the relationship between initial kinematics and conflict resolution time. While minor discrepancies existed, such as slightly larger velocity deviations in the model due to edge-case miscommunications, the overall correlation between human and model behavior was strong. The significance of this work lies in providing a comprehensive explanation of human merging interactions that does not assume rationality or explicit communication. By demonstrating that communication-based beliefs and risk-perception mechanisms can explain behavior across multiple levels, the study offers a robust foundation for interaction-aware automated driving. This approach allows AVs to understand implicit traffic communication and adhere to social norms, leading to more predictable and acceptable automated behavior. Furthermore, the model provides insights into the fundamental mechanisms of human interactive capacities, potentially generalizing to other traffic scenarios beyond merging.

Key finding

A communication-enabled interaction model based on risk perception accurately predicts human merging behavior across individual control inputs, joint safety margins, and high-level decisions without assuming rational utility maximization.

Methodology

simulator

Sample size: 18

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-27.

StageOutcomeToolModelPromptAttemptsCompleted
discover success author_sweep — — 2 2026-05-27
archive success canonical_url — — 6 2026-06-06
extract success cached — — 3 2026-06-10
clean success clean — — 1 2026-06-07
chunk success chunk — — 1 2026-06-07
embed success embed Qwen/Qwen3-Embedding-8B — 1 2026-06-07
enrich skipped — — — 4 2026-07-02
promote success — — — 1 2026-06-04
summarize success llm qwen3.6-27b-prismaquant summ-v5 2 2026-06-10
tag success vector_similarity — — 15 2026-06-11
verify success — — — 2 2026-06-10

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).