Transferability of Car-Following Models Between Driving Simulator and Field Traffic

Papadimitriou, Stavros; Choudhury, Charisma F. · 2017 · Crossref

DOI: 10.3141/2623-07

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the critical research gap regarding the transferability of car-following models between driving simulators and real-world field traffic. While driving simulator data offers controlled, reproducible conditions suitable for testing hypothetical scenarios, it often lacks behavioral realism. Conversely, field data reflects true driving behavior but suffers from measurement errors and a lack of driver-specific information. Previous validation studies have focused on isolated measures like speed or reaction time, yielding mixed results on absolute validity. This paper rigorously investigates whether mathematical models of driving behavior developed using simulator data can be transferred to real-life conditions. The researchers developed stimulus-response based car-following models using three distinct microscopic data sources: trajectory data from the University of Leeds Driving Simulator (UoLDS), video-derived trajectory data from the M1 motorway in the UK, and video-derived data from Interstate 80 in the USA. The simulator data involved 40 participants driving a Jaguar S-type cab in a controlled motorway scenario, while the field data represented comparable motorway segments with similar flow levels. Model parameters were estimated using Maximum Likelihood Estimation, assuming a reaction time of 0.5 seconds. To assess transferability, the authors employed two statistical methods: t-tests of individual parameter equality to compare specific coefficients, and the Transferability Test Statistic (TTS) to evaluate the overall statistical equivalence of the models. The results indicate that while the models exhibit transferability at the structural level, they do not fully transfer at the parameter level for either the UK or US field scenarios. Specifically, the t-tests revealed statistically significant differences in key parameters, such as relative speed and sigma (error variance), between the simulator and field models. For instance, the relative speed coefficient differed significantly between the simulator and both field datasets, with t-statistic differences exceeding the critical threshold of 1.96. Similarly, the TTS results rejected the null hypothesis of statistical equivalence, confirming that the parameter estimates from the simulator model were not statistically equivalent to those derived from field data. Although aggregate characteristics like mean acceleration and time headway showed similarities, the underlying behavioral parameters governing driver responses to stimuli varied significantly between the controlled simulator environment and real-world traffic. The significance of these findings lies in the validation of driving simulator utility and limitations. The study confirms that while simulators can capture the general structure of car-following behavior, they cannot perfectly replicate the specific parameter values observed in real traffic. This suggests that simulator-based models may require calibration or adjustment before being applied to real-world traffic simulation, safety analysis, or Intelligent Transportation Systems deployment. The research provides a rigorous methodological framework for future studies aiming to validate and transfer behavioral models across different data sources, highlighting the need for caution when applying simulator-derived parameters to field applications.

Key finding

Car-following models developed from driving simulator data are transferable to real-world field traffic at the model structure level but exhibit significant differences in parameter estimates, indicating incomplete transferability at the parameter level.

Methodology

mixed_methods

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.

StageOutcomeToolModelPromptAttemptsCompleted
discover success Crossref 1 2026-06-06
archive success canonical_url 1 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
promote success 1 2026-06-06
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).