Driver behavior in traffic.

Abbas, Montasir; Medina, Alejandra; Chong, Linsen; Higgs, Bryan; McGhee, Catherine; Fontain, Mike; Goodall, Noah; Kautzsch, Lukas; Leonhardt, Axel; Ova, Kiel · 2012 · ROSA P / United States. Federal Highway Administration

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 report addresses the limitation of existing traffic analysis and management tools, which fail to adequately model how drivers recognize their environment and respond with behaviors that vary by situation. Current literature on driver behavior is often narrow in scope, limited to specific locations, and lacks the ability to simulate both normal and safety-critical driving. The research aimed to develop methods for modeling driver behavior in traffic using naturalistic driving data, specifically focusing on creating more accurate and sensitive traffic simulation models that account for driver heterogeneity. The study employed a multi-stage methodology involving naturalistic driving data from car and truck databases. First, the researchers calibrated and extended existing car-following models, specifically the Wiedemann model, by reconstructing its logic and introducing new thresholds based on longitudinal trajectory data. They developed a hybrid model combining the Wiedemann model with the Generalized Humanistic Rule (GHR) model. Second, the team utilized agent-based modeling with neuro-fuzzy reinforcement learning (NFACRL), an artificial intelligence technique, to train driver agents. These agents were trained on extracted safety-critical events from the naturalistic database to learn acceleration and steering actions. Robust agent activation techniques were developed using discriminant analysis to determine when specific behavioral agents should be triggered. Finally, the developed agents were implemented in the VISSIM simulation platform. The results demonstrated that the reconstructed Wiedemann model and the hybrid model provided better fits to observed driver data than the original model, with specific threshold equations derived for individual drivers. The NFACRL-trained agents successfully simulated events from different drivers, proving the existence of behavior heterogeneities. Cross-validation results showed that the agents could replicate the acceleration and yaw angle trajectories of real drivers during safety-critical events. When implemented in VISSIM, the agents exhibited behavior closely resembling the original driver data. Quantitative assessments of vehicle behavior in the simulation environment confirmed that the agents accurately reproduced the kinematic properties of the naturalistic data. Discriminant analysis successfully identified conditions for activating specific driver behaviors, distinguishing between normal car-following and near-crash events. The significance of this work lies in providing the transportation industry with methods to develop more accurate traffic simulation models. By integrating agent-based artificial intelligence with naturalistic data, the study enables the simulation of complex traffic situations, including incidents and safety-critical events, which traditional models often overlook. The findings support the development of next-generation traffic simulation tools that can better predict driver responses in diverse scenarios. The report recommends future research to train agents using more extensive data to cover a wider region in the Wiedemann regime space and to perform sensitivity analysis on agent training parameters to further refine model accuracy.

Key finding

Individual driver agents trained with NFACRL on VTTI naturalistic data closely reproduced observed car-following and near-crash trajectories in VISSIM, and a hybrid Wiedemann–GHR model with new pass and hook thresholds calibrated to naturalistic data reduced car-following error by 5–43% versus standard Wiedemann.

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. Discovered via bulk_ingest_rosap on 2026-05-23 (5 acquisition events logged).

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.

• following distance
• traffic density
• mental model of traffic
• naturalistic crash near crash
• situational awareness
• speed choice

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).

• Empirical Findings: behavioral performance data
• Methodological Resource: dataset resource
• Theoretical Contribution: computational model