A Simulator Dataset to Support the Study of Impaired Driving

Gideon, John; Tamura, Kimimasa; Sumner, Emily; Dees, Laporsha; Reyes Gomez, Patricio; Haq, Bassamul; Rowell, Todd; Balachandran, Avinash; Stent, Simon; Rosman, Guy · 2025 · arXiv

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 introduces a novel driving simulator dataset designed to study the effects of alcohol intoxication and cognitive distraction on driver behavior. Motivated by the high societal cost of impaired driving and the difficulty of measuring cognitive distraction, the authors address the lack of comprehensive datasets that combine these impairments with controlled road hazards. The study aims to support the development of systems capable of diagnosing various forms of impairment and understanding their impact on safety-critical driving behaviors. The experimental design involved 52 human subjects completing 23.7 hours of simulated urban driving using the CARLA simulator. The study comprised two versions: one involving alcohol intoxication (targeting a 0.10% blood alcohol content) and one involving only cognitive distraction. Participants performed two types of cognitive distraction tasks (1-back and sentence parsing) while driving. The dataset captures multimodal data, including vehicle controls, ground-truth scene geometry and semantics, driver gaze tracking, audio, and self-reported surveys. Crucially, the experiment included eight specific, scripted road hazards (e.g., pedestrian crossings, vehicle cut-ins) to elicit and record driver responses under normal, intoxicated, distracted, and combined impairment conditions. Validation of the dataset revealed distinct behavioral changes associated with each impairment. Alcohol intoxication significantly increased longitudinal acceleration, steering reversal rates, and collision frequencies, while cognitive distraction significantly reduced forward acceleration and increased pupil diameter. Gaze metrics, such as yaw standard deviation, fixation counts, and saccade counts, decreased significantly under both impairment types, indicating increased visual concentration on the road. Notably, the combined effect of intoxication and distraction showed complex interactions; for instance, the acceleration increase from intoxication was mitigated by the distraction-induced reduction in acceleration. Collision rates more than doubled for intoxicated drivers compared to sober drivers, whereas cognitive distraction alone had a marginal effect on collisions. The significance of this work lies in providing the first publicly available dataset that jointly explores alcohol intoxication, cognitive distraction, and hazard responses. By offering ground-truth labels for impairment states and hazard events, the dataset enables researchers to disentangle the overlapping effects of different impairments and evaluate detection algorithms. The authors conclude that gaze features are generally useful for detecting both impairment types, while vehicle control features are more effective for detecting intoxication. This resource supports future research into online impairment prediction, hazard response modeling, and the design of improved driver monitoring systems to enhance road safety.

Key finding

Releases the first publicly available driving-simulator dataset combining alcohol intoxication and cognitive distraction with controlled hazard responses, and shows that gaze-based features (fixation/saccade rates) decline under both impairment types with additive effects when impairments co-occur.

Methodology

simulator

Sample size: N=52 (v1 + v2 cohorts); ~25 hours of simulated urban driving

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 discover_arxiv on 2026-05-04 (4 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.

• polydrug
• distraction detection algorithms
• alcohol
• visual
• situational awareness
• visual occlusion

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).

• Methodological Resource: tool software, dataset resource
• Theoretical Contribution: computational model