Quantifying the Individual Differences of Driver' Risk Perception with Just Four Interpretable Parameters

Chen, Chen; Lan, Zhiqian; Zhan, Guojian; Lyu, Yao; Nie, Bingbing; Li, Shengbo Eben · 2022 · 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 addresses the challenge of modeling individual differences in driver risk perception to enable autonomous vehicles (AVs) to exhibit human-like, customized behaviors that foster trust and acceptance. Existing risk models are typically statistical or rely on simple distance metrics, failing to capture the nuanced, individual-specific ways drivers assess risk in non-conflict scenarios. The authors propose the Potential Damage Risk (PODAR) model, a physically interpretable framework that defines risk as the projection of future potential collision damage, attenuated by spatial and temporal factors. The study utilizes an open-access dataset from a driving simulation experiment involving eight participants who performed 308 obstacle avoidance trials each. The experimental design included 77 obstacles placed at varying longitudinal and lateral positions. Two indicators of perceived risk were recorded: an objective signal, the maximum steering angle (MSA) applied within one second of obstacle appearance, and a subjective signal, an oral response number (ORN) indicating perceived necessary steering effort. The PODAR model was calibrated for each driver using gradient descent to fit four interpretable parameters: prediction horizon ($T$), damage scale coefficient ($k$), temporal attenuation coefficient ($A$), and spatial attenuation coefficient ($B$). These parameters define how drivers discount risk over time and distance. The results demonstrate that the PODAR model effectively captures individual risk perception differences with high accuracy, achieving an average $R^2$ of 0.93 for objective signals and 0.88 for subjective signals, outperforming the previously proposed Driver’s Risk Field (DRF) model which required 47 parameters per driver. The calibrated parameters revealed distinct cognitive profiles: for instance, some drivers exhibited steep temporal attenuation, focusing only on near-future risks, while others maintained attention on far-future collisions. Spatially, drivers varied significantly in their sensitivity to lateral distance, with some indifferent to objects beyond 1 meter and others maintaining high risk perception for distant obstacles. Notably, subjective prediction horizons averaged 6 seconds, whereas objective actions reflected a horizon of approximately 4 seconds, indicating a delay between perception and reaction. Furthermore, drivers acted directly on spatial risks but did not instantly react to temporal risks, suggesting a decoupling of immediate action from future-oriented perception. The significance of this work lies in providing a concise, four-parameter model that quantifies individual risk perception with physical meaning, unlike the qualitative or computationally heavy alternatives. This approach allows AVs to customize their decision-making and planning algorithms to match specific human drivers’ risk profiles, potentially improving safety and user trust. The findings also highlight that handling obstacles within 0.2–1 meters significantly impacts driver acceptance, offering actionable insights for AV behavior design. The study lays a foundation for personalized autonomous driving by demonstrating that complex human risk assessment can be modeled efficiently and accurately.

Key finding

PODAR model with four physical-interpretable parameters successfully captures individual differences in perceived driving risk, enabling autonomous vehicles to develop customized, human-like behaviors that match individual driver risk perception patterns.

Methodology

lab_experiment

Sample size: 8

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.

• risk taking
• telematics crash prediction
• decision making risk perception
• situational awareness
• exposure measurement
• induced exposure

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: crash risk outcomes
• Theoretical Contribution: computational model, theory or model