Stochastic Mixture Modeling of Driving Behavior During Car Following
DOI: 10.6109/jicce.2013.11.2.095
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of modeling stochastic driver behavior during car-following maneuvers, specifically focusing on predicting pedal operations (gas and brake pressure) based on vehicle velocity and following distance. The research is motivated by the limitations of existing models, such as Gaussian Mixture Models (GMM), which require pre-specifying the number of latent components and often suffer from overfitting or local maxima issues. Furthermore, individual driver models trained on sparse data fail to generalize to unseen driving situations, while universal models lack individual specificity. The authors propose a stochastic framework that aggregates individual and general driving characteristics to improve prediction accuracy. The methodology employs a combined probabilistic model consisting of two components. First, a Dirichlet Process Mixture Model (DPM), a non-parametric Bayesian approach, is used to model individual driving styles. DPM automatically determines the optimal number of mixture components from sparse individual observations, avoiding the model selection problem inherent in finite GMMs. Second, a universal background model is trained using a standard GMM with Expectation-Maximization (EM) on a large dataset from multiple drivers to capture general driving patterns. These two distributions are combined into a single aggregate model using a weighted linear aggregation, allowing the system to emphasize individual characteristics while backing off to general patterns for unmatched parameter spaces. The model was evaluated using real-world driving data from 64 drivers, comprising approximately 300 minutes of car-following events. The input features included vehicle velocity, following distance, and pedal patterns with their first- and second-order derivatives. Experimental results demonstrated that the proposed combined model significantly outperformed individual DPM models, universal GMM models, and driver-adapted models using Maximum A Posteriori (MAP) adaptation. The individual DPM model alone showed the worst performance due to a lack of coverage for unseen driving situations in the test data. However, merging the individual model with the universal background model yielded substantial improvements. The optimal performance was achieved with a weighting scale of approximately 0.3 for the background model, resulting in a Signal-to-Deviation Ratio (SDR) of 19.95 dB for gas-pedal prediction. The combined model effectively balanced the specificity of individual driving styles with the robustness of general driving patterns, providing superior prediction accuracy across various mixture configurations. The significance of this work lies in its ability to create robust, driver-dependent behavior models without requiring extensive individual training data or manual model selection. By leveraging non-parametric Bayesian methods for individual modeling and combining them with universal background models, the approach offers a practical solution for anticipating driver actions in intelligent transportation systems. This framework enhances the reliability of driver behavior prediction, particularly in scenarios where individual training data is sparse or contains unseen driving contexts, thereby advancing the field of stochastic driver modeling and autonomous vehicle interaction.
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.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | Crossref | — | — | 1 | 2026-06-24 |
| archive | success | canonical_url | — | — | 1 | 2026-06-26 |
| extract | success | pdftotext | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-26 |
| chunk | success | chunk | — | — | 1 | 2026-06-26 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-26 |
| enrich | success | openalex | — | — | 1 | 2026-06-26 |
| promote | success | — | — | — | 1 | 2026-06-24 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-26 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; 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).
- Methodological Resource: tool software
- Theoretical Contribution: computational model, theory or model