Modeling driver behavior at roundabouts: Results from a field study
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 challenge of predicting driver behavior at roundabouts to support the development of Advanced Driving Assistance Systems (ADAS). The primary motivation is to enhance safety, particularly for vulnerable road users like cyclists, by enabling warning systems that can reliably predict whether a driver intends to exit the roundabout or continue circulating. Existing research has largely focused on highway or intersection scenarios, leaving a gap in modeling specific roundabout behaviors. The authors aim to develop a classification model that determines if a vehicle will leave at an upcoming exit or stay within the roundabout, using real-world driving data. To achieve this, the researchers conducted a field study in Braunschweig, Germany, involving seven participants who drove through a track containing three roundabouts with varying geometries. Each participant completed at least 30 drives, covering all entry-exit combinations. Data were collected at 50 Hz using sensors for steering angle, steering angle velocity, acceleration, velocity, and GPS position. The authors selected steering angle and its velocity as the primary features for prediction, arguing that velocity and acceleration are too heavily influenced by external traffic conditions. The modeling process involved mapping time-series data to distance-based features, scaling inputs, and splitting the dataset into training (80%) and testing (20%) sets. The study distinguished three geometric scenarios based on the angle between entry and exit points, developing specific sub-models for each. Support Vector Machines (SVM) were employed as the classification algorithm, with parameters optimized via 5-fold cross-validation. The results demonstrate that SVM is a robust and efficient method for this binary classification problem. Using a linear kernel with a parameter C = 100, the model achieved high recognition accuracies across all scenarios. Specifically, the model reached 98.1% accuracy in Scenario 1 (adjacent exits, angle < 110°) at 13.8 meters before the exit, 97.4% in Scenario 2 (adjacent exits, angle > 110°) at 11.4 meters before the exit, and 98.5% in Scenario 3 (non-adjacent exits, angle > 110°) at 14.1 meters before the exit. The classification process required only 0.01 seconds per site. The findings confirm that steering angle and steering angle velocity provide sufficient information to predict driver intent with greater than 95% accuracy approximately 11 meters before the exit. The significance of this work lies in filling a critical research gap regarding driver behavior modeling at roundabouts. By proving that driver intent can be predicted with high accuracy and low latency using simple vehicle dynamics data, the study supports the feasibility of implementing real-time warning systems in ADAS. These systems could alert drivers to potential risks, such as overlooking cyclists, thereby reducing accident rates. The authors conclude that future work should explore additional feature inputs, such as speed and acceleration, test other classifiers like Hidden Markov Models, and refine the model for generic roundabout geometries to further optimize the timing of assistance warnings.
Key finding
Support Vector Machine models using steering angle and steering angle velocity as inputs can predict whether a driver will exit or stay in a roundabout with over 95% accuracy at distances of approximately 11 to 14 meters before the exit.
Methodology
field_study
Sample size: 7
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 author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 7 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| 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).
- Theoretical Contribution: computational model