Development of Machine-Learning Models for Autonomous Vehicle Decisions on Weaving Sections of Freeway Ramps

Lin, Brian T W · 2023 · ROSA P / University of Michigan. Center for Connected and Automated Transportation

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 study addresses the challenge of developing safe, human-data-driven automated lane-change models for autonomous vehicles (AVs) navigating freeway weaving sections. These sections, characterized by limited auxiliary lanes where vehicles must merge and diverge simultaneously, are high-risk areas for crashes. Existing research often relies on algorithm-driven optimization or macroscopic traffic flow models that lack driver-centric behavioral insights. The authors aimed to create decision-making models that predict when and how an AV should execute a weaving maneuver based on naturalistic human driving data, ensuring the strategies are both safe and acceptable to human drivers. The research utilized data from the Integrated Vehicle-Based Safety System Field Operational Test (IVBSS FOT), comprising over 200,000 miles driven by 108 adult drivers. The team identified 53 weaving sections in southeastern Michigan and extracted variables including vehicle dynamics, roadway geometry, and surrounding vehicle positions. Due to the time-varied nature of lane-change decisions and the absence of "no-go" counterfactuals in the dataset, the authors employed survival analysis using Cox proportional hazards models rather than traditional machine learning or logistic regression. These models were validated using five-fold repeated cross-validation and evaluated through MATLAB Simulink computer simulations. Finally, the models were implemented on an AV platform at the Mcity test facility using an augmented reality environment to simulate interactions with virtual "ghost" vehicles, as physical traffic testing was restricted for safety reasons. The Cox models identified significant predictors for lane-change probability, including longitudinal speed, lateral speed, weaving position, and relative speed (range rate) to surrounding vehicles. For weaving toward an exit, the model achieved an area under the curve (AUC) of 0.89, identifying 87% of weaving events with 81% overall accuracy. For weaving toward an entrance, the AUC was 0.88, with 83% identification of weaving events and 81% accuracy. Computer simulations revealed critical safety limitations: collisions were likely if the AV matched the speed of the vehicle in the target lane. Furthermore, if the AV traveled faster than 70 mph while the surrounding vehicle exceeded 55 mph, the model often failed to initiate a lane change before the end of the weaving section, resulting in missed exits or entrances. The augmented reality demonstration successfully executed maneuvers without virtual collisions, confirming the platform's capability to interact with simulated traffic. The study concludes that while survival analysis effectively models human weaving decisions, current implementations require improvement to ensure complete collision-free environments. The findings highlight the necessity of incorporating speed adjustment strategies for AVs before entering weaving sections and expanding the variety of weaving scenarios in training data. This work provides a foundational, driver-centric approach for automated driving systems to handle complex ramp maneuvers, bridging the gap between theoretical traffic models and practical, safe autonomous vehicle deployment.

Key finding

Cox proportional hazards models derived from naturalistic driving data achieved at least 81% accuracy in predicting autonomous vehicle lane-change decisions in freeway weaving sections, with speed, lateral speed, weaving position, and relative speed to other vehicles identified as significant predictors.

Methodology

naturalistic

Sample size: 108

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 (6 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.

• lane changing

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