Lane-Change Initiation and Planning Approach for Highly Automated Driving on Freeways

Arbabi, Salar; Dixit, Shilp; Zheng, Ziyao; Oxtoby, David; Mouzakitis, Alexandros; Fallah, Saber · 2020 · Crossref

DOI: 10.1109/vtc2020-fall49728.2020.9348630

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 quantifying and encoding human driver preferences for tactical decision-making in highly automated vehicles, specifically for lane-change initiation on freeways. Traditional approaches often rely on hand-crafted objective functions or rule-based systems that are difficult to tune and lack generalization. While machine learning methods like Reinforcement Learning and Supervised Learning offer alternatives, they typically require high-fidelity simulators or large-scale annotated data. The authors propose a low-complexity approach that learns decision conditions directly from naturalistic driving data, eliminating the need for engineered objective functions or explicit expert rules. The methodology utilizes vehicle trajectories captured from CCTV cameras on US 101 and I-80 in California. From this dataset, 600 discretionary lane-change maneuvers were automatically extracted and labeled based on lateral speed thresholds. The decision model employs a Random Forest classifier to map driving context features to two maneuver classes: lane keeping and discretionary lane change. Key features include relative positions and velocities with surrounding vehicles, time-to-collision metrics, and a novel "utility" feature that models driver contentment by accumulating dissatisfaction when following a slower vehicle. The system integrates this decision model with a Model Predictive Control (MPC) framework for trajectory generation. The MPC solves a finite-horizon optimization problem with safety constraints, using artificial potential fields to define collision-free zones and ensuring the vehicle tracks target states for lateral position, heading, and velocity. The results demonstrate that the Random Forest decision model replicates human drivers’ discretionary lane-change decisions with up to 92% accuracy. This represents a significant improvement over a baseline accuracy of 67% achieved without the engineered utility and time-to-collision features. Feature importance analysis identified the longitudinal distance to the rear-left vehicle and the utility feature as the most critical predictors. Simulations of the integrated system, where the autonomous vehicle’s actions were logged against ground truth data, showed that the generated trajectories closely matched human behavior in overtaking scenarios. The system successfully initiated lane changes when the probability exceeded an 80% confidence threshold, maintaining safe distances and smooth control inputs. The significance of this work lies in its ability to learn complex, context-dependent driving behaviors from real-world data without relying on subjective rule engineering or expensive simulation environments. By using a data-driven decision model combined with a robust MPC planner, the approach offers a scalable solution for autonomous highway driving. The study highlights the importance of incorporating psychological factors, such as driver contentment, into feature sets to improve classification accuracy. Furthermore, the modular architecture allows for potential personalization of the decision model based on individual driver disengagements, paving the way for more adaptable and human-like autonomous driving systems.

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.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• lane changing
• situational awareness

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