Extended Longitudinal Motion Planning for Autonomous Vehicles on Highways Including Lane Changing Prediction

Khelfa, Basma; Tordeux, Antoine · 2020 · Crossref

DOI: 10.1007/978-3-030-55973-1_61

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 predicting lane-changing maneuvers for autonomous vehicles and advanced driver assistance systems (ADAS) on highway ramps. Accurate prediction is critical for safety, as improper lane changes contribute significantly to crashes and fatalities. While existing models often rely on explicit reasoning processes or single-time-step observations, this study proposes a generalized machine learning approach that utilizes high-dimensional, sequential feature data to predict whether an ego vehicle will change lanes or maintain its current lane. The researchers utilized the Next Generation Simulation (NGSIM) dataset, which contains detailed vehicle trajectory data from US-101 and I-80 highways in California. They defined a specific scenario involving an ego vehicle and three surrounding vehicles: one directly ahead in the current lane, and one preceding and one following in the target lane. Features included longitudinal and lateral distances, relative velocities, and accelerations for these vehicles. To capture temporal dynamics, the authors concatenated data from six time steps (3.0 seconds) prior to the event, resulting in 84-dimensional feature vectors. The dataset was filtered to include 645 lane-changing events and 663 lane-keeping events. Three classification algorithms were employed: Logistic Regression (LR), Support Vector Machine (SVM), and Artificial Neural Network (ANN). Hyperparameters for each model were optimized using a genetic algorithm to maximize the Area Under the Receiver Operating Characteristic curve (AUROC). The results demonstrate that lane-changing behavior is highly predictable using these methods. The ANN achieved the best performance, with a test set accuracy of 96.9%, sensitivity of 98.9%, precision of 94.9%, and an AUROC of 0.9961. The SVM followed with 85.7% accuracy and an AUROC of 0.9078, while LR achieved 82.1% accuracy and an AUROC of 0.9047. Feature selection analysis identified longitudinal distance to the preceding vehicle in the target lane and lateral distance to the front center vehicle as the most indicative features. Error analysis revealed that misclassifications primarily stemmed from sensor errors in the dataset causing discontinuous data or from ego vehicles hesitating during the maneuver, causing ambiguous lateral position signals. The study concludes that machine learning models, particularly ANNs, can effectively predict lane changes with high accuracy by leveraging sequential traffic environment data. This capability supports the development of safer autonomous driving systems by allowing vehicles to anticipate human driver behavior. The authors suggest future work should extend the model to predict multiple outputs, such as lane-changing direction and velocity, and further generalize the model to broader traffic scenarios.

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-prismaquant on 2026-06-26; 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