Characterization and Validation of Lane-Change Precursors Based on Vehicle Kinematic States for Proactive Safety in Autonomous Driving
Year: 2026
Type: article
Abstract: Lane changes (LCs) and cut-ins can increase collision risk and reduce an ego vehicle’s response time. Existing intention and trajectory prediction studies provide limited explanation of how preceding kinematic changes develop relative to normal lane keeping (LK). This study defines a lane-change precursor as a state in which kinematic deviations that are not readily accounted for by normal LK under comparable driving conditions persistently develop and strengthen. Two complementary indices quantify the rarity of cumulative directional lateral motion, evaluated in both lateral directions, and the departure of the joint lateral velocity, lateral acceleration, heading, and yaw-rate state from normal LK. Application to naturalistic trajectories from Next Generation Simulation (NGSIM) and the highD highway trajectory dataset showed increasing differences between lane-changing and normal LK vehicles toward lane transition. Multivariate state discrimination increased toward lane transition in NGSIM, whereas clear separation was observed from the beginning of the analysis window in highD. Precursors were also observed before individual lane transitions, with the two indices showing different detection and false-positive characteristics. Linking precursor information to logistic regression and long short-term memory outputs improved prediction precision and suppressed false positives during normal LK by requiring subsequent confirmation after an observed precursor. Cumulative motion and multivariate state deviation provide complementary kinematic and statistical information for interpreting and confirming maneuver predictions through observed vehicle motion. This information improves prediction reliability and can support future interaction risk assessment.
Source: Machines
Authors: Joong Ho Lee, Kiwon Yeom
Institutions: Keimyung University, Sangmyung University
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7225610124
FWCI:
Cites:
Cited by:
Related to: