Interaction-Aware Probabilistic Trajectory Prediction of Cut-In Vehicles Using Gaussian Process for Proactive Control of Autonomous Vehicles

Yoon, Youngmin; Kim, Changhee; Lee, Jong‐Min; Yi, Kyongsu · 2021 · OpenAlex-citations

DOI: 10.1109/access.2021.3075677

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 accurately predicting the trajectories of cut-in vehicles to enhance the safety and ride quality of autonomous vehicles (AVs). Lane-change maneuvers are a significant source of traffic accidents and latency, yet conventional prediction methods often fail to account for the complex interactions between the cutting-in vehicle and surrounding traffic. The authors propose an interaction-aware probabilistic trajectory prediction framework that utilizes Gaussian Process Regression (GPR) to estimate behavioral parameters, thereby enabling proactive control via Model Predictive Control (MPC). The methodology consists of two primary modules: behavioral parameter estimation and vehicle state prediction. First, GPR models are trained using real-world trajectory data from the Next Generation Simulation (NGSIM) database, specifically the I-80 and US-101 datasets, which contain 537 extracted cut-in events. The GPR models estimate three key behavioral parameters representing lane-change aggressiveness: the remaining longitudinal distance, the final lateral offset, and the time remaining to complete the maneuver. These models incorporate input features describing both the cut-in vehicle’s progress and the relative configuration of adjacent vehicles in the target lane. Second, an Extended Kalman Filter (EKF) predicts the future states of the cut-in vehicle. The EKF utilizes a path-following model to generate "virtual measurements" of yaw rate and acceleration based on the probabilistic outputs of the GPR, allowing for a recursive estimation of the vehicle's trajectory and uncertainty. The study validates the proposed predictor through computer simulations and real-world autonomous driving vehicle tests. The GPR models demonstrated reliable estimation performance, with Mean Absolute Error and Root Mean Square Error statistics bounded within acceptable levels, despite higher variance in left lane-change scenarios due to overtaking dynamics. When integrated into an MPC framework, the interaction-aware predictor significantly improved prediction accuracy in multi-vehicle scenarios compared to conventional methods that ignore inter-vehicle interactions. The significance of this work lies in its demonstration that accurate, interaction-aware prediction directly enhances control performance. The results indicate that the proposed system allows the autonomous vehicle to reduce control effort and improve passenger ride quality during cut-in events while maintaining safety guarantees. By explicitly modeling the influence of surrounding vehicles on lane-change behavior, the approach provides a more robust solution for proactive maneuvering in dense traffic environments, addressing a critical gap in current autonomous driving algorithms.

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.

• anticipation