Scholarlink

<p>This paper proposes an integrated collision-avoidance framework based on sparse Gaussian process regression (Sparse-GPR) and stochastic model predictive control (SMPC) for Level 4 autonomous-driving systems. Conventional collision-avoidance systems rely on deterministic metrics such as time-to-collision, which can cause unnecessary or delayed interventions under uncertain conditions. To overcome that limitation, a Sparse-GPR-based probabilistic trajectory-prediction method is introduced to quantify uncertainties in surrounding-vehicle behaviors. The predicted trajectories and their associated uncertainties are explicitly incorporated into an SMPC framework to simultaneously optimize braking and steering controls for cooperative emergency maneuvers. The proposed framework is evaluated in a camera-based over-the-air hardware-in-the-loop simulation environment for three representative scenarios: abrupt lane changes, subtle lateral movements, and abrupt deceleration following lane changes. Comparative analyses demonstrate improvements in prediction accuracy, proactive collision risk management, and integrated braking–steering control performance, compared with deterministic approaches. The integrated Sparse-GPR and SMPC method reduces unnecessary interventions and enhances emergency-handling capabilities, highlighting the need for probabilistic prediction and integrated control optimization to improve the safety of autonomous driving.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrated Collision Avoidance Framework for Autonomous Vehicles Using Sparse Gaussian Process Regression and Stochastic Model Predictive Control

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Yunseok Baek,
  • Hyeongcheol Lee

摘要

This paper proposes an integrated collision-avoidance framework based on sparse Gaussian process regression (Sparse-GPR) and stochastic model predictive control (SMPC) for Level 4 autonomous-driving systems. Conventional collision-avoidance systems rely on deterministic metrics such as time-to-collision, which can cause unnecessary or delayed interventions under uncertain conditions. To overcome that limitation, a Sparse-GPR-based probabilistic trajectory-prediction method is introduced to quantify uncertainties in surrounding-vehicle behaviors. The predicted trajectories and their associated uncertainties are explicitly incorporated into an SMPC framework to simultaneously optimize braking and steering controls for cooperative emergency maneuvers. The proposed framework is evaluated in a camera-based over-the-air hardware-in-the-loop simulation environment for three representative scenarios: abrupt lane changes, subtle lateral movements, and abrupt deceleration following lane changes. Comparative analyses demonstrate improvements in prediction accuracy, proactive collision risk management, and integrated braking–steering control performance, compared with deterministic approaches. The integrated Sparse-GPR and SMPC method reduces unnecessary interventions and enhances emergency-handling capabilities, highlighting the need for probabilistic prediction and integrated control optimization to improve the safety of autonomous driving.