Scholarlink

To address the uncertainty challenges in traversability mapping and path planning in off-road environments, this study proposed a unified framework integrating Bayesian Kernel Inference (BKI), Model Predictive Path Integral (MPPI) algorithm, and Conditional Value at Risk (CVaR). By employing BKI for elevation regression, 2.5D grid maps are constructed to quantify terrain uncertainties. While planning, a CVaR-based MPPI cost function is designed to enhance algorithmic robustness through risk-sensitive metrics. Simulation shows that: (1) The planning success rate of CVaR-MPPI outperforms baseline MPPI by 40%; (2) In real-time planning for distant targets, the average Hausdorff distance of trajectory samples has decreased by 2.016 m (0.671 m vs. 2.687 m). Furthermore, flexible switching between aggressive and conservative strategies is achieved by setting up different \(\alpha \) values.

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

Risk-Aware MPPI Path Planning Based on Bayesian Kernel Inference

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jingshi Zhang,
  • Yunxiao Shan,
  • Xundao Zhou

摘要

To address the uncertainty challenges in traversability mapping and path planning in off-road environments, this study proposed a unified framework integrating Bayesian Kernel Inference (BKI), Model Predictive Path Integral (MPPI) algorithm, and Conditional Value at Risk (CVaR). By employing BKI for elevation regression, 2.5D grid maps are constructed to quantify terrain uncertainties. While planning, a CVaR-based MPPI cost function is designed to enhance algorithmic robustness through risk-sensitive metrics. Simulation shows that: (1) The planning success rate of CVaR-MPPI outperforms baseline MPPI by 40%; (2) In real-time planning for distant targets, the average Hausdorff distance of trajectory samples has decreased by 2.016 m (0.671 m vs. 2.687 m). Furthermore, flexible switching between aggressive and conservative strategies is achieved by setting up different \(\alpha \) values.