One Diffusion Model, Two Roles: Guided Trajectory Planning and Safety-Critical Scenario Generation in Closed-Loop Simulation
研究使用单一预训练扩散交通模型解决自动驾驶中的轨迹规划和安全关键场景生成问题,通过SSDS解码器和DAPSE方案提升规划性能及生成挑战性测试场景。
研究使用单一预训练扩散交通模型解决自动驾驶中的轨迹规划和安全关键场景生成问题,通过SSDS解码器和DAPSE方案提升规划性能及生成挑战性测试场景。
为了解决城市环境中精细跨视角定位问题,本文通过构建大规模、多样化的OpenCVL数据集,并开发数据管理框架来过滤和修正姿态注释,从而提高模型性能。
本文通过优化场景表示、GPU执行等方法,解决了光线追踪在实时渲染中速度慢的问题,实现了比现有方法更快的渲染速度。
This work addresses the challenge of accurately modeling highly variable or elongated objects—such as ropes—in instance segmentation, where conventional anchor-based methods often fail to capture complex shapes. To overcome this limitation, the authors propose a novel anchor-free paradigm that leverages neural networks to predict, at each pixel, multi-directional distances to the nearest object boundary. These predictions are aggregated to approximate a signed distance function (SDF), from which a foreground mask is obtained via thresholding. By performing pixel-wise directional distance regression, the method flexibly represents arbitrary object geometries without relying on predefined anchors. Evaluated on the COCO dataset, the approach achieves superior segmentation IoU compared to state-of-the-art methods like YOLACT, demonstrating significantly enhanced adaptability to irregular object shapes.
This work addresses the challenge of achieving scalable and robust ego-vehicle trajectory prediction without reliance on high-definition maps. To this end, we propose an end-to-end system that integrates front-view images, vehicle kinematics, and navigation paths derived from standard-definition (SD) maps. We introduce SD map paths as a semantic prior for trajectory prediction for the first time, and design a dual-hypothesis fusion architecture with a gated classifier to handle challenges such as route corruption or visual ambiguity. Evaluated on 480,000 real-world driving scenarios spanning ten European countries and the United States, our method reduces the average displacement error (ADE) over an 8-second horizon by 16.9% compared to a baseline using only images and kinematics. We also release an open-source toolkit for SD path generation to support community benchmarking.
研究使用单一预训练扩散交通模型解决自动驾驶中的轨迹规划和安全关键场景生成问题,通过SSDS解码器和DAPSE方案提升规划性能及生成挑战性测试场景。
为了解决城市环境中精细跨视角定位问题,本文通过构建大规模、多样化的OpenCVL数据集,并开发数据管理框架来过滤和修正姿态注释,从而提高模型性能。
本文通过优化场景表示、GPU执行等方法,解决了光线追踪在实时渲染中速度慢的问题,实现了比现有方法更快的渲染速度。
This work addresses the challenge of accurately modeling highly variable or elongated objects—such as ropes—in instance segmentation, where conventional anchor-based methods often fail to capture complex shapes. To overcome this limitation, the authors propose a novel anchor-free paradigm that leverages neural networks to predict, at each pixel, multi-directional distances to the nearest object boundary. These predictions are aggregated to approximate a signed distance function (SDF), from which a foreground mask is obtained via thresholding. By performing pixel-wise directional distance regression, the method flexibly represents arbitrary object geometries without relying on predefined anchors. Evaluated on the COCO dataset, the approach achieves superior segmentation IoU compared to state-of-the-art methods like YOLACT, demonstrating significantly enhanced adaptability to irregular object shapes.
This work addresses the challenge of achieving scalable and robust ego-vehicle trajectory prediction without reliance on high-definition maps. To this end, we propose an end-to-end system that integrates front-view images, vehicle kinematics, and navigation paths derived from standard-definition (SD) maps. We introduce SD map paths as a semantic prior for trajectory prediction for the first time, and design a dual-hypothesis fusion architecture with a gated classifier to handle challenges such as route corruption or visual ambiguity. Evaluated on 480,000 real-world driving scenarios spanning ten European countries and the United States, our method reduces the average displacement error (ADE) over an 8-second horizon by 16.9% compared to a baseline using only images and kinematics. We also release an open-source toolkit for SD path generation to support community benchmarking.