nuPlan-v1.1

nuPlan是世界上第一个自动驾驶的大型规划基准。尽管越来越多的基于ML的运动计划人员，但缺乏已建立的数据集，仿真框架和指标限制了该领域的进展。用于自动驾驶车辆运动预测的现有基准 (Argoverse，Lyft，Waymo) 集中在其他代理的短期运动预测上，而不是自我车辆的长期计划。这导致以前的工作使用具有L2-based指标的开环评估，这些指标不适合公平评估长期计划。我们的基准测试通过提供培训框架来开发基于机器学习的计划者，轻量级闭环模拟器，特定于运动计划的指标以及可视化结果的交互式工具，克服了这些限制。

nuPlan is the world's first large-scale planning benchmark for autonomous driving. While ML-based motion planners have proliferated, the absence of established datasets, simulation frameworks, and standardized metrics has constrained progress in the field. Existing benchmarks for autonomous vehicle motion prediction—including Argoverse, Lyft, and Waymo—focus on short-term motion forecasting of surrounding traffic agents rather than long-term planning of the ego vehicle. Consequently, prior studies have resorted to open-loop evaluations using L2-based metrics, which are ill-suited for the fair assessment of long-term planning tasks. Our benchmark addresses these limitations by offering a training framework for developing machine learning-based planners, a lightweight closed-loop simulator, motion planning-specific evaluation metrics, and an interactive tool for result visualization.

搜集汇总

背景与挑战

背景概述

nuPlan-v1.1是世界上第一个自动驾驶大型规划基准数据集，专注于长期运动规划而非短期预测，以克服现有基准的局限性。它提供了机器学习规划者的培训框架、轻量级闭环模拟器、专用指标和可视化工具，由Motional于2021年6月发布，数据量达1114.3GB，采用Apache 2.0许可证。