GeorgiaTech/HERCULES
收藏资源简介:
HERCULES是一个多机器人光真实感合成SLAM数据集,专为多机器人SLAM和协作感知研究设计。数据集包含2架无人机和2辆Husky UGV在四个大规模环境(澳大利亚内陆中心路线、澳大利亚内陆周边路线、城市街区和茂密森林)中运行的时间同步多模态传感器流。数据是合成的,通过基于Unreal Engine 5的HERCULES仿真框架生成,扩展了AirSim和Cosys-AirSim作为UE5插件,使用Lumen全局照明和Nanite几何实现光真实感渲染。它提供光真实感图像以及几何、语义和轨迹的完美无噪声真值。所有传感器流共享共同的时间基准(同步采集)。数据集适用于SLAM、LiDAR惯性/视觉惯性里程计、多机器人/协作感知、深度估计和语义分割等任务。每个机器人配备相同的同步记录传感器套件,包括RGB相机(752×480,20 Hz)、立体相机(基线0.11米)、深度图(浮点32位,米)、语义分割(320个类别)、LiDAR(16通道,20 Hz)和IMU(最高500 Hz)。数据集总大小约1.1 TB,轨迹长度设计为每序列359-945米,包含机器内和机器间闭环。数据以.tar.zst存档格式提供,需下载后解压使用。
HERCULES is a photorealistic synthetic multi-robot SLAM dataset dedicated to multi-robot SLAM and collaborative perception research. The dataset contains time-synchronized multimodal sensor streams from 2 drones and 2 Husky UGVs operating in four large-scale environments: Central Australian Inland Route, Peripheral Australian Inland Route, Urban Block, and Dense Forest. The data is synthetic, generated via the HERCULES simulation framework based on Unreal Engine 5, which extends AirSim and Cosys-AirSim as UE5 plugins, and adopts Lumen global illumination and Nanite geometry to achieve photorealistic rendering. It provides photorealistic images alongside perfect noise-free ground truths for geometry, semantics and trajectories. All sensor streams share a unified time base (synchronized acquisition). This dataset is applicable to tasks including SLAM, LiDAR inertial/visual-inertial odometry, multi-robot/collaborative perception, depth estimation and semantic segmentation. Each robot is equipped with an identical synchronously recorded sensor suite, comprising an RGB camera (752×480, 20 Hz), stereo camera (baseline 0.11 m), depth map (32-bit float, in meters), semantic segmentation (320 categories), 16-channel LiDAR (20 Hz) and IMU (up to 500 Hz). The total size of the dataset is approximately 1.1 TB, with the trajectory length designed to be 359–945 meters per sequence, containing both intra-robot and inter-robot loop closures. The data is provided in .tar.zst archive format and requires decompression after downloading for usage.




