theairlabcmu/TartanGround

TartanGround是一个大规模多模态地面机器人感知和导航数据集，包含63个多样化模拟环境中的878个轨迹和超过144万样本。支持多种机器人平台和传感器模态，包括RGB立体相机对、深度图、语义分割、光流、立体视差、LiDAR点云、IMU数据、真实位姿和语义占用地图等。适用于语义占用预测、视觉SLAM、神经场景表示、俯视图预测和导航等多种机器人感知和导航任务。

TartanGround is a large-scale, multi-modal dataset for ground robot perception and navigation, containing 878 trajectories across 63 diverse simulation environments and over 1.44 million samples. It supports various robot platforms and sensor modalities, including RGB stereo camera pairs, depth maps, semantic segmentation, optical flow, stereo disparity, LiDAR point clouds, IMU data, ground truth poses, and semantic occupancy maps. It is suitable for a wide range of robotic perception and navigation tasks such as semantic occupancy prediction, visual SLAM, neural scene representation, birds-eye-view prediction, and navigation.