TartanGround: A Large Scale Dataset for Ground Robot Perception and Navigation Dataset Description TartanGround is a large scale, multi modal dataset designed to advance the perception and autonomy of ground robots operating in diverse environments. Collected across 63 photorealistic simulation environments, it provides comprehensive data streams for various robotic tasks. Key Features Environments : 63 diverse simulation environments categorized into: Indoor Nature Rural Urban Industrial/Infrastructure Historical/Thematic Trajectories : 878 trajectories captured across the environments. Samples : Over 1.44 million samples. Robot Platforms : Omnidirectional ( P0000 , P0001 , ...) Differential Drive ( P1000 , P1001 , ...) Quadrupedal ( P2000 , P2001 , ...) Sensor Modalities : RGB Stereo Camera Pairs (front, back, left, right, top, bottom) Depth Maps Semantic Segmentation Optical Flow Stereo Disparity LiDAR Point Clouds IMU Data Ground Truth Poses (6 DOF) Semantic Occupancy Maps (3D voxel grids) Proprioceptive Data (for quadruped trajectories) Applications TartanGround supports a wide range of robotic perception and navigation tasks, including: Semantic Occupancy Prediction Open Voca…
Runs entirely in your browser via DuckDB-Wasm — this dataset's real data file is loaded once, then queried locally. Nothing is sent to a server.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy