[Question] How to Access Ground Truth Environment Geometry in IsaacLab #2353
Replies: 2 comments
-
|
Thank you for posting this. For ground truth labeling you have to add a sensor to the scene, such as a camera, and use Replicator to request ground truth as needed. This doc may be a good place to start. |
Beta Was this translation helpful? Give feedback.
-
|
Hello, @RandomOakForest @eugene0429 |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
How to Access Ground Truth Environment Geometry in IsaacLab
Hi NVIDIA Team,
I'm currently working on a deep learning project using IsaacLab, and I want to generate training data for a scene reconstruction model. The idea is to use sensor data (RGB, depth, LiDAR) from a robot (e.g., ANYmal-C) as input, and train a model to reconstruct the surrounding terrain or environment in 3D.
✅ My Setup:
TerrainImporterCfgto procedurally generate terrains (e.g., rough terrain).❓ My Key Question:
What is the proper or recommended way to directly access the terrain/environment geometry (i.e., full 3D point cloud or mesh) for ground truth labeling?
📌 Specifically:
TerrainImporterCfg, is it expected to be aMeshprim under a known path like/World/ground?Meshtypes, is there a recommended way to convert or sample them for GT data?👀 What I've Tried:
Meshprims.UsdGeom.Mesh(...).GetPointsAttr().Get()and transforming to world coordinates.XformorConetypes, which are not directly sampleable./World/groundis missing).🎯 My Goal:
For every simulation frame, generate a pair:
What’s the best practice for this in IsaacLab?
Thank you so much in advance — any pointers or sample code would be appreciated!
Beta Was this translation helpful? Give feedback.
All reactions