Skip to content

Dataset · nvidia

NVIDIA Robot Motions and Locomanipulation

nvidia/PhysicalAI-Robotics-Locomanipulation-GRAIL

A multimodal dataset with 3D data, images, and videos of robot motions, including stair climbing and descending.

What's inside

The NVIDIA dataset includes 3D assets, images, and video recordings of various motions for robotics (1,000 to 10,000 records). A recent update adds 12,512 sequences of ascending and descending stairs lasting 5 to 12 seconds across 2,000 procedurally generated staircases, alongside tabletop object interaction data.

What you can do with it

  • Training tracking policies for legged and humanoid robots
  • Training locomanipulation models in simulators
  • Testing robot navigation algorithms on stairs and uneven terrain
Size
1K–10K rows
Downloads this month
54.7K
Likes
62

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/PhysicalAI-Robotics-Locomanipulation-GRAIL")