Dataset · allenai
3D scenes, objects, and grasps for robot simulation
allenai/molmospaces
Large multimodal dataset from AllenAI featuring 3D assets, scenes, grasps, and benchmarks for robotics.
What's inside
The repository contains between 1 and 10 million records in tabular and text formats, including descriptions of objects, robots, scenes, grasp points, and benchmarks. The dataset includes scene versions for MuJoCo simulators (including iTHOR, ProcTHOR, and Holodeck with occupancy maps) and NVIDIA Isaac USD-compatible objects. The data is updated regularly and provided with download scripts.
What you can do with it
- Training robots for navigation and object manipulation in MuJoCo and Isaac simulators
- Evaluating multimodal models and VLA agents on spatial understanding
- Generating synthetic environments for reinforcement learning (RL)
- Size
- 1M–10M rows
- Downloads this month
- 42.6K
- Likes
- 53
How to load it
from datasets import load_dataset
ds = load_dataset("allenai/molmospaces")