jcoholichMIT

MANGO Demonstrations

A dataset of 150 human-teleoperated, 150 synthetic MANGO, and 150 sim demonstrations for four manipulation tasks (stack blocks, close laptop, stack cups, pick coke), including image translation training data with paired sim and real images.

Downloads594
Episodes450 total (150 per source type × 3 sources)

Why This Matters for Physical AI

This dataset demonstrates the use of synthetic data generation and domain adaptation through image translation to improve imitation learning for robotic manipulation tasks across sim-to-real transfer.

Technical Profile

Modalities
rgbproprioception
Action Space
end_effector_delta
Environment
simulationlab
Task Types
manipulationpick_and_placestacking
Episodes
450 total (150 per source type × 3 sources)
Data Format
HDF5
Annotation Types
segmentation
License
MIT
Part of the MANGO family

Community Signals

Top 25% by downloads
HuggingFace Discussions1

Access

Need custom rgb data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets