Source: https://actionable.ai/dataset-guide.html

DATASET SELECTION

# Choosing an embodied AI dataset

Compare the signals your model needs with the sample specifications in the Actionable AI catalogue.

## What data does Actionable AI provide?

Actionable AI offers eight collections across egocentric video, bimanual UMI demonstrations and real-robot data. These support evaluation for visual understanding, imitation learning, vision-language-action research and long-horizon task learning.

## Which collections include robot actions?

Bimanual UMI pairs visual observations with gripper pose and actions. G1/G2, Galbot and long-horizon robot collections include robot observations and action or state information. Egocentric video alone should not be assumed to include robot action labels.

## How should I compare datasets?

Start with the target task, required camera views, action or state signals, timing, file format and annotation needs. Compare the sample specifications on each dataset page and confirm recording-level coverage with the team before choosing a collection.

## Are the listed specifications guaranteed for every recording?

The detailed specifications describe samples. Camera configuration, resolution, frame rate, depth and audio can vary by collection and recording. Inventory hour labels are rounded; confirm current availability and the exact delivery scope when enquiring.

## Can I download the data or assume a licence?

Public pages and the catalogue feed describe the collections. Sample library access requires an approved account; licensing and delivery terms are agreed for each enquiry. A public description does not grant a licence to the underlying recordings.

## How do I request a sample or a custom collection?

Choose a dataset and use its enquiry link, or contact info@actionable.ai. Share your target tasks, required signals, environment and timeline. The team will discuss sample access, fit and a delivery scope, including custom collection where appropriate.

## Explore the collections

- [Stereo Ego · 1080p](https://actionable.ai/datasets/E01.html) — Human perspective, captured in stereo with audio and motion sensing.
- [Stereo Ego · Wide-view](https://actionable.ai/datasets/E02.html) — A wider perspective on complete manufacturing and craft workflows.
- [Monocular Ego · Android](https://actionable.ai/datasets/E03.html) — Everyday skills and workplace tasks, seen from a human point of view.
- [Monocular Ego · iPhone](https://actionable.ai/datasets/E04.html) — Continuous first-person recordings that follow activities from start to finish.
- [Bimanual UMI](https://actionable.ai/datasets/U01.html) — Connect human demonstrations to precise gripper movement and action.
- [Real Robot · G1 / G2](https://actionable.ai/datasets/R01.html) — Whole-body and dual-arm experience from robots in real environments.
- [Galbot Robot Data](https://actionable.ai/datasets/R02.html) — Multi-view robot demonstrations across household manipulation tasks.
- [Long-Horizon Robot Data](https://actionable.ai/datasets/R03.html) — Multi-step, bimanual demonstrations for learning sequences of action.

[Discuss your dataset requirements ↗](https://actionable.ai/#contact)