InternData-A1
Shanghai AI LaboratoryInternData-A1 is a simulated robot manipulation dataset from Shanghai AI Laboratory: over 630k trajectories and 7,433 hours across four single- and dual-arm embodiments (Franka, AgileX Split ALOHA, ARX Lift-2 and an AgiBot robot), 70 tasks and 227 scenes, generated by a physics-validated motion-planning pipeline and published in LeRobot format for pretraining vision-language-action policies.
Scoped to InternData-A1. The InternRobotics organization also publishes InternData-M1 (a separate simulated dataset) and InternData-N1 (navigation), which are distinct releases and not part of this row. The card calls the dataset hybrid synthetic-real, but only the simulation data is released; the real-world post-training data is listed as to be released. The Hub card metadata carries no license field; the terms are in the gate text.
Openness
2 high confidence- license
- cc-by-nc-sa-4.0(stated in the card body and the gate prompt
- access
- auto(automatic Hub gate collecting name, email, affiliation, phone and job title)
- dataset_card
- present
Downloadable by anyone who fills in the automatic gate, but only for non-commercial use: the card and the gate both put all data and code under CC BY-NC-SA 4.0.
- https://huggingface.co/api/datasets/InternRobotics/InternData-A1?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=lastModified&expand[]=gated&expand[]=downloadsAllTime recorded 2026-09-27
Hub API for InternData-A1: gated auto; gate prompt "InternData-A1 COMMUNITY LICENSE AGREEMENT ... All the data and code within this repo are under CC BY-NC-SA 4.0", with fields for name, email, country, affiliation, phone and job title; cardData has no license field.
- https://huggingface.co/datasets/InternRobotics/InternData-A1 recorded 2026-09-27
Dataset card: "All the data and code within this repo are under CC BY-NC-SA 4.0"; release list "[2025.11] Released: 632k simulation pretraining data", "[2026.01] Released: 589k simulation pretraining data with full annotations", "To be released: real-world post-training data".
Adoption
4 medium confidenceHugging Face downloads of the dataset; for a multi-terabyte set these count file fetches, not training runs.
- https://huggingface.co/api/datasets/InternRobotics/InternData-A1?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=lastModified&expand[]=gated&expand[]=downloadsAllTime recorded 2026-09-27
Hub API for InternData-A1: 113,672 downloads in the trailing 30 days.
Capability
4 medium confidenceInternData-A1 is hundreds of thousands of simulated trajectories across four robot bodies, the same scale and spread as AgiBot World. It is all simulation, and four embodiments is far from the twenty-plus pooled in Open X-Embodiment.
- https://arxiv.org/abs/2511.16651 recorded 2026-09-27
Abstract: "Our synthetic dataset, InternData-A1, contains over 630k trajectories and 7,433 hours across 4 embodiments".
- https://huggingface.co/datasets/InternRobotics/InternData-A1 recorded 2026-09-27
Dataset card: "over 630k trajectories and 7,433 hours across 4 embodiments, 18 skills, 70 tasks, and 227 scenes"; "ARX Lift-2, AgileX Split Aloha, A2D, Franka".
- https://internrobotics.github.io/interndata-a1.github.io/ recorded 2026-09-27
Project page: "4 single or dual-arm embodiments, 70 diverse tasks ... and 227 rooms"; "630k episodes, 401.4M frames and 7433.9 hours".
Verified 2026-09-27