All Robots In One

A unified embodied-data format standard from Peng Cheng Laboratory and partners, and a roughly 3-million-item dataset built to that standard.

ARIO was released in August 2024 by the Institute of Multi-Agent and Embodied Intelligence at Peng Cheng Laboratory, together with Southern University of Science and Technology, Sun Yat-sen University, and others; it is both a data-format standard and a large dataset assembled under it. The authors argue that aggregating datasets such as Open X-Embodiment still leave formats inconsistent and modalities incomplete, so ARIO specifies: control data from robots of different forms recorded in one unified format, sensors of differing frequency aligned by timestamp, organization by a three-level “series–task–episode” hierarchy, and support for five modalities — image, 3D, sound, text, and touch. The dataset contains about 3 million episodes, 258 series, and more than 320,000 tasks, drawn from three sources: 3,662 episodes from a self-built real-robot platform; about 700,000 from simulators such as Habitat and MuJoCo; and roughly 2.33 million converted from existing open datasets.

ExampleARIO's real-robot portion used the AgileX Cobot Magic dual-arm platform to collect more than 30 kinds of manipulation tasks in real household scenes.

Also called

ARIO, ARIO Data Standard

Related

Cross-Embodiment Data · Open X-Embodiment · Multimodal Data · Heterogeneous Data · AgileX Cobot Magic · RLDS (Reinforcement Learning Datasets)

Sources

All Robots in One: A New Standard and Unified Dataset for Versatile, General-Purpose Embodied Agents (arXiv 2408.10899)
ARIO 项目主页 (Chinese)

As of

2024-08