DexHub and DART: Toward Internet Scale Robot Data Collection

Younghyo Park, Jagdeep Singh Bhatia, Lars Lien Ankile, Pulkit Agrawal

Keywords: Robot Teleoperation, Augmented Reality, Cloud Robotics

TL;DR: DART is an AR teleoperation platform using cloud simulation for widely available robot data collection. It improves throughput, reduces fatigue, and supports Sim2Real transfer.

Abstract: The field of robotics has long grappled with a critical challenge: the scarcity of diverse, high-quality data that can be used to train a generalist robot policy. While real-world data collection efforts exist, requirements for robot hardware, physical environment setups, and frequent resets significantly impede the scalability needed for modern learning frameworks. To address these limitations, this paper introduces DART, a novel teleoperation platform that reimagines robotic data collection by leveraging cloud-based simulation and augmented reality (AR). Our user studies highlight that DART enables higher data collection throughput and lower physical fatigue compared to real-world teleoperation frameworks. In addition, our policy training experiments using DART-collected datasets demonstrate successful Sim2Real transfer with robust trained behaviors. Most importantly, all data collected through DART is automatically stored in our cloud-hosted database, DexHub, and publicly available to anyone.

Previous Publication: No

Submission Number: 36