← ML Research Wiki / 2403.12945

DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset

Alexander Khazatsky, Karl Pertsch [email protected], Suraj Nair, Ashwin Balakrishna, Sudeep Dasari, Siddharth Karamcheti, Soroush Nasiriany, Mohan Kumar Srirama, Lawrence Yunliang Chen, PeterKirsty Ellis, Peter David Fagan, Joey Hejna, Masha Itkina, Marion Lepert, Jason Ma, Patrick Tree Miller, Jimmy Wu, Suneel Belkhale, Shivin Dass, Huy Ha, Abraham Lee, Youngwoon Lee, Arhan Jain, Marius Memmel, Sungjae Park, Ilija Radosavovic, Kaiyuan Wang, Albert Zhan, Kevin Black, Cheng Chi, Kyle Hatch, Shan Lin, Jingpei Lu, Abdul Rehman, Pannag R Sanketi, Archit Sharma, Cody Simpson, Quan Vuong, Homer Walke, Blake Wulfe, Ted Xiao, Jonathan Yang, Arefeh Yavary, Tony Z Zhao, Christopher Agia, Rohan Baijal, Mateo Guaman Castro, Daphne Chen, Qiuyu Chen, Trinity Chung, Jaimyn Drake, Ethan Paul Foster, Jensen Gao, David Antonio Herrera, Minho Heo, Kyle Hsu, Jiaheng Hu, Donovon Jackson, Charlotte Le, Yunshuang Li, Kevin Lin, Roy Lin, Zehan Ma, Abhiram Maddukuri, Suvir Mirchandani, Daniel Morton, Tony Nguyen, Abby O'neill, Rosario Scalise, Derick Seale, Victor Son, Stephen Tian, Andrew Wang, Yilin Wu, Annie Xie, Jingyun Yang, Patrick Yin, Yunchu Zhang, Osbert Bastani, Glen Berseth, Jeannette Bohg, Ken Goldberg, Abhinav Gupta, Abhishek Gupta, Dinesh Jayaraman, Joseph J Lim, Jitendra Malik, Roberto Martín-Martín, Subramanian Ramamoorthy, Dorsa Sadigh, Shuran Song, Jiajun Wu, Yuke Zhu, Thomas Kollar, Sergey Levine, Chelsea Finn, Mohan Kumar Srirama, David Fagan, Jimmy Wu, Jonathan Yang, Jimmy Wu, Jimmy Wu, Jingyun Yang, Abhinav Gupta, Abhishek Gupta, Roberto Martín- Martín, Jiajun Wu (2024)

Paper Information
arXiv ID
Venue
Robotics: Science and Systems
Domain
robotics
Code
Available
Reproducibility
7/10

Abstract

BathroomDROIDDistributed Robot Interaction Dataset 564 Scenes 76k Episodes 52 Buildings 86 Tasks / Verbs 13 Institutions Dining Room Bedroom Laboratory Laundry Room Office Kitchen

Summary

This paper presents DROID (Distributed Robot Interaction Dataset), a large-scale robot manipulation dataset which includes 76k demonstration trajectories across 564 scenes and 86 tasks, aimed at improving robot manipulation policies through enhanced training data diversity. The dataset consists of 350 hours of interaction data collected over 12 months by 18 research labs worldwide, utilizing three synchronized RGB camera streams, camera calibration, depth information, and natural language instructions. DROID seeks to address the limitations of existing datasets, predominantly collected in controlled settings, by ensuring a wide range of tasks and environments for better policy generalization. The authors emphasize the substantial improvements in policy performance, robustness, and generalization achieved through training with DROID, demonstrating its superior capabilities over state-of-the-art existing datasets. The paper also delineates the data collection setup, methodology, and experiments, showing how the diversity of the dataset leads to improved policy outcomes in various real-world scenarios.

Methods

This paper employs the following methods:

  • Robot Manipulation
  • Data Collection Protocol
  • Policy Learning
  • Diffusion Policies

Models Used

  • Franka Panda

Datasets

The following datasets were used in this research:

  • DROID

Evaluation Metrics

  • None specified

Results

  • DROID demonstrates significant improvements in policy performance (20% increase in success rates on average) and robustness compared to existing datasets.
  • The open-sourcing of DROID is expected to catalyze further research in robot manipulation policies.

Limitations

The authors identified the following limitations:

  • Creating diverse and large robot manipulation datasets is logistically challenging and requires substantial investment.
  • Even existing methods exhibit limitations in generalization across unfamiliar environments.

Technical Requirements

  • Number of GPUs: None specified
  • GPU Type: None specified

Keywords

robot manipulation large-scale dataset generalization policy learning in-the-wild data

Papers Using Similar Methods

External Resources