RoboTurk Pilot#


The first RoboTurk paper released large-scale pilot datasets collected with robosuite v0.3. These datasets consist of over 1000 task demonstrations each on several Sawyer PickPlace and NutAssembly task variants, collected by several human operators. This repository is fully compatible with these datasets.



When working with these datasets, please make sure that you have installed robosuite from source and are on the roboturk_v1 branch.

Download the dataset here (~9 GB download), and unzip the file, resulting in a RoboTurkPilot folder.


First, the dataset must be updated to a format compatible with robomimic. Run the following script (these arguments are provided as an example):

# convert the Can demonstrations, and also create a "fastest_225" filter_key (prior work such as IRIS has trained on this subset)
$ python conversion/ --folder /path/to/RoboTurkPilot/bins-Can --n 225

Then, the dataset must be postprocessed since there are no observations stored. You must run For more information, see this page.


  title={Roboturk: A crowdsourcing platform for robotic skill learning through imitation},
  author={Mandlekar, Ajay and Zhu, Yuke and Garg, Animesh and Booher, Jonathan and Spero, Max and Tung, Albert and Gao, Julian and Emmons, John and Gupta, Anchit and Orbay, Emre and others},
  booktitle={Conference on Robot Learning},