Contributing
Why contribute?
The Oopsie Dataset is a community effort. We are collecting failure data from diverse robots, tasks, and environments.
Our core focus for the first iteration of the project is on single-arm and bi-arm manipulation settings. To ensure that the data is broadly useful across different labs, we are especially interested in data from standardized setups, such as the Franka-arm setup used in the Droid project, or the Aloha arms distributed by Trossen Robotics. However, if you have other systems that you believe will be useful for the community, please submit your data! Failure data on non-standard systems is still very useful for many research projects, such as building reliable failure detection or reward models.
Why should YOU contribute?
First and foremost, because open-data and cross-lab collaboration has proven to be one of the best ways for academics to collaborate on ambitious robotics projects. But we acknowledge that data collection needs some additional incentives.
Specifically, all labs that are involved will get several benefits. First of all, we want to make this data useful for your projects, so we are happy to discuss adaptations that make our tool and data useful for your specific lab use-case. If you help us, you get a say in how the data is processed and what utilities we provide.
Second, we will regularly share pre-release versions of the dataset with all contributors so that they can prototype and test ideas with the data. We are tentatively aiming for an alpha release by the CoRL deadline, until then you can use our data exclusively if you contribute. In addition, having your own robot in a major dataset can make future projects that build on this data much easier to deploy, as you have already provided lab-specific fine-tuning data.
Finally, we are happy to invite all volunteers who collect substantial amounts of data and provide feedback as co-authors on our public release. Volunteering for data gathering is laborious and we acknowledge that by celebrating everybody who contribute significant time and effort to the project.
Who can contribute?
Anyone running robot manipulation experiments!
If you are evaluating policies, you are already generating the data we need. Furthermore we want it to be as easy as possible for labs around the world to contribute. If you find that your workflow is currently not well-supported by the provided software, let us know. We will make sure to help you share your evaluation data.
Finally, please contact us for any suggestions! We want to make this data useful for you and your research.