|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
🦾 🤖 Project CrowdBot |
|
|
|
|
|
|
|
|
 |
|
|
|
|
|
# Problem: Limited Access to High-Quality Training Data |
|
|
1. State-of-the-art foundational robotics models today rely on large datasets of private teleoperation data, accessible only to well-funded organizations, creating a barrier to innovation. |
|
|
<img src="https://cdn-uploads.huggingface.co/production/uploads/67b01cae6ccf462ccafdc4e8/ELOW6hNBPz03P8KDcDI3E.png" width="400"/> |
|
|
<img src="https://cdn-uploads.huggingface.co/production/uploads/67b01cae6ccf462ccafdc4e8/Qb6K1ooyvJCHCkF35Hex3.png" width="400"/> |
|
|
|
|
|
2. The datasets available today are all static and made by universities using robots > $10,000. There is no high-quality dataset that *anyone* can contribute to who doesn't have a professional budget. |
|
|
|
|
|
# Solution: Crowdsourcing an Open Robotics Dataset |
|
|
Project CrowdBot seeks to democratize robotics by building a high-quality, open dataset for training state-of-the-art models. By enabling anyone to access and contribute data, CrowdBot fosters broader advancements in robotics. |
|
|
|
|
|
# How YOU can Contribute |
|
|
1. **Build** your own low-cost robot |
|
|
2. **Collect** labeled teleoperation data, such as folding a size-M t-shirt |
|
|
3. **Upload** your data to CrowdBot's Huggingface page via pull request |
|
|
|
|
|
# How to use the dataset |
