Hugging Face навчила робота складати футболки за допомогою людей-телеоператорів

Currently, we're doing an experiment with folding t-shirts with robots, fully autonomous. So, we're doing every little step from collecting data to training the models, doing evaluations, and also trying to learn and democratize all the little steps and details that are needed to do this. So, you can now put different kinds of t-shirts, different sizes, and different colors, etc. in front of the the robot initially in a flattened state, and it will fold it nicely and put it on the side. And the robot is also being able to fold it when you just throw at it, and the robot will unfold it and then fold your t-shirt for you. With this, [music] we can also train the robot to do other types of clothes folding using all the technology that we provide. So, this is a project we kicked last [music] year with the objective of going on an adventure and try to to train a robot to perform a complex task. We started from scratch. We picked [music] a robot, and then we picked sensors, and then we tackled the tricky part, that is data collection. But when you're tackling a task such as shirt folding, you need much more manpower. So, we decided to hire teleoperators. There were roughly 10 teleoperators working with us on folding t-shirts. So, they were working by pairs, [music] taking shifts, and trying to fold as many t-shirts as they could. And then using this data and using these demonstrations, we learn in a a model to reproduce basically these folds [music] to eventually learn itself on how to fold t-shirts. So, this is really the first stepping stone where we want to show that you can automate a quite complex task fully end-to-end with all the open-source tools that are out there.