GELLO Teleop

To collect demonstrations for robot policy learning, we built two GELLO UR replica’s and created a dual-arm teleop setup.

You can see an example of the fine-grained manipulation tasks that can be demonstrated with such a teleop setup in the video below:

Using Behavior Learning algorithms, such as ACT, we can train our robot to execute fine-grained manipulation tasks. Below is an example of how the robot is able to autonomously open our (instrumented) Nespresso machine;

The codebase is available here. A big thank you to Philipp Wu for open-sourcing GELLO.