Hi everyone,
I am trying to integrate lerobot with ROS2 humble to evaluate VLA models:
For each time instant (frame) i have collected the position of the joints, gripper and the photo from the camera .
I have some problems converting this "raw" dataset , collections of image,joints and gripper states to lerobot format so then i can train the models. Then i will deploy a node containing the trained model in order to perform inference. I am still working on the simulator (Gazebo ) and then i will switch to the real robot FAIRINO FR10, collecting the dataset BY move it manually.
The raw dataset is composed of a collection, for each episode, of jpg frames images and csv file containing the states.
Do you have any suggestions for how to perform this conversion? has this strategy some sense or I should use some other tools?
Thank you