Still coming back to the LeRobot announcement. I'm interested in the gap between watching a manipulation demo and being able to explain why it failed. People who've started exploring this: what was worth learning before the hardware arrived? Not after a perfect six-month syllabus. The first useful thing.
https://blogs.nvidia.com/blog/hugging-face-lerobot-open-source-robotics/
Keeping a genuinely separate test set would be my pick. Otherwise I can keep improving my score without learning whether the setup copes with anything new.
