Datasets:
How was this dataset generated?
the description says generated with lerobot. However, I can't find a script to generate this dataset. Was it ported from here? https://github.com/real-stanford/diffusion_policy?tab=readme-ov-file if so, there is no script showing how this was done?
It looks like it was ported from https://diffusion-policy.cs.columbia.edu/data/training/. But the problem I have found with the dataset on Hugging Face (lerobot/pusht) is that while it has the environment state for the pusher position (motor coordinates x, y), it is missing the environment state for the T position (x, y position and rotation), which is critical for replaying expert episodes for training! This is in the original dataset. Or perhaps I just didn't know where to look, so I ended up downloading the training dataset files from the Columbia website and using those.