so RDT train via 3 stage? here is my opinions and other questions :) #111
Description
pretrain---46 open dataset---1 M step-----and then get robotics-diffusion-transformer/rdt-1b
finetune---Mobile‑ALOHA ---6 K traj-----no release
finetune_sim--- ManiSkill ---5 K traj -----and then get robotics-diffusion-transformer/maniskill-model/mp_rank_00_model_states.pt]
is that correct ?
and what is more, on your hugging face , [robotics-diffusion-transformer]/[rdt-sim-eval] is nearly empty.
Are Mobile‑ALOHA and rdt‑ft‑data the same underlying dataset, differing only in format?