Project Page | Paper | GitHub
RoboTwin 2.0 aloha-agilex demonstrations used to train
FlowWAM, a dual-stream (RGB +
optical-flow) world model with an IDM action expert. This is the raw
dataset (before latent pre-encoding).
50 tasks, dual-arm aloha-agilex embodiment.
Two variants per task:
aloha-agilex_clean_50 — 50 episodes/task (2,500 total).
aloha-agilex_randomized_500 — 500 episodes/task (25,000 total).
Each (task, variant) is a tar.gz archive… See the full description on the dataset page:
https://huggingface.co/datasets/YixiangChen/FlowWAM_RoboTwin.
