datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
quickdraw-calvin-d-d
QuickDraw CALVIN D-to-D
The official six-hour CALVIN environment-D play corpus, repackaged as LeRobot v2.1 for QuickDraw.
Splits are made only at native rollout boundaries. train and validation are a deterministic
90/10 partition of upstream training rollouts; eval_offline preserves the official validation
recordings. CALVIN's standard online D-to-D benchmark remains simulator evaluation.
Each frame contains the unchanged 39-dimensional robot-plus-scene state, unchanged… See the full description on the dataset page: https://huggingface.co/datasets/madang6/quickdraw-calvin-d-d.quickdraw-robocasa-scene4-4h
QuickDraw RoboCasa Scene 4 Eight-Task Composite (4h)
This dataset contains 4.008 hours of RoboCasa expert demonstrations in one
exact kitchen environment: layout 4, style 4. It is packaged as a LeRobot v2.1
training split.
What is in every episode?
Every episode contains the combined object inventories of the same eight
selected tasks:
DeliverStraw
GatherTableware
GetToastedBread
HeatKebabSandwich
LoadDishwasher
MakeIceLemonade
PrepareCoffee
WashFruitColander… See the full description on the dataset page: https://huggingface.co/datasets/madang6/quickdraw-robocasa-scene4-4h.