datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
gr00t-n15-robocasa-gr1-eval
GR00T N1.5 on RoboCasa GR-1 Tabletop — Evaluation Trajectories
Per-simulator-step recordings of 1,200 evaluation episodes (24 tasks × 50 episodes) of
NVIDIA's GR00T N1.5 vision-language-action model on the
RoboCasa GR-1 Tabletop Tasks benchmark.
Each episode stores every low-level transition — ego-view frames, robot state, executed actions,
rewards, full MuJoCo state, and 3D poses of every object and scene body — so that rollouts can be
re-analyzed or re-rendered without… See the full description on the dataset page: https://huggingface.co/datasets/junhee1998/gr00t-n15-robocasa-gr1-eval.robocasa_pretrain_human300_v4_annotated5This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 20,
"features": {
"observation.images.robot0_agentview_left": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"video_info": {… See the full description on the dataset page: https://huggingface.co/datasets/pepijn223/robocasa_pretrain_human300_v4_annotated5.robocasa24-cache-batch1-base50
RoboCasa Atomic-24 Batch 1 Cache
Final Geometry and Semantic cache artifacts for the first 50 (base50) episodes
of every RoboCasa Atomic-24 task.
Release summary
Tasks: 24
Episodes per task: 50
Total policy frames: 332,859
Geometry: 24/24 PASS, float32, 2,048 dimensions, zero NaN/Inf
Semantic: 24/24 PASS
Dataset revision: 7236e704a04ebe477cc06d0a06ad540cd968fa5d
Cache tools commit: e61d47c97bb81ed37c5fc7c72454d0144fd0f740
The source RGB/mask HDF5 payload, Motion… See the full description on the dataset page: https://huggingface.co/datasets/xc1204/robocasa24-cache-batch1-base50.robocasa_pretrain_human300_v4This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 20,
"features": {
"observation.images.robot0_agentview_left": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"video_info": {… See the full description on the dataset page: https://huggingface.co/datasets/pepijn223/robocasa_pretrain_human300_v4.deva_robocasa
DeVA — RoboCasa (processed)
Processed RoboCasa data for DeVA, a robot
video-world model with affordance + depth physical guidance. Derived from
robocasa/robocasa —
please cite the original benchmark as well.
Episodes
1,199
Frames
316,995
Video
448 × 448, 16 fps (3 views on a 2×2 grid)
Views
agentview_left, agentview_right, eye_in_hand
Action dim
7 (absolute)
Proprio state
not included
Size
56 GB
Download
hf download… See the full description on the dataset page: https://huggingface.co/datasets/mengqz9/deva_robocasa.robocasa_mg_gr00t_100failbench-robocasa-v2
FailBench RoboCasa v2 — contact-prediction dataset
Labeled robot-failure trials built on RoboCasa kitchen demos (PandaMobile / Franka).
Each trial injects a hardware failure partway through a teleop/MimicGen demo, then records the
contacts the failure causes during a 1-second settle. The supervised target is a
240×320 force-weighted contact heatmap in the agentview camera — the model learns to predict
where a failure at a given pre-failure configuration will drive the… See the full description on the dataset page: https://huggingface.co/datasets/aaronngx/failbench-robocasa-v2.robocasa_ptrobocasa-rldx-demospeedup-slow2-fast4
RoboCasa DemoSpeedup: RLDX-1 entropy, slow2 / fast4
Derived from kimtaey/robocasa_mg_gr00t_300 using prehj/rldx1-robocasa-img-gb64-60k-baseline.
Read FORMAT.md before training. Targets are explicit per-observation action chunks; the original action column is retained for provenance and is not the speedup target.
7200 episodes; 2073457 source observations; 2066257 usable anchors.
No model was trained for this release. See meta/demospeedup.json and per-episode entropy files.
robocasa_lerobotThis dataset was created using LeRobot.
Dataset Structure
Originally comes from https://robocasa.ai/docs/build/html/datasets/datasets_overview.html
the script to build this dataset : https://github.com/yanan1116/flow_matching/blob/master/examples/push_lerobot.py#L171
this dataset includes
Atomic: such as pick and place, open door, open drawer, turn faucet, navigate to kitchen, turn microwave ...
Composite : such as waffle reheat, transfer pan, make ice lemon, select a… See the full description on the dataset page: https://huggingface.co/datasets/yananchen/robocasa_lerobot.robocasa-openfridge-astra-low-eval
OpenFridge: Kinex and Codex evaluation
Watch the evaluation website · Dataset repository · Browse all files · Website source
Ten recorded OpenFridge episodes, evaluated on September 18, 2026 with GPT-6 Astra / low, standard service tier, fast mode disabled. Five fresh conversations per harness; seeds 0–4; existing cross-episode tool/skill/memo sharing retained.
Harness
Native successes
Outcomes E1–E5
Harness execution
Kinex
1/5
fail, success, fail, fail, fail
Four… See the full description on the dataset page: https://huggingface.co/datasets/RLE-Bench/robocasa-openfridge-astra-low-eval.arp-decision131-robocasa-libero-20
Decision-131 complete RoboCasa + LIBERO-PRO subset
This repository contains the complete data package requested for the 20 task
identities selected in ARP decision 131.
RoboCasa human demonstrations
10 target tasks
5,101 episodes
2,680,229 frames at 20 Hz
Actions, robot state, episode/task metadata, and three synchronized 256×256
H.264 camera streams
176 MP4 files and 8 Parquet files
The episodes were selected from
lerobot/robocasa_target_human_unified
at… See the full description on the dataset page: https://huggingface.co/datasets/mojimoji61/arp-decision131-robocasa-libero-20.robocasa-100demos-5tasks-for-dgrobocasa_20260430T030150Z_full_run_prepare_cocktail_station ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_prepare_cocktail_station.robocasa_mg_gr00t_300_transferrobocasa_20260430T030150Z_full_run_line_up_condiments ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_line_up_condiments.robocasa_20260430T030150Z_full_run_beverage_organization ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_beverage_organization.robocasa_20260430T030150Z_full_run_set_bowls_for_soup ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_set_bowls_for_soup.robocasa_20260430T030150Z_full_run_spicy_marinade ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_spicy_marinade.robocasa_20260430T030150Z_full_run_setup_wine_glasses ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_setup_wine_glasses.robocasa-ratio-labels-contact
배속 라벨 (ratio labels)
로봇 정책이 내는 액션 청크를 시점마다 몇 배속으로 압축해도 되는지 적은
라벨입니다. robocasa 벤치마크용입니다.
붙이는 법
git clone https://github.com/rakybond007/quantization_label_gr00t
python quantization_label_gr00t/vlm_gate/scripts/apply_ratio_labels.py \
--labels labels/robocasa.parquet \
--dataset /내/경로/lerobot/robocasa \
--out /내/경로/robocasa_with_ratio
pandas 와 pyarrow 만 있으면 됩니다. GPU 도 VLM 도 필요 없습니다.
--dry-run 을 먼저 주면 아무것도 쓰지 않고 붙는 비율만 보여 줍니다.
붙는 열… See the full description on the dataset page: https://huggingface.co/datasets/prehj/robocasa-ratio-labels-contact.robocasa_20260430T030150Z_full_run_setup_bowls ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_setup_bowls.robocasa_20260430T030150Z_full_run_veggie_dip_prep ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_veggie_dip_prep.robocasa-ratio-labels
배속 라벨 (ratio labels)
로봇 정책이 내는 액션 청크를 시점마다 몇 배속으로 압축해도 되는지 적은
라벨입니다. robocasa 벤치마크용입니다.
붙이는 법
git clone https://github.com/rakybond007/quantization_label_gr00t
python quantization_label_gr00t/vlm_gate/scripts/apply_ratio_labels.py \
--labels labels/robocasa.parquet \
--dataset /내/경로/lerobot/robocasa \
--out /내/경로/robocasa_with_ratio
pandas 와 pyarrow 만 있으면 됩니다. GPU 도 VLM 도 필요 없습니다.
--dry-run 을 먼저 주면 아무것도 쓰지 않고 붙는 비율만 보여 줍니다.
붙는 열… See the full description on the dataset page: https://huggingface.co/datasets/prehj/robocasa-ratio-labels.robocasa_20260430T030150Z_full_run_distribute_chicken ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_distribute_chicken.robocasa-trajectories-1k
RoboCasa Trajectories Single
This dataset contains flat RoboCasa step rows with sidecar episode JSON.
Structure
Each row is one tool step.
Episode-level JSON is not duplicated into parquet. Instead, each row
carries repo-relative references:
adapted_trajectory_path
original_trajectory_path
execution_metadata_path
Image columns stay inline and viewable in the dataset table:
room_view
top_view
map
agentview_center
agentview_left
agentview_right
wrist
Summary… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa-trajectories-1k.robocasa-100demos-5tasks-for-dg-inferencerobocasa_20260430T030150Z_full_run_tong_buffet_setup ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_tong_buffet_setup.robocasa_20260430T030150Z_full_run_bowl_and_cup ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_bowl_and_cup.robocasa_20260430T030150Z_full_run_garnish_cupcake ---
pretty_name: RoboCasa Trajectories Single
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# RoboCasa Trajectories Single
This dataset contains one row per RoboCasa trajectory / episode.
## Structure
Each row is one trajectory / episode.
Episode-level JSON is stored inline:
adapted_trajectory
original_trajectory
execution_metadata
Step-level data is stored in aligned sequence columns:… See the full description on the dataset page: https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_garnish_cupcake.