china-sae-robotics/gr1_arena_sequential_task_replay

GR1 Arena — Ranch Bottle Into Fridge (new camera pose, ego + wrist, replay)

LeRobot-format teleoperation/replay dataset for the GR1 humanoid performing the put_item_in_fridge_and_close_door task in Isaac Lab Arena.

Task: Place the ranch dressing bottle on the top shelf of the fridge, and close the fridge door. (object: ranch_dressing_hope_robolab)

What this dataset is

This is a re-rendered / replayed version of the official NVIDIA Arena dataset. The source trajectories were taken from:

hf download \
  nvidia/Arena-GR1-Manipulation-PlaceItemCloseDoor-Task \
  --include "ranch_bottle_into_fridge/ranch_bottle_into_fridge_generated_100/lerobot/*" \
  --repo-type dataset \
  --revision arena_v0.2_lab_v3.0 \
  --local-dir "$_tmp"

On top of the official data we updated the head (ego) and wrist camera poses and replayed the recorded actions in Isaac Lab Arena (Lab 3.0) to regenerate the camera observations. The result:

• Two camera streams: observation.images.ego_view (head) and observation.images.wrist_view (wrist) — both at the new poses.
• Actions / joint states are the replayed source trajectories.

This makes the dataset suitable for training VLA policies (e.g. GR00T N1.6) that consume both an ego and a wrist camera.

Dataset stats

Field	Value
Format	LeRobot v2.1
Robot	gr1
Episodes	70
Frames	22,873
Tasks	1
FPS	50
Videos	140 (70 ego + 70 wrist)

Features

Key	Type	Shape	Notes
observation.images.ego_view	video	512×512×3	head camera (h264)
observation.images.wrist_view	video	512×512×3	wrist camera (h264)
observation.state	float32	(26,)	14 arm + 12 hand joints
action	float32	(26,)	14 arm + 12 hand joints
timestamp	float64	(1,)
next.reward	float64	(1,)
next.done	bool	(1,)
annotation.human.action.task_description	int64	(1,)	task index

Joint layout (both observation.state and action): left_arm[0:7], right_arm[7:14], left_hand[14:20], right_hand[20:26] (see meta/modality.json).

Layout

.
├── data/chunk-000/episode_{000000..000069}.parquet
├── videos/chunk-000/observation.images.ego_view/episode_*.mp4
├── videos/chunk-000/observation.images.wrist_view/episode_*.mp4
└── meta/
    ├── info.json          # LeRobot dataset info (features, fps, paths)
    ├── episodes.jsonl      # per-episode task + length
    ├── tasks.jsonl         # task index -> language instruction
    ├── modality.json       # GR00T modality mapping (state/action/video split)
    ├── stats.json          # per-feature statistics
    └── relative_stats.json # relative (delta) statistics

Loading

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

ds = LeRobotDataset("china-sae-robotics/gr1_arena_sequential_task_replay")
print(ds)
sample = ds[0]

Provenance & license

• Source: nvidia/Arena-GR1-Manipulation-PlaceItemCloseDoor-Task (ranch_bottle_into_fridge/ranch_bottle_into_fridge_generated_100, revision arena_v0.2_lab_v3.0).
• Modification: updated head/wrist camera poses + Isaac Lab Arena (Lab 3.0) replay to regenerate camera observations (dual ego + wrist views).
• License: inherits the license of the upstream NVIDIA Arena dataset. Please refer to the source dataset for the authoritative license terms.