datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
RoboDojoGOAI-2026robodojo-assets-packed
RoboDojo Assets — verified transport shards
This public dataset is a byte-preserving transport repack of Assets/ from
RoboDojo-Benchmark/RoboDojo, pinned at source commit
aada55a7a58b520a2004b9db02e884144b6e83e7. It reduces 15,335 source files to 11 independently downloadable and
restorable tar.zst shards. No payload file was changed.
这是 RoboDojo Assets/ 的公开传输分片：固定源 commit、减少小文件数量、每片可独立下载/重试，
完整复原后会逐文件做 SHA-256 校验。
Contents
Source files: 15,335
Source logical… See the full description on the dataset page: https://huggingface.co/datasets/didfd/robodojo-assets-packed.RoboDojo-taco-visual-gemini
RoboDojo-taco-visual-gemini
The visual-grounding variant of RoboDojo-taco-gemini: the same
RoboDojo long-horizon episodes (8 tasks, 800 episodes, 25 fps) with the same dense high-level labels (Gemini 3.7 Flash under the
task-specific context induced for each task), except that a target position leaves the label text and is drawn into the
low-level policy's keyframe slot.
In the source labels a target that words cannot identify is named by its image coordinates on the 0–1000… See the full description on the dataset page: https://huggingface.co/datasets/Myungkyu/RoboDojo-taco-visual-gemini.robodojo-ablation-bundle
RoboDojo L4 Inspect EEF ablation videos + traces
Organized export of general_pickup / ICL probe rollouts (eval MP4 + planner traces).
Dir
Content
01_skill_bimanual_gray_ok
Early gray-block bimanual lift L4 (OK)
02_negate_xyz
XYZ negate + prior L4 Inspect layout0
03_vision_flip
flipud + fliplr
04_jitter_mean10cm
Sphere mean-10cm per-move jitter
05_camera_mask
Wrist / right-wrist / L9×2 masks
06_icl
Image+EEF and text ICL (34 tasks)
See MANIFEST.txt and… See the full description on the dataset page: https://huggingface.co/datasets/Solomonz/robodojo-ablation-bundle.astra-robodojo-rollouts
Astra RoboDojo Evaluation Records
Rollout records from the evaluations in GPT 6 Astra as an Embodied Policy,
by Jiayi Su, Yixin Zheng, Mi Yan, Li Yi, Zhizheng Zhang, and He Wang. This archive
includes action proposals, executed actions, observations, robot states,
model-provided explanations and reasoning summaries, and metadata for reproducing
the evaluation settings, together with a reader and documentation.
Report
Public controller source
Data schema and alignment… See the full description on the dataset page: https://huggingface.co/datasets/YuMoool/astra-robodojo-rollouts.RoboDojoData
RoboDojo Reviewed Dataset
This dataset contains human-reviewed robot manipulation episodes generated with RoboDojo.
It is organized by task and robot, with 50 valid episodes selected for each task/robot pair.
本数据集包含由 RoboDojo 生成、经过人工视频审核的机器人操作数据。每个 task/robot
组合选取 50 条有效 episode。
Tasks and robots
Tasks:
pick_anything
stack_blocks
dual_bottles_pick
general_pickup
insert_tube — legacy single-tube insertion (formerly insert_tubes)
Robots:
x5
piper_x
franka… See the full description on the dataset page: https://huggingface.co/datasets/dd1581334/RoboDojoData.RoboDojo_demo_dataRoboDojoData_demo
RoboDojoData Demo
This repository contains a small review-oriented subset of RoboDojo data.
It is intended for inspecting data structure, camera views, task coverage, and
embodiment compatibility before production data is published.
Production-scale data is published separately in
dd1581334/RoboDojoData.
Original/production dataset:
Repository ID: dd1581334/RoboDojoData
URL: https://huggingface.co/datasets/dd1581334/RoboDojoData
Included tasks
deposit_coin… See the full description on the dataset page: https://huggingface.co/datasets/dd1581334/RoboDojoData_demo.IsaacLab-Arena_RoboDojo_Assets
IsaacLab-Arena × RoboDojo Assets
Assets for evaluating RoboDojo dual-arm tasks on IsaacLab-Arena with the ARX X5 (dual_x5) embodiment.
Contents
Path
Description
Embodiment/x5/
ARX X5 dual-arm robot USD / meshes
texture/
Dome / material textures used by eval
Layouts/RoboDojo/
Scene layouts for 10 RoboDojo tasks
Objects/RoboDojo/
Object USD instances referenced by those layouts
Tasks
build_tower, classify_objects… See the full description on the dataset page: https://huggingface.co/datasets/littleTang/IsaacLab-Arena_RoboDojo_Assets.RoboDojo-taco-gemini
RoboDojo-taco-gemini
RoboDojo long-horizon episodes (8 tasks, 800 episodes, 25 fps) with dense high-level labels produced by the TACOR offline annotator:
Gemini 3.7 Flash reads each whole episode as one video clip (one sample every 25 frames) and labels every sampled frame under the
task-specific context (taco) induced for that task, in its hybrid form: labels name the target object's image coordinates only
where words cannot identify it (a random instance of a class that… See the full description on the dataset page: https://huggingface.co/datasets/Myungkyu/RoboDojo-taco-gemini.RoboDojo-preset-gemini
RoboDojo-preset-gemini
RoboDojo long-horizon episodes (8 tasks, 800 episodes, 25 fps) with dense high-level labels produced by the TACOR offline annotator:
Gemini 3.7 Flash reads each whole episode as one video clip (one sample every 25 frames) and labels every sampled frame given only the
subtask preset of the task: the label list of the task-specific context (its hybrid form, so some labels carry a coordinate slot),
without the context's boundary criteria, sequence rule or… See the full description on the dataset page: https://huggingface.co/datasets/Myungkyu/RoboDojo-preset-gemini.robodojo-sim
RoboDojo Sim Dataset
Dexdata format RoboDojo simulation dataset.
Download and Extract
If the dataset is uploaded as split tar archives:
cat robodojo_sim.tar.gz.part-* | tar -xzf -
After extraction, the dataset directory will be:
robodojo_sim/
Dataset Structure
robodojo_sim/
├── jsonl/
│ └── <task_name>/
│ └── episode_XXXXXXX.jsonl
└── video/
└── <task_name>/
└── episode_XXXXXXX/
├── cam_head.mp4
├──… See the full description on the dataset page: https://huggingface.co/datasets/Dexmal/robodojo-sim.robodojo-hlp-local-sttp
RoboDojo HLP Local-STTP — v3 training set
One question per row: has the robot moved past the subtask it was given, and if so, what comes
next? A high-level planner is called every 16 simulator steps, sees a few head-camera frames
and the sentence the controller is currently holding, and answers with a flag plus the next
subtask sentence.
41,530 rows built from 995 expert demonstration episodes across 10 RoboDojo tasks. Derived from
RoboDojo-Benchmark/RoboDojo;
see the licence… See the full description on the dataset page: https://huggingface.co/datasets/ghkim-rlwrld/robodojo-hlp-local-sttp.RoboDojo-taco-visual2-gemini
RoboDojo-taco-visual2-gemini
visual2 variant (2026-09-14): unlike RoboDojo-taco-visual-gemini (marker in a separate keyframe video), here the point marker (red disc, radius 1.9 % of the width, white ring) is drawn into the head camera video itself for every frame whose governing label names a position, and the label text says ... the location marked with the red dot ... (tic-tac-toe: ... <cell> marked with the red dot.). The dataset keeps the three live views only (no keyframe… See the full description on the dataset page: https://huggingface.co/datasets/Myungkyu/RoboDojo-taco-visual2-gemini.RoboDojo-taco-luna
RoboDojo-taco-luna
RoboDojo long-horizon episodes (8 tasks, 800 episodes, 25 fps) with dense high-level labels produced by the TACOR offline annotator:
GPT-5.6 Luna (gpt-5.6-luna) reads the frames of each episode sampled every 25 frames as labelled images and labels every sampled frame under the
task-specific context (taco) induced for that task, in its hybrid form: labels name the target object's image coordinates only
where words cannot identify it (a random instance of a… See the full description on the dataset page: https://huggingface.co/datasets/Myungkyu/RoboDojo-taco-luna.RoboDojo-preset-luna
RoboDojo-preset-luna
RoboDojo long-horizon episodes (8 tasks, 800 episodes, 25 fps) with dense high-level labels produced by the TACOR offline annotator:
GPT-5.6 Luna (gpt-5.6-luna) reads the frames of each episode sampled every 25 frames as labelled images and labels every sampled frame given only the
subtask preset of the task: the label list of the task-specific context (its hybrid form, so some labels carry a coordinate slot),
without the context's boundary criteria… See the full description on the dataset page: https://huggingface.co/datasets/Myungkyu/RoboDojo-preset-luna.RMBench_Robodojo_for_ProgressLMV2robodojo-hybrid-episodes
RoboDojo Hybrid GPT and Pi 0.5 Compact Episodes
This release retains all 258 started attempts: 216 canonical complete cases
(54 task variants, four repetitions each), eight complete pilot episodes, and 34
superseded or interrupted attempts. Canonical results include 57
successes and 159 failures. Failed complete episodes are retained.
Seven attempts have no aligned recording; their explicit manifests and available
model behavior remain present. indexes/episodes.jsonl lists every… See the full description on the dataset page: https://huggingface.co/datasets/shuaishuaicdp/robodojo-hybrid-episodes.RoboDojo_RealStellaVLA-Data-RoboDojo
StellaVLA — RoboDojo trajectories
robodojo_lerobot.tar 120 GB unpack in place -> robodojo_lerobot_v30/
Byte-identical copy of the joint variant of the RoboDojo LeRobot v3.0 export, kept
next to the StellaVLA results so the trajectories and their annotations are guaranteed to
line up. Contents of the tar:
path
size
notes
robodojo_lerobot_v30/meta/
54 MB
includes the modality.json upstream does not ship
robodojo_lerobot_v30/data/
262 MB
14-D joint state + action… See the full description on the dataset page: https://huggingface.co/datasets/siyuhsu/StellaVLA-Data-RoboDojo.RoboDojo_realRoboDojo_simRoboDojoLibero-mem_RMBench_Robodojo_Manualrobodojo_lerobot_v3.0Benchmark_dataset_RMBench_RobodojoG05-RoboDojo-12taskrobodojo-lerobot-v3.0
RoboDojo LeRobot v3.0
This dataset packages RoboDojo robot-manipulation demonstrations in LeRobot v3.0 format for EasyWAM training. It contains synchronized high-view and dual-wrist RGB video, bimanual joint state, actions, and natural-language task metadata.
Dataset Summary
Episodes
Frames
Tasks
Recording rate
3,500
1,856,102
35
25 FPS
Structure
robodojo-lerobot-v3.0/
├── data/ # frame-level Parquet files
├── meta/… See the full description on the dataset page: https://huggingface.co/datasets/OpenMOSS-Team/robodojo-lerobot-v3.0.