DROID Dataset
In-the-wild real-world robot manipulation
A geographically distributed Franka manipulation dataset collected across diverse scenes, tasks, operators, and institutions.
VLA fine-tuning
Strong real-world diversity; verify release subset completeness before processing.
Inspect schema and run a bounded sample audit before committing to the full release.
Has observation, action/state proxy, and task or language context.
fit 85 · confidence 55
Has visual observations plus geometry, calibration, depth, or reconstruction cues.
fit 82 · confidence 55
Contains observation, intent, and action/state, but feedback/correction signal is weak.
fit 72 · confidence 55
Verified facts and provenance
Claims, metadata verification, and sample verification are shown separately.
Unknown — no machine-readable schema facts have been captured.
Unknown — metadata conclusions do not prove sample coverage, alignment, or file integrity.
Declared loop signal coverage
Signals inferred from official metadata; Data pipeline verification is still pending.
Observation / ego video
video · camera calibration
Action / hand pose / robot state
actions · robot state · In-the-wild real-world robot manipulation · A geographically distributed Franka manipulation dataset collected across diverse scenes, tasks, operators, and institutions.
Gaze / attention
No decision-grade evidence captured yet.
Language intent / task phase
language · A geographically distributed Franka manipulation dataset collected across diverse scenes, tasks, operators, and institutions.
Feedback / correction / failure
No decision-grade evidence captured yet.
Sim-real pairing
camera calibration · A geographically distributed Franka manipulation dataset collected across diverse scenes, tasks, operators, and institutions.
License / format / access
Open · RLDS · HDF5
Model and task fit · OpenBot inference
Has observation, action/state proxy, and task or language context.
fit 85 · confidence 55
Has visual observations plus geometry, calibration, depth, or reconstruction cues.
fit 82 · confidence 55
Contains observation, intent, and action/state, but feedback/correction signal is weak.
fit 72 · confidence 55
Failure, correction, intervention, and recovery annotations have not been verified.
fit 50 · confidence 15
Good tasks
Blockers and unresolved evidence
- Gaze / attentionunknownNot enough evidence to classify this signal. Verify metadata or a bounded sample.
- Feedback / correction / failureunknownNot enough evidence to classify this signal. Verify metadata or a bounded sample.
Raw dataset signals
OpenBot fit
- VLA fine-tuning
- Environment generalization
- Operator drift analysis
Related models and papers
Model references linked to similar loop signals.
pi0.5
A VLA model designed for open-world generalization by combining heterogeneous robot data, semantic subtask prediction, and high-level knowledge transfer.
Gemini Robotics 1.5
An agentic robotics model combining advanced multimodal reasoning with vision-language-action control for multi-step physical tasks.
LingBot-Depth
A masked depth model for refining incomplete sensor depth into metric geometry for reconstruction, tracking, and manipulation.
X-VLA
A 0.9B soft-prompted VLA that adapts a shared backbone across heterogeneous embodiments and action spaces.
WALL-OSS 0.5
An open 4B VLA pretrained across more than 20 embodiments, with physical-hardware evaluation of pretrained robotic capability.
OpenVLA-OFT
An optimized OpenVLA fine-tuning recipe with continuous action chunks, multi-image input, and faster high-frequency control.
Integration notes
- Strong real-world diversity; verify release subset completeness before processing.