Step-By-Step Guide
1. Joining the Campaign
Each data collection effort starts with a campaign, defining what data is needed, where it’s collected, and how contributors are compensated.
Campaigns specify:
Target objective (e.g. indoor navigation, perception)
Required sensors
Geographic scope and duration
Compensation rate
Campaign types include: navigation, perception, environmental sensing, and motion capture.
2. Data Collection
Collectors join campaigns through the RoboX mobile app.
Data is collected passively in the background during normal activity
Sensor sampling rates are campaign-defined
Data is encrypted and buffered locally
Collection pauses automatically if quality thresholds aren’t met
Uploads occur automatically when device and network conditions allow.
3. On-Device Anonymization
All anonymization happens before upload.
Locations are coarse-grained and hashed
Faces, license plates, and identifying visuals are masked on-device
Hardware identifiers are stripped
Pseudonymous IDs rotate per campaign
Timestamps are slightly randomized to prevent correlation
Raw, identifiable data never leaves the device.
4. Upload & Validation
Anonymized data is uploaded over encrypted connections and validated automatically.
Validation checks include:
Completeness and continuity
Sensor plausibility
Duplicate detection
Cross-validation against nearby collectors
Only validated data qualifies for compensation.
5. Aggregation & Dataset Creation
Validated data is standardized, annotated, and grouped into datasets by:
Campaign
Geography
Sensor modality
Intended use case
Metadata includes collection context, device characteristics, and quality scores.
6. Data Access
Authorized users access datasets via the RoboX API:
Streaming for continuous ingestion
Batch downloads for offline training
Query interface for exploration and preview
Datasets are provided in ML-ready formats (TensorFlow, PyTorch, or raw).
Each dataset includes documentation covering methodology, anonymization, limitations, and recommended use.
Last updated