Documentation
Adoption Path
rFabric is designed to be adopted in stages. Teams start where infrastructure friction is highest — usually the path from robot data to training-ready datasets — and extend into model development, release management, and live operations as their program grows. The architecture stays the same at every stage; only the depth of what you run on it changes.
Adoption Stages
Stage 1 — Data foundation
Establish one trustworthy record of robot data intake, labeling, and dataset preparation while keeping the rest of your stack unchanged.
- Data Ingestion & Preprocessing
- Annotation & Labeling
- Dataset Curation & Versioning
- Unified Data Model & Traceability for core data entities
- Platform API, CLI & SDKs for data operations
- Identity & Access and Governance & Tenancy from day one
Stage 2 — Dataset quality at depth
Deepen the highest-leverage surface in robot learning: dataset quality, curation rules, and review rigor.
- Automated quality scoring and policy-driven selection
- Multi-operator coordination and correction loops
- Orchestration for data-to-dataset automation
- Governance policy for retention, regions, and approval paths
- Metering for storage and media-processing costs
Stage 3 — Model development
Extend the same lineage from governed datasets into reproducible training runs, evaluation results, and promoted model identity.
- Model Training & Evaluation
- Model Registry
- Training lineage inside the shared data model
- Orchestration for dataset-to-model workflows
- Compute cost visibility through Metering & Billing
Stage 4 — Release management
Turn approved models into governed production artifacts and controlled rollout flows.
- Release Packaging
- Deployment & Updates
- Environment progression and deployment approvals
- Release and deployment entities in the shared graph
- Promotion, rollout, and rollback flows orchestrated end-to-end
Stage 5 — Live operations
Close the loop from deployed fleets back to data, evaluation, and better future releases.
- Fleet Management
- Monitoring, Alerting & Maintenance
- Human Intervention & Robot Control
- Operations lineage across telemetry, incidents, maintenance, and interventions
- Full lifecycle orchestration and operations-aware metering
The Backbone Grows With You
Every stage runs on the same shared backbone. Adopting a new stage does not introduce a new system — it deepens the one already in place.
Unified Data Model & Traceability
Expands from datasets and episodes into training runs, promoted models, artifacts, deployments, telemetry, and interventions — one graph across the lifecycle.
Orchestration
Starts with data-to-dataset automation and grows into promotion, rollout, rollback, maintenance, and human intervention workflows.
Governance & Tenancy
Deepens from workspace isolation into region policy, environment promotion rules, operational approvals, and regulated deployment controls.
Metering & Billing
Covers storage, compute, media processing, rollout, and fleet operations so cost stays attributable as usage grows.
Why Staged Adoption Works
Start where the pain is
The data-to-training path is where most robotics teams feel infrastructure burden first, so that is where adoption usually begins.
One architecture throughout
Each stage deepens the same backbone and component map. There is no replatforming decision between stages.
Value compounds
Every additional stage enriches the same system of record, which makes the parts you already run more valuable.
Your existing tools keep working
Visualization tools, training code, and operational systems you rely on today stay in place. The platform standardizes lineage and handoffs around them.