Features
A closed-loop platform for robotic factories: twin the line, generate edge cases, retrain policies, validate safety, deploy to the edge, and learn from telemetry.
Six nodes that keep robots aligned with reality.
Kilnara turns the digital twin into an operating loop, not a static model.
Physics-accurate twin
A calibrated operational model of the cell that stays aligned with real sensors, robots, fixtures, and cycle constraints.
- OpenUSD line model — geometry, robot kinematics, fixtures, and work envelopes.
- Sensor calibration — camera, force, and controller data reconciled to the twin.
- Continuous drift checks — telemetry flags when reality diverges from simulation.
Generate the hard cases before they hit production.
Create labeled edge-case datasets for new SKUs, lighting shifts, occlusion, wear, and layout drift without risking throughput.
- Scenario expansion — thousands of controlled variations per cell.
- Automatic labels — segmentation, 6-DoF pose, and contact events from simulation.
- Reality-gap controls — randomize materials, lighting, and physics within bounds.
New SKU
2,800 scenes
Glare
1,900 scenes
Occlusion
3,400 scenes
Wear
2,100 scenes
Nightly retraining
Turn new telemetry and generated data into candidate policies while the line is offline, then rank them against real production targets.
- RL and imitation learning — choose the right method for each task.
- Experiment lineage — trace datasets, parameters, and policy versions.
- Scheduled runs — retrain nightly or on change events.
Safety validation before deployment
Every candidate policy must pass measurable safety, performance, and process constraints before operators approve release.
- Configurable thresholds — success rate, force, cycle time, and collision limits.
- Human approval — route sign-off to robotics, controls, or operations.
- Audit and rollback — every release is traceable and reversible.
| Collisions | 0 |
| Peak force | 18N / 25N |
| Cycle time | 4.1s / 4.5s |
| Status | Passed |
Edge deployment
Ship validated policies close to robots for deterministic inference, production telemetry, and controlled release management.
- Low-latency inference — optimized for cell-side NVIDIA edge hardware.
- Canary and rollback — release gradually and revert instantly.
- Telemetry stream — cycle outcomes, exceptions, and drift back to the twin.
- Cell 04 armed
- Policy v38 active
- Rollback v37 ready
Agentic operations copilot for the robotic line.
Ask why a cell slowed down, what changed in the twin, or whether a policy is safe to deploy. KilnOps works across telemetry, simulation runs, validation results, and release history.
See KilnOpsWhy did cell-04 drop below 96% after shift change?
- Identified glare variance — camera 2 exposure drifted 14%.
- Generated retrain plan — add 1,600 glare scenarios tonight.
Enterprise controls around the loop.
Deployment modes
Cloud, VPC, on-prem, or air-gapped factory environments.
Integrations
Robots, PLCs, cameras, MES, data lakes, and ticketing systems.
Governance
RBAC, approvals, release records, and full policy lineage.
Observability
Telemetry, run history, drift alerts, and outcome dashboards.
Simulation assets
Versioned geometry, materials, calibration, and scenario packs.
Model registry
Track candidate, validated, deployed, and retired policies.
Change analysis
Estimate impact of parts, fixtures, lighting, and throughput targets.
Security
SSO/SAML, audit logs, private networking, and data ownership.
Kilnara vs. visualization-only twins vs. one-shot sims.
| Capability | Kilnara | Visualization-only twins | One-shot sims |
|---|---|---|---|
| Continuously syncs with telemetry | — | ||
| Generates labeled synthetic data | — | ||
| Retrains policies nightly | — | — | |
| Validates safety before release | — | — | |
| Deploys to edge robots | — | — | |
| Feeds production outcomes back | — | — |
See the closed loop on a line like yours.
Bring a recent changeover, failure mode, or throughput target. We will show how Kilnara models, retrains, validates, and deploys.