Home/Features
Platform

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.

Closed loop

Six nodes that keep robots aligned with reality.

Kilnara turns the digital twin into an operating loop, not a static model.

Twin
Physics-accurate
Generate
Synthetic data
Retrain
RL / IL policies
Validate
Safety gate
Deploy
Edge release
Telemetry
Back to twin
KilnSim

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.
KilnSim · cell model
Robot
6-axis arm
Camera
calibrated
Fixture
±1.2mm
Synthetic data engine

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.
scene batch · variants

New SKU

2,800 scenes

Glare

1,900 scenes

Occlusion

3,400 scenes

Wear

2,100 scenes

KilnTrain

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.
KilnTrain · run 114
38
candidate policies
7hrs
train + rank
Validation gate

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.
policy v38 · validation
Collisions0
Peak force18N / 25N
Cycle time4.1s / 4.5s
StatusPassed
KilnEdge

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.
KilnEdge · deploy
  • Cell 04 armed
  • Policy v38 active
  • Rollback v37 ready
KilnOps

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 KilnOps
KilnOps · query

Why 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.
Capabilities

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.

Comparison

Kilnara vs. visualization-only twins vs. one-shot sims.

CapabilityKilnaraVisualization-only twinsOne-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
0hrs
Typical overnight retrain-and-validate cycle for design-partner cells.
0
Synthetic scenes generated per changed line per night.
0%
Median pick success after validated retraining.
0
Policies deployed without passing the safety gate.
Get started

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.