Loading
The brain behind smarter robotics.
Deploy and scale AI robotic solutions without deep robotics expertise. ORNA is the intelligence layer that gives industrial robots the ability to see, think, and act.
<1 day
Time to live picks for a new SI partner
100%
Edge-native, offline-first, air-gapped capable
4
Validated use cases at launch
Most AI robotics platforms need a head start. ORNA doesn't.
Traditional vision systems require CAD models, weeks of per-part scan training, or a robotics specialist on staff. ORNA handles new items on Day 1 — no re-programming, no CAD library, no dedicated robotics hires. Your existing engineers configure it. Your existing robots run it.
Traditional Industrial Robotics
Vision
Trained per part, per project
Understands what it sees — describe the part, it finds it
Programming
Robot code and teach pendants
Natural language — tell the robot the task
Who Operates it
Robotics Specialist
Your existing engineers
Hardware
Custom integration per robot and camera brand
One platform — robots and cameras are interchangeable
Time to live
Weeks of integration
Under 1 day
The SDK that speaks robot.
OrnaPy is the single source of truth for all robot execution. Write, test, and deploy anything — from traditional motion control and PLC logic to full AI agents — without touching the driver layer.
Abstracts all motion primitives
Perception-action integration handled. Focus on the application, not the plumbing.
Integrate the full workcell
MES, PLC, camera, gripper, AI Agent — same SDK, same mental model, full-stack coverage.
IDE-first development
Full Python IDE in the browser.
pick_and_place.pyOrnaPy
from ornapy import orna_api, vision, plc
# Connect to robot and peripherals
orna = orna_api.ORNA(robot="UR10e")
cam = vision.Camera(model="orbbec-335lg")
line = plc.Modbus(host="192.168.1.50")
# Configure the pick task
task = orna.configure_pick(
target="m8_bolt",
confidence_min=0.92,
retries=3,
camera=cam,
)
# Run with full reasoning trace
result = task.run(trace=True)
# All execution data stays on-edge
print(result.summary)
# → 11 picks · 0 errors · 98.2s
# Signal PLC: job done
line.write_coil("job_complete", True)
One conversation. One day to deploy.
Tell us about your workcell. We'll tell you exactly how ORNA fits — and what it takes to go live.
Contact for Quote
One platform. Full-stack intelligence.
A single operator interface to configure workcells, train vision models on your parts, and monitor every robot in real time. SIs deploy in under a day. Operators run lights-out without touching code.
Robots that explain every decision.
ORNA's neurosymbolic architecture combines foundation model perception with deterministic execution. Every action is auditable. Every failure is traceable. Nothing happens in a black box.
User
Pick all M8 bolts from the bin and place in tray A. Skip anything that looks damaged or partially hidden. We need at least 8 clean picks to continue the line.
ORNA
→ PERCEIVE3D scan — 14 objects detected
→ CLASSIFY11 × M8 bolt · conf 0.97
→ PLANgrasp sequence optimised
→ VALIDATEcollision-free path confirmed
11 M8 bolts identified. Executing pick sequence — tray A reachable.
Executing — 4 / 11 completeUser
I can see there's still a bolt on the left edge of the bin — it looks reachable from here. Why did you skip it and can we retry?
ORNA
Confidence was 0.71 — below threshold 0.92. Part was occluded by bin wall. Marked for re-scan next cycle.
Chat with AI Agent
09:41:02
3D scan initiated
Orbbec 335Lg · depth 480ms · 14 objects in FOV
Perceivedepth_res: 640×480fps: 30exposure: auto
09:41:02
11 × M8_bolt identified
VLM match · SAM2 segmentation · 3 objects rejected
Classifymodel: m8_bolt_v3latency: 312msthreshold: 0.92
09:41:03
Object_03 excluded
Occlusion detected — bin wall overlap >40%
Skipconf: 0.71reason: occludedaction: re-scan queued
09:41:03
Grasp sequence computed
MoveIt2 · 8 waypoints · collision-free
Planplanner: RRTConnectwaypoints: 8plan_time: 44ms
09:41:04
Path validation passed
No collision detected · joint limits OK
Validatejoints: 6-DoFspeed: 180mm/sforce_limit: 50N
Decision Trace · Audit Log
Teach it once. Deploy it anywhere.
VLA (Vision-Language-Action) lets you train ORNA on new tasks through demonstration — no code, no CAD, no scripting. Show the robot what to do, and it learns.
Record the task.
Connect your cameras and gripper. An operator demonstrates the task by hand — ORNA records every frame, joint angle, and gripper state. Typically 50–200 episodes per task.
Train the policy.
Select your dataset, set epochs, and start training. ORNA runs imitation learning on your episodes and shows loss curves in real time. When val loss converges — your policy is ready
Watch it live
Pick the task and the trained model, and the robot runs it live. Approve each move yourself, let it run step by step, or let it run continuously. Stop and return-to-start are always one click away.
Built for real production floors.
Six validated use cases. Same platform, same SDK, configured to your parts and process.
Contact for Quote3D VisionNo FixturesAdaptive Grasp
Random bin picking
Unstructured parts, no fixed orientation. ORNA finds, classifies, and grasps without per-part training.
Mixed SizeDynamic LayoutRapid Reconfig
Kitting & assembly
Sequenced multi-part picks for assembly-line feeding.
Force-GuidedPLC IntegrationCB3 Compatible
Palletizing/Depalletizing
6-DoF pose estimation locates parts precisely. ORNA coordinates with PLC for machine-ready signals and uses wrist F/T sensing for compliant chuck insertion.
Surface MappingDefect DetectionReal-time Feedback
Machine tending
ORNA's adaptive surface mapping enables real-time trajectory correction. Sustained cutting loads monitored via wrist F/T sensor for consistent quality assurance.
Path FollowingReal-time FeedbackConsistent Bead
Adhesive application
ORNA's adaptive surface mapping tracks the contour in real time, keeping the nozzle at a consistent standoff and speed. Bead width and continuity stay stable across curved and irregular surfaces.
Torque ControlVision AlignmentPLC Integration
Screw fastening
Vision aligns each screw to the hole; torque feedback confirms seating. ORNA logs every fastener event and signals the PLC on completion, catching cross-threads and missed screws before they reach the next station.
Common questions, answered directly.
No. ORNA uses zero-shot perception, meaning it can identify and grasp new parts without a CAD model or per-part scan training. New SKUs can typically go live within a day of being added to a workcell.
ORNA is arm-agnostic by design, built to work across major industrial robot platforms. We're UR+ certified today, covering the full Universal Robots range— with support for additional manufacturers actively expanding. No new hardware purchase required in most cases.
Most system integrator partners reach live picks in under a day. ORNA's edge-native platform deploys without cloud setup or lengthy integration — your existing engineers configure pick zones and confidence thresholds directly.
No. ORNA is designed to be configured and operated by the production engineers and technicians you already employ — not dedicated robotics or AI specialists.
Yes. ORNA is 100% edge-native and air-gapped capable. All perception, planning, and execution run locally — no cloud round-trip and no internet dependency required for operation.
CAD-trained systems require a 3D model or scan of every part before they can recognize it — adding integration time for every new SKU. ORNA's zero-shot perception generalizes to new geometry without that step.
Ready to put intelligence on your line?
Tell us about your workcell and your existing hardware. We'll show you exactly how ORNA deploys — usually within a day.
Get in touch
Talk to us about an SI partnership, a deployment, or to learn more about the platform.