TECHNOLOGY PLATFORM
VisionAI Platform: Industrial Computer Vision for Robotics
VisionAI Platform is Viroteq’s industrial computer vision engine — the AI-powered vision layer that powers VisionAI Sorting, RobotDepalr, and every Viroteq cell that needs to see, classify, and act in real time. Sub-10ms classification, fused 2D and 3D scene understanding, on-premise inference on edge IPCs, and brand-agnostic robot integration make VisionAI Platform the production-grade alternative to off-the-shelf vision libraries. As a result, vision is never the bottleneck inside a Viroteq robotic cell — and the same engine scales from a single inspection station to a multi-cell plant rollout without changing the software stack.
- ✓ Sub-10ms classification at production resolutions
- ✓ Fused 3D + 2D vision with point cloud processing
- ✓ Edge inference — on-premise IPC, no cloud dependency
100+
Production deployments
<10 ms
Real-time classification latency
3D + 2D
Fused vision capability
On-prem
Edge inference on Industrial PCs
Why VisionAI Platform Outperforms Off-the-Shelf Vision
Generic vision libraries — open-source toolkits, framework-only stacks, or vendor SDKs limited to a single camera brand — solve the easy half of an industrial vision problem. They identify items in clean light, on a still conveyor, with curated training data. The hard half is everything a real production cell throws at the camera: changing light, dirty optics, partial occlusion, mixed SKUs, and millisecond-bounded cycle times. The engine was built from the start for that hard half.
However, raw algorithmic capability is only the start. A production-grade vision system needs deterministic latency budgets, repeatable training data pipelines, model versioning, on-line monitoring, and a clean integration path to the robot or conveyor controller. Off-the-shelf libraries leave all of that as an exercise for the integrator. Therefore, every cell built on a generic stack ships a one-off middleware layer that no one owns long-term.
Furthermore, training data is where many vision projects stall. Hand-labelling thousands of frames is expensive, and the resulting model drifts the moment product packaging changes. The platform includes an adaptive training pipeline — operators flag misclassifications through the console, the platform retrains overnight, and the new model is validated against a hold-out set before promotion. As a result, vision accuracy improves continuously rather than degrading with time.
Edge latency is the final differentiator. Vision running in the cloud — or on a non-real-time OS — cannot meet sub-10ms decision budgets. According to ISO/IEC 22989 AI terminology and trustworthiness standards, deterministic, auditable AI behaviour is a precondition for industrial deployment. Viroteq’s edge-first technology platform places VisionAI Platform inference next to the robot controller, with full classification telemetry exposed to integrators.
How VisionAI Platform Works in Production Cells
The pipeline begins at the camera. Industrial sensors — Basler, Cognex, Intel RealSense, Photoneo, or any GenICam-compliant unit — stream frames into the edge IPC. The runtime preprocesses, normalises, and routes each frame to the appropriate model: classification, defect detection, pose estimation, or 3D segmentation.
Next, AI-powered inference runs locally on the IPC. Advanced AI algorithms, trained on production data from real Viroteq cells, return classification labels, bounding boxes, segmentation masks, and 3D coordinates inside the sub-10ms budget. The runtime then publishes results to the robot controller — FANUC, ABB, KUKA, Universal Robots, Yaskawa, or Stäubli — over REST API, WebSocket, or OPC-UA.
Additionally, the platform handles the messy reality of a production floor. Misclassifications are logged with the original frame and operator override. The platform learns from those corrections automatically. For depalletizing, sorting, or quality control, the same engine powers every vision-driven Viroteq product through a single runtime and a single set of APIs.
Viroteq Products Powered by VisionAI Platform
The platform is the shared vision engine behind every Viroteq application that needs to see and decide in real time. The same runtime, the same APIs, and the same operator console, training tools, and edge IPC hardware power sorting, depalletizing, and vision-assisted stacking — three distinct production workflows running on a single AI engine. As a result, customers running multiple Viroteq products operate one vision stack across cells, not three, which simplifies training, maintenance, spare parts, and software upgrades across the plant. Furthermore, integrators reuse calibration and commissioning playbooks across applications.
VisionAI Sorting
VisionAI Sorting is the production application of the platform for parcel, food, and material recovery sorting. Real-time classification routes items to outbound lanes at conveyor speed with sub-10ms decisions per frame.
RobotDepalr
RobotDepalr uses the platform’s 3D scene understanding to identify items on inbound mixed pallets. Depth cameras and point cloud processing return item geometry, pose, and grasp coordinates to the robot in real time.
RobotStackr Vision
RobotStackr Vision combines vision inspection with mixed-case stacking. The system validates incoming case identity, dimensions, and damage in real time, then feeds the data to StackrBrain for optimal pallet placement.
Core Capabilities of VisionAI Platform
3D Scene Understanding
Depth cameras, stereo vision, and structured light feed the 3D pipeline. Point cloud processing returns item geometry, pose, and grasp coordinates in the same frame budget as 2D classification, which is essential for bin picking and depalletizing.
Real-Time Defect Detection
Surface defects, packaging damage, label errors, and dimensional outliers are identified inline at conveyor speed. The runtime exposes confidence scores and bounding boxes per defect class, so downstream cells can sort, reject, or quarantine without human review.
Adaptive Model Training
Operators label edge cases through the console. The platform retrains overnight against a hold-out validation set and promotes the new model only when accuracy improves. As a result, vision performance climbs with deployment time instead of drifting downward.
VisionAI Platform Use Cases Across Industries
The platform powers vision-driven cells across the full Viroteq customer base — from food processing and parcel logistics to material recovery and inbound supply. The same runtime serves every industry, and the same APIs feed every robot brand. Furthermore, the platform’s adaptive training pipeline means a new product family is onboarded with sample images rather than custom code. In addition, integrators reuse calibration profiles and operator workflows across cells, which shortens commissioning and keeps maintenance simple. As a result, vision-driven automation scales from pilot cell to full plant without rebuilding the stack per industry vertical.
Food Quality Inspection
The platform classifies produce by size, ripeness, surface defects, and foreign objects at conveyor speed. Reject lanes are triggered inline without slowing the upstream line.
Parcel Sorting & Routing
Real-time classification reads barcodes, addresses, and parcel features at sortation speed. The runtime routes items to the correct outbound lane with sub-10ms decisions.
Material Recovery
The platform separates plastics, metals, paper, and contaminants on recycling lines by polymer type, colour, and shape. Operator feedback continuously refines fraction purity.
Inbound Pallet Inspection
Supplier pallets are scanned on arrival. The engine identifies items, validates layer integrity, and flags damage before the depalletizer touches the load.
Random Bin Picking
3D scene understanding identifies pickable items in unstructured bins. The platform returns grasp coordinates, confidence scores, and collision-free trajectories per item.
Quality Assurance
The platform validates finished-goods packaging, labels, and seals against the production spec. Pass/fail telemetry feeds the MES with full image evidence per unit.
Production Line Inspection
Inline quality and defect detection on packaging lines. Every unit is photographed, classified, and either passed downstream or routed to a reject lane with full image evidence.
Distribution Center Routing
Parcel sortation at conveyor speed. Barcodes, addresses, dimensions, and shape are read in a single frame to drive diverters and put walls without operator intervention.
Material Recovery Facility
Polymer-, colour-, and shape-aware separation on mixed-waste streams. Vision-guided robots pick targeted fractions while operator feedback refines purity per shift.
<10 ms
Classification latency per frame
100+
Production deployments worldwide
99%+
Classification accuracy in production
Engineered for Production-Grade Reliability
The platform is built around a deterministic runtime, not a research notebook. Every inference call carries a measured latency budget, a model version, and a confidence score. As a result, integrators and plant teams can audit vision behaviour against PLC and robot controller cycles instead of guessing at a black box. Moreover, model versioning means every promotion is reversible — a regression rolls back in seconds, not days.
Training methodology follows a strict pipeline. Sample images are collected from the live cell, labelled in the operator console, validated against a hold-out set, and only then promoted to the production model. Additionally, edge cases flagged by operators are captured with the original frame and the override decision, which feeds the next training cycle. Therefore, accuracy improves continuously while the production line runs unaffected.
Monitoring is built in. Per-frame timing, confidence distributions, model version, and drift indicators are exposed through the runtime API and the operator HMI. Furthermore, integration with OPC-UA, REST, and standard MES dashboards means vision telemetry sits alongside conveyor, robot, and quality metrics rather than in a separate silo. Plant engineers see exactly when, where, and why a frame was classified — full transparency, no proprietary middleware.
Edge deployment closes the reliability loop. The runtime ships on industrial-grade IPCs validated for 24/7 production, with redundant networking and supported by Viroteq’s remote diagnostics through Viroteq’s product portfolio. In addition, no cloud dependency means the engine survives external outages, restricted IT environments, and air-gapped factories without compromise. According to public computer vision benchmarks from OpenCV, real-time inference under deterministic budgets is the central challenge in industrial vision — VisionAI Platform was engineered to clear that bar on day one.
Integrating VisionAI Platform With Your Cell
The platform integrates with your cell through the cameras, networks, and controllers you already have. On the optical side, the runtime is camera-agnostic and supports Basler, Cognex, Intel RealSense, Photoneo, IDS Imaging, and any GenICam-compliant industrial sensor — 2D area-scan, 3D depth, stereo, and structured-light units alike. Furthermore, custom integrations for specialty sensors are available without rebuilding the core pipeline.
On the controls side, the runtime exposes classification, coordinates, and confidence scores over REST API, WebSocket, and OPC-UA. Robot brand support spans FANUC, ABB, KUKA, Universal Robots, Yaskawa, and Stäubli through standard protocols — no proprietary teach pendants, no vendor SDKs. In addition, vision-to-robot calibration is handled by a guided procedure during commissioning, so the coordinate frame between camera and robot is locked down without bespoke maths. As a result, deployments are predictable in scope and timeline. To see the platform integrated against your specific cell, book a Viroteq demo or talk to our solutions team.
Camera Brands
Basler, Cognex, Intel RealSense, Photoneo, IDS Imaging, plus all GenICam-compliant sensors.
REST API
Standard JSON request/response over HTTPS for classification, coordinates, and telemetry.
OPC-UA
Bridge to plant historians, MES, and SCADA via the industrial data standard for vision telemetry.
Edge IPC
Inference on dedicated Industrial PC inside the cell. No cloud dependency, deterministic latency.
VisionAI Platform — Frequently Asked Questions
Ready to See VisionAI Platform in Action?
Book a personalised demo and see how VisionAI Platform delivers sub-10ms classification, 3D scene understanding, and adaptive training inside your existing robot cell. Explore vision sorting solutions or compare against RobotDepalr.
Industrial Computer Vision Built for Modern Robotics
Bring your cameras, your robot brand, and your cycle-time target — Viroteq specialists will map a VisionAI Platform deployment that fits your cell footprint and keeps production running.
