Application note · Specialised robotics

Application Note: Robot Perception Training, Digital Twin and Deployment

A closed-loop workflow from calibrated data and simulation through GPU training to bounded, low-latency perception beside the robot cell.

This design trains pose, segmentation or inspection models on AxiForge, validates them against a digital twin and production captures, then serves them on AxiOverseer while AxiGovernor supports controls integration.

Dataset and frames

Version images, depth/point clouds, annotations, camera intrinsics/extrinsics, robot/tool frames, product CAD and lighting state. Keep train/validation/test splits independent by physical scene or production period. Include negative and uncertain examples.

Simulation and training

Use the twin to generate labelled rare poses, occlusion and fault states, but randomise only physically plausible variables. Train on AxiForge with two 96GB GPUs and record seeds, image digest and data version. Compare synthetic-only, real-only and mixed training to quantify the sim-to-real gap.

Perception service

Optimise the approved model with TensorRT where appropriate and deploy behind a bounded interface on AxiOverseer. DeepStream can handle multiple video streams; ROS 2/NITROS can reduce supported copy paths. Timestamp every input/output and reject stale or uncalibrated frames.

Cell integration

AxiGovernor hosts TwinCAT engineering, ADS integration and virtual commissioning. The perception service returns class/pose/confidence; the robot/controller applies reachability, collision, interlock and safety checks. Network or AI failure leads to a defined state, not an unbounded command.

Production acceptance

Test every part family, pose, lighting condition and line rate. Measure sensor-to-decision p99 latency, dropped frames, pose error and failure recovery. Run shadow mode and supervised trials before enabling commands. Preserve the previous model for immediate rollback.

Primary technical references

References are provided for software architecture and implementation planning. Validate the versions, licences, support matrix and regulated-use requirements applicable to the final deployment.

From technical concept to production system

Apply this technology through an Axiotech engineering work package

Axiotech can connect the compute, AI or analytics platform to the machine controls, data contracts, validation evidence and lifecycle-support model required for industrial use.

Relevant Hyperion platforms

Hyperion X1260

AxiForge

A professional two-GPU training, visualisation and applied-research server for larger datasets and models.

One 4U node; 96 AMD EPYC cores; 768GB ECC DDR5; two NVIDIA RTX PRO 6000 96GB GPUs; 192GB aggregate VRAM and high-capacity NVMe scratch.

Hyperion X1150-C

AxiGovernor

A controls-engineering and virtual-commissioning platform for TwinCAT, PLC, motion, simulation and private engineering AI.

One 4U node; 96 AMD EPYC cores; 768GB ECC DDR5; one NVIDIA RTX PRO 5000 48GB; engineering storage and remote management.

Hyperion X1250

AxiOverseer

A production-intelligence node for machine vision, telemetry, predictive maintenance and low-latency factory AI.

One 4U node; 96 AMD EPYC cores; 768GB ECC DDR5; two NVIDIA RTX PRO 5000 48GB GPUs; protected local/NAS data services.

Configuration and quotation

Validate this workload on Hyperion

Final architecture and price depend on representative code and data, concurrency, storage, networking, site infrastructure, component availability and export compliance.

Request Formal Quotation