| Before rolling out routing changes in a live warehouse, we built a digital twin in Omniverse to simulate AGV and picker traffic under different layout and routing-algorithm assumptions. Goal was to catch throughput bottlenecks (congestion at intersections, picker-AGV conflicts) before touching the physical floor. |
|---|
The USD-based workflow made it fairly straightforward to bring in the existing CAD layout and iterate on routing logic without needing a full physics-accurate simulation for every test — we only needed physics accuracy for collision/congestion behavior, not for anything downstream.
One open question: we’re currently validating the twin against real sensor data from the physical warehouse (comparing simulated vs actual throughput at a few checkpoints) to calibrate how much to trust the sim before trusting a routing change enough to deploy it. Is there a more standard approach people use for this kind of sim-to-real calibration in logistics/warehouse contexts specifically, versus the robotics-focused calibration approaches that seem more commonly documented?
This came out of logistics software work at Dev Technosys — happy to go into more detail on the routing algorithm side if useful, but the calibration question above is the part we’re stuck on.