Define the decisions the twin must support
Start with questions: can the robot reach the station, meet cycle time, perceive the object, avoid mission conflicts and recover from a failure? Give each question a metric and tolerance. Otherwise the model becomes a demonstration that cannot be accepted objectively.
Inputs include current geometry, kinematics, load mass and centre of gravity, drive characteristics, sensor fields of view, network latency and upstream interfaces. Link CAD, maps and configuration to a named project version.
Move from a model to a closed loop
The first level checks reachability, collisions and cycle time. Software-in-the-Loop runs production control logic against the simulator. Hardware-in-the-Loop moves a controller, compute unit or network onto target hardware while keeping the environment virtual. Each layer covers a different risk class and does not replace the next.
Calibrate sensor models with real measurements. Synthetic images and randomised lighting, materials and object poses can widen perception tests, but the simulation-to-reality gap must be measured rather than concealed.
Build a scenario library, not one perfect run
Include normal cycles, edge positions, a blocked aisle, lost marker, unstable object, message delay, camera loss, malformed mission and recovery after an emergency stop. A fleet also needs dense traffic, priority conflicts, charging and removal of one robot from service.
Store every result with model, software and parameter versions. This turns a demonstration into a reproducible regression test after updates.
Complete virtual acceptance on the real site
Simulation reduces dangerous and expensive testing but cannot replace validation of floors, lighting, traction, electromagnetic interference, human behaviour and mechanical tolerances. Repeat critical scenarios on site and compare the measurements with the model.
Keep the twin after launch for upgrades, new routes, operator training and incident reconstruction. Its value is the number of changes validated before touching a live process.