Test the production conditions around the robot.
An event-model comparison visualised in Isaac Sim: compare shorter transfer times with a second tray drawer, then introduce longer inspection times or bursts of incoming parts. A separate contact trial tests a limited handling sequence.
Watch the cell and the comparison.
Replay the same production window.
The first three cases use the same 24 parts arriving every six seconds, two-second serial inspection and a fixed 22-second external tray-exchange delay. Each clip compresses 180 simulated seconds into 30 seconds. Choose a case, then press Play.
One four-pocket tray. The cell waits while the full tray is exchanged.
Recorded scenario, not a live connection to physical equipment. The queue count includes parts waiting upstream of the four-position conveyor.
- Simulated time
- 0.0 s
- Completed
- 0
- Waiting to enter the station
- 1
Waiting for playback
12 of 24 parts completed
66 seconds waiting for a tray
What changes the result?
| Case | Time to clear all 24 | Completed / 24 at 180 s | Waiting for tray | Peak waiting queue |
|---|---|---|---|---|
| One tray | 330.8 s | 12 | 66 s | 13 |
| Faster robot | 278.96 s | 16 | 67.36 s | 12 |
| Two tray drawers | 220.8 s | 19 | 0 s | 8 |
| Longer inspection | 364.8 s | 11 | 0 s | 14 |
| Bursty arrivals | 220.8 s | 19 | 0 s | 12 |
The second drawer lets the robot continue while the other tray is exchanged. Faster robot motion also improves output here, but the single tray still interrupts production. These outcomes apply to this control policy and the declared durations. Zero tray waiting in the two-drawer case is a result of this nominal run, not a guarantee under variable exchange times.
Bursty arrivals increase the peak queue without changing completed output in this particular run. Extra storage can absorb a burst; it does not increase the inspection station’s processing capacity.
Timing assumptions and scope
Each part uses two serial transfers, each assumed to take 3.6 seconds, plus inspection. The shorter-transfer case scales each transfer to 2.52 seconds; this is 30% less time, not a verified controller setting. Four pockets fit each tray. Transfer and exchange durations are illustrative inputs, without measured equipment or vendor-controller calibration.
The ordinary arrivals end at 138 seconds. Six-second arrivals are intentionally faster than the nominal 9.2-second processing cadence, so a queue grows. The upstream queue is a logical waiting buffer without a capacity limit; the replay shows four conveyor positions. This study does not establish a cell sized to meet that incoming rate.
The baseline waits for a tray before loading the nest. Preparing the next part during exchange and larger feasible trays are alternatives requiring their own layout and sequence checks; this comparison does not establish the lowest-cost design. Concurrent tray exchange assumes a suitable segregated, interlocked arrangement, which is not implemented here.
Check the mechanism behind the motion.
The component trial uses gravity, a sensor-stopped conveyor, a finite-stroke stop, finite-force jaw drives and workpieces that remain dynamic during pickup and release. Fresh normal and shorter-transfer four-part trials pass a 1 mm placement criterion and a 0.3-second settling check with position, tilt and velocity limits. A smaller time-step trial and a deliberately low-friction failure check are run separately.
Named geometric checks cover full drawer travel, the stop's swept volume, its separation from the end drum, the driven conveyor nose and selected mounting interfaces. These checks do not establish complete machine interference, structural strength or safety-system validation.
The robot geometry comes from official Epson GX8 CAD. The setting uses NVIDIA warehouse assets. The conveyor, camera mounting, fixture, gripper and drawers are custom concept assemblies.
Mass and inertia come from the modeled geometry and assumed material densities. Friction and servo settings are assumptions. Pickup uses perfect simulated pose feedback; a vision algorithm is not implemented. The four-part contact trial uses stationary trays and a Cartesian wrist-drive model; it does not identify the robot’s motors, structural compliance or vendor controller.
Inspection is a timed operation, not a demonstrated vision algorithm. The three-minute comparisons are event-model studies with authored motion. Gripping, sensing, tray exchange, guarding and production rates require checks on the selected physical equipment.
Bring one automation decision to the model.
A layout, part drawing, cycle-time observations and a clear success criterion provide a starting point for comparing alternatives. Agree what the model needs to answer and what must be measured on equipment.