How it works
One panel,
start to finish.
Below is the real output of the control software running a single panel wrap end to end. Every figure was produced by executing the code. None was written by a person, and none was rounded to look better.
The run, stage by stage
Job AW-2026-DEMO · panel DOOR_FL · material AW-114 from roll ROLL-001 · robot SIM-ROBOT-001 (SIMULATED).
Two clocks appear here. The cycle time above is the modelled duration of the job. The column below is compute time — how long the software itself took to produce the run, which is wall-clock and varies between executions. Neither is a measurement of a physical robot.
| Compute time | Stage | What it does | Recorded |
|---|---|---|---|
| +0.000s | SCAN | Take the panel geometry. | source=SIMULATED_KNOWN_GEOMETRY, is_perception=False, coverage=1.0 |
| +0.000s | DIGITAL_TWIN | Build the vehicle's digital twin. | twin_version=1 |
| +0.000s | PANEL_IDENTIFIED | Identify the panel and compute its area. | panel_id=DOOR_FL, area_m2=0.434 |
| +0.000s | MATERIAL_RESERVED | Reserve film from a specific roll. | roll_id=ROLL-001, required_m=0.8 |
| +0.000s | WORKCELL_RESERVED | Reserve a cell and a robot. | workcell_id=WC-01, robot_id=SIM-ROBOT-001 |
| +0.000s | INTERLOCKS | Check safety interlocks before anything moves. | satisfied=True |
| +2.886s | WRAP_PLAN | Generate the trajectory. | waypoints=18281 |
| +2.886s | MATERIAL_CONSUMED | Record film actually used against film planned. | actual_m=0.824, planned_m=0.8, variance_m=0.024 |
| +2.886s | INSPECTION | Inspect, and correct what is found. | defects=1, corrections=1, passed=True |
| +2.886s | PANEL_INSTALLED | Mark the panel done. | panel_id=DOOR_FL |
| +2.886s | FINAL_SCAN | Re-scan to confirm. | source=SIMULATED_KNOWN_GEOMETRY, is_perception=False |
| +2.891s | JOB_COMPLETE | Close the job and write the record. | — |
What the run recorded
0.824 m used
Against 0.8 m planned — a variance of 0.024 m, recorded rather than absorbed. Film is the consumable that decides job margin, so the difference between planned and actual is tracked per job and per roll.
1 defect, 1 correction
wrinkle at 1.235,-0.221 — found during the operation and corrected within it, not flagged afterwards. Final status PASS.
Why not an industrial arm and a camera?
The usual first question, and a fair one. The distinction is not marketing — it is what the granted claims actually recite. Each row cites the element it comes from.
| Dimension | Conventional cell | What the claims recite | Element |
|---|---|---|---|
| Robot platform | One arm, bolted down, programmed for that cell. | An interface module to humanoid, articulated or mobile platforms, operable in fixed workspace AND remote mobile deployments without owning, modifying or customising the robot hardware. | '243 claim 1(b), 1(g) |
| Tooling | One end-effector per cell. Changing the operation means re-tooling and re-programming. | An automated tool magazine of interchangeable heads, selected autonomously, switched mid-process. | '243 claim 1(d), claim 3 |
| The tool heads themselves | Generic grippers. Film work is done by hand because the motion is the skill. | Heat-controlled roller, PPF applicator with heat and pressure sensors, motorised film tensioner, knifeless tape, and a vibration-assisted squeegee that reproduces the 10 and 2 hand angle. | '243 claim 1(e) |
| Vision | A camera for pick-and-place. Fixtured parts in known positions. | A 3D vision system of stereo cameras AND laser grid projectors generating high-resolution surface models of a curved, variable body. | '243 claim 1(c) |
| Defects | Detected at the end, by a person, after the fact. | Detected during the operation, with real-time corrective action and no human intervention. | '243 claim 1(f) |
| Learning | Re-programmed by an integrator when the vehicle changes. | Manipulation sequences learned from expert demonstration, with generative-AI synthetic training data and a closed-loop feedback mechanism. | '240 claim 1(b)(ii), claim 3 |