The dark factory does not arrive overnight. It is a process of building the perceive-decide-execute loop and widening autonomy only as far as it has been verified. We do not promise an "unmanned factory" — we build, step by step, a factory that does not stop even with fewer people.
A dashboard shows and stops there, but an autonomous factory has to carry what it senses through decision and on to execution as one. VEXPLOR builds the loop of perceive (equipment and vision data) → decide (AI decision-making) → execute (direct equipment control) on the five CPS layers.
FIG — THE LOOP SIMPLY RUNSA volume-production metal forming factory. MES, digital twin, AI autonomous control, AAS and EDGE as one stack. AMR and EMS are the customer's existing equipment, connected for AI supervision.
STACKMES · UNITY 3D WEBGL TWIN (100MS) · AI AUTONOMOUS CONTROL · AAS(BASYX) · EDGEThis stack is not built for one particular factory; it is designed to be ported to a new site as it is. Extending the stack verified on a volume-production press process to different process types is the next step.
STACKMES + SCADA · DIGITAL TWIN/CPS · IOT — PORTING THE STACK VERIFIED AT THE AUTONOMOUS FACTORY DEMONSTRATION SITEThis is the last step of the dark factory journey. Foundation, integration, intelligence — a factory that has passed acceptance of the earlier steps begins the handover of control here.
AI only decides; people execute — the accuracy of its decisions is verified.
MODESHADOW — AI DECISION ONLY · DECISION AGREEMENT RATE VERIFIEDExecution after human approval. Handover of control takes place on AAF (L0~L4).
MODEHUMAN-IN-THE-LOOP — AAF L2, EXECUTION AFTER APPROVALAI executes within the verified scope. Commands that violate physical constraints are designed to be checked with SHACL before execution.
MODEBOUNDED AUTONOMY — AAF L3 · AI→PLC DEMONSTRATION [VERIFIED]Handover of control proceeds through five stages of verification: data consistency verification, which confirms that the collected data matches the shop floor; PoC performance verification, which verifies AI performance within a limited scope; pilot Shadow verification, which verifies AI decisions alongside people on a real line; whole-process loop verification, which confirms that the perceive-decide-execute loop is connected across the whole process; and accredited-body testing and standards conformity verification. Passing each one is the condition for moving to the next stage — from Shadow Mode to decision support, and then to autonomous control. Unmanned operating hours and the number of interventions are also reported as figures every month.
Control is not the only thing step-by-step autonomy leaves behind. If an expert's judgement stays only in that person's head, part of the factory leaves on the day they do. The records comparing AI decisions with expert decisions side by side in Shadow Mode, and the history of approvals, rejections and corrections, keep as data the judgement that until now was passed on only by word of mouth — an approach that hands judgement over to the system before the person leaves.
Handover of control takes place on the authority framework (AAF L0~L4) of VEXPLOR Core, and virtual commissioning and twin synchronisation on VEXPLOR Vision. The autonomous factory also starts with a diagnosis — a factory that has not been standardised cannot be made autonomous, so the VEXPLOR diagnosis is the first step. Many customers start alongside a government programme — we design it with you from the programme planning stage.
There are three cases. If you want a fully unmanned factory right now — we promise only step-by-step autonomy. If you want office automation only, with no manufacturing — our expertise is in manufacturing. If you are looking for a supplier of hardware such as robots or equipment — hardware is supplied by verified partners, and we take on integration and operation.
We review your current data and equipment and design the autonomy roadmap with you.