Digital Twin–Driven Multi-Objective Layout Optimisation of Flexible Robotic Cells with NSGA-II

Robotic cells in modern factories must be laid out for efficiency, flexibility, and fast adaptation to shifting product mixes [1]. Cycle time and safety hinge on where workstations, robot bases, stock bins, and tools sit, yet finding a balanced layout is challenging because static choices alter dynamic behaviour: moving a fixture a few centimetres can force a longer robot arc or an extra human step [3]. We present an automated layout-synthesis framework that couples a high-fidelity digital twin with an NSGA-II genetic algorithm. Each component is represented by its true 3-D shape, so the twin checks reach, clearance, and cycle time. For every chromosome—specified only by the positions and orientations of all cell elements—the twin returns two fitness values: (1) the cumulative distance from every object to the robot’s workspace envelope, ensuring collision-free reachability; and (2) the total distance the robot and the human operator must travel to execute the task sequence [4]. Human loading and unloading are simulated to capture uncertain human behavior [2]. Demand variability is handled by evaluating each layout under several mixes of product sizes; the robot’s best route depends on which items occupy each stock slot, so layouts are scored by average performance across weighted scenarios. Fast twin evaluation lets the GA explore hundreds of layouts over many generations. In an industrial pick-and-assemble study, the framework generated layouts that concurrently reduced robot travel and operator walking while maintaining throughput. The resulting Pareto front reveals a clear trade-off: compressing the robot’s motion envelope can lengthen operator’s task sequencing, showing the need to balance automation speed and human well-being. The method scales to multi-robot cells and, thanks to the live digital twin, supports adaptive re-layout triggered by shop-floor data.