Simulation-based procedure for robotic cell layout optimisation

Robotic systems are a cornerstone of Industry 4.0, with adoption steadily increasing due to cost reductions and performance improvements. As these systems continuously interact with other elements in their surroundings, their performance strongly depends on parameters such as the pose, or shape of these components. Despite its importance, robotic system design often relies on manual methods or simplified simulations that yield suboptimal results. To address this gap, we present an optimiser-agnostic standardised simulation-based workflow for robotic system design optimisation. Within this workflow, we introduce two contributions: the use of high-fidelity simulations for motion-feasibility evaluation, ensuring that solutions remain valid with respect to detailed simulation-based collision constraints, and a fast hybrid two-step collision-checking approach where simulation-generated data is used to train a machine-learning-based self-collision classifier that reduces the required computational cost. The contributions are validated through two case studies: a mobile manipulator designed to unload a truck container filled with boxes, and a robotic cell integrating a horizontal lathe that must be loaded and unloaded with metallic cylinders, with input and output conveyors supplying and receiving the parts. Results demonstrate the existence of optimal poses for the elements in the scene. The computational benefit of the proposed collision checker is demonstrated.