A Planning Method for Design-Space Exploration of Multi-Robot Assembly Cells

Abstract

Designing a multi-robot assembly cell requires evaluating many candidate structural configurations. These configurations differ in robot set, robot placement, tooling, and the layout of task-relevant resources. Their performance is measured by the makespan of the best assembly procedure they admit. Procedure generation requires jointly solving task-to-robot assignment, sequencing, scheduling, and collision-aware motion planning, making the evaluation computationally expensive. Existing methods target single fixed configurations and do not support efficient repeated evaluation across design alternatives. This paper presents an assembly-specific planning method for that purpose. Assembly tasks are decomposed into discretized motion primitives with precomputed pairwise collision relations. For each candidate design, executable procedures are generated under collision-free constraints and evaluated by makespan. A structural change invalidates only the directly affected primitives, enabling selective reuse of the collision memory across structurally different cell configurations. The method achieves makespans competitive with reference methods while substantially reducing per-configuration cost. A sweep of 3675 candidate component layouts is completed in 6.3 hours, indicating the method's suitability for large-scale design-space exploration.