PDE-Constrained MPC of Motility-Induced Phase Separation in Robotic Swarms: The Role of Model Fidelity and Local Information

Motility-induced phase separation enables swarm aggregation to be regulated without prescribing a target spatial distribution. This paper develops a partial differential equation-constrained model predictive control framework for a vibration-driven robotic swarm. Three continuum models retaining progressively less orientation information are considered to study the trade-offs among model fidelity, information requirements, and computational cost. The framework is evaluated under centralized and decentralized settings, with focus on how limited sensing and local-to-global phase-separation estimation affect the performance of predictive control.