Abstract
Active matter systems consist of self-driven units that convert energy into motion, producing diverse non-equilibrium behaviors such as defect dynamics, spontaneous pattern formation, and chaotic flows. Unlike passive materials, active matter generates autonomous movement, offering potential for novel functional materials with applications in fields like biotechnology and robotics. However, achieving control over these systems remains challenging due to their inherent instability and turbulence.
This thesis develops two complementary approaches to address these challenges: optimal control theory and model-free reinforcement learning (RL). Using the Toner-Tu model for polar active fluids, we demonstrate that adjusting self-propulsion speed can steer the system into desired states, such as relocating asters, reorienting waves, and dynamically switching between stationary and traveling patterns. We also demonstrate the controllability of active nematic fluid by stabilizing states such as Couette flow, coherent flow in bulk, and defect-free subdomains, which are otherwise inaccessible without control. These findings highlight how model-based control can reveal generic principles for controlling active fluid behaviors.
In parallel, we introduce an RL framework to control active nematics, systems prone to chaotic defect proliferation. Unlike traditional methods, the RL framework does not require a detailed physical model, making it more robust to noise and experimental uncertainties. The RL-based controller autonomously discovers spatiotemporal sequences that stabilize coherent flows, suppress defects, and reconfigure the system between different states.
Our findings provide a comprehensive roadmap for designing programmable active materials with precise spatiotemporal control. By bridging theoretical control methods with practical implementations, this work advances the understanding and manipulation of non-equilibrium systems, paving the way for future innovations in active matter research.