Presentation description

This project models the MOAB microfluidic valve network as a FrozenLake-inspired grid environment, where user-defined fluid routes are learned by a Q-learning agent navigating from source to destination valve. Stochastically injected faults act as frozen holes blocking sample flow. Risk Factor based reward shaping penalizes high-bottleneck valves and rewards safer detours, enabling the agent to autonomously discover fault-tolerant routes whit high efficiency and without explicit fault detection.