For RoboSub 2026, Stanford RoboSub is deploying two next-generation AUVs, Crush and Gill, featuring a completely overhauled, modular architecture. To maximize mission uptime, the mechanical design introduces a rapid-access sliding electronics chassis, while the electrical system utilizes a custom dual-battery management rail for hot-swappable power transitions. Driven by a fully agnostic ROS2 infrastructure, the software stack integrates a single-stage YOLOv8 perception pipeline with a dual-microcontroller acoustic architecture. To overcome underwater optical distortions, the team developed a synthetic data augmentation framework using JTA-GAN style transfer, ensuring robust perception generalization across a comprehensive, tiered testing cycle.