In this project, I designed a modular system to enable underwater robots to detect and retrieve black boxes autonomously. Using ROS2 and Unity simulation, I created an environment for testing. The system achieved real-time detection of ArUco markers for locating objects and used PD controllers for precise depth control. To handle complex decision-making, I implemented a behavior tree framework, allowing the robot to switch between manual and autonomous modes seamlessly. This project combined hardware and software integration for real-world applications.