A real-time and effective double-loop feedback control system for underwater flexible manipulators is raised in this paper. The research object is a kind of underwater flexible manipulator driven by McKibben water hydraulic artificial muscle (WHAM) that can grasp, swallow, and disgorge target objects in its interior space. To make up for the lack of flexibility, an underwater flexible manipulator collaborative working strategy is proposed. A more flexible and smaller flexible manipulator is placed inside the flexible manipulator to assist it in performing difficult underwater works. The control system feeds back the position of internal objects through a built-in binocular camera and the working state of the manipulator through displacement sensors. The control system setups including underwater flexible manipulator subsystem, hydraulic drive subsystem, PLC control subsystem, displacement sensor subsystem, built-in binocular vision subsystem, and upper computer subsystem is built. PYTHON-based built-in binocular vision software and C++-based underwater flexible manipulator control software are also developed to facilitate observation and recording. The underwater flexible manipulator collaborative experiment is designed to verify the performance of the control system and the control algorithm.