This paper introduces a digital-twin-based system for foam cleaning robots in spent fuel pools, aiming to efficiently clean foam in spent fuel pools. The system adopts a four-layer architecture, including the physical entity layer, twin data layer, twin model layer, and application service layer. Initially, the robot was modeled in two dimensions, encompassing physical and kinematic aspects. Subsequently, data collection and fusion were carried out using laser radar and depth cameras, establishing a virtual model of the working scenario and mapping the physical entity to the digital twin model. Building upon this foundation, improvements were made in applying the full-coverage path planning algorithm by integrating a pure tracking algorithm, thereby enhancing the cleaning efficiency. Obstacle detection and localization were conducted using infrared and depth cameras positioned above the four corners of the spent fuel pool, with the digital twin platform transmitting coordinates to the robot for obstacle avoidance operations. Finally, comparative experiments were conducted on the robot’s full-coverage algorithm, along with simulation experiments on the robot’s position and motion direction. The experimental results indicated that this approach reduced the robot’s overall cleaning time and energy consumption. Furthermore, it enabled motion data detection for the digital twin robot, reducing the risk of collisions during the cleaning process and providing insights and directions for the intelligent development of foam cleaning robots.