<p>For decades, the use of fish traps was the gold standard for investigation on fish migration. The availability of high-resolution digital camera systems has laid the foundation for digital fish counting/monitoring. Using observation windows in the fish migration aid, migrating fish can be captured and recorded without contact and therefore in a way that is particularly gentle on the fish. However, the water-wetted areas are subject to constant natural pollution. The build-up of biofilm on the water side of the observation windows, the rear wall and on the bottom of the capturing tunnels creates a coating that negatively affects the quality of the recorded images. However, a high image quality is advantageous for analysing the images and video sequences to ensure reliable identification and counting. Especially when using software-supported evaluation and AI, the best possible image quality is of great value in order to minimise false hits. This article describes the efforts to develop a cleaning robot for a highly automated fish monitoring system in the research tunnels at the new fish migration aid at the Drau power plant Feistritz-Ludmannsdorf. The division of the problem into the two tasks of cleaning and movement and the subsequent research into solutions in similar areas led to rapid success.</p>