Underwater cave search and entry using a robotic fish with embedded vision

Underwater cave exploration has always been a tough problem for autonomous underwater vehicles. As a primary step towards this problem, we explore the underwater cave search and entry with a free-swimming robotic fish characterized by flexible control and embedded vision. In particular, we use a colored ring instead of the cave and acquire its contour by its binary image. Then we propose an algorithm combining the Meanshift algorithm with the bump characteristic of the ring contour to eliminate the mirror image effects and background interference. Eventually a bio-inspired Central Pattern Generator control method is adopted to smoothly drive the robotic fish to swim towards the ring. All proposed algorithms are implemented in real time with a hybrid control system consisting of two embedded microprocessors (TI DM3730 and STMicroelectronics STM32F407). Preliminary aquatic experiments demonstrate that a fairly good exploration effect is resulted and the interference caused by mirror image effect and background is largely eliminated. The proposed robotic fish-based scheme offers an alternative to cave exploration in complex aquatic environments.