Underwater 3D Perception for Robotic Manipulation
An advanced underwater ROV system is designed to facilitate robotic arm control for underwater interactions, such as grasping objects. The system relies on a stereo vision setup to collect environmental data, dynamically construct 3D scenes, and identify target objects within the generated environment. Based on the reconstructed 3D scene, the system determines optimal grasping points for precise and efficient manipulation.
However, underwater vision models often suffer from performance degradation and poor generalization due to challenges such as light attenuation, scattering, and turbidity Additionally, real-time control constraints impose further limitations on computational resources and system responsiveness.
The goal of this project is to enhance the robustness and reliability of underwater visual perception by improving 3D scene reconstruction and object recognition under challenging aquatic conditions. Ultimately, this research aims to advance underwater ROV technology by providing a more stable and effective solution for underwater robotic manipulation, contributing to its broader application in exploration, maintenance, and scientific research.