Integrated Stereo Vision and Compliance Control With an Underwater Manipulator for Hydraulic Structure Inspection and Maintenance

Hydraulic structure inspection and maintenance are essential for structural health and operational safety. To address the limitations of underwater intervention, this article introduces an integrated underwater manipulator system that supports both teleoperation and automatic operation modes. The automatic mode incorporates visual position estimation for locating structures, robust motion capability for accurate trajectory tracking, and compliance and force tracking for safe interaction. An uncertainty-based underwater disparity estimation network (UWNet) is developed to enhance position estimation accuracy. This network combines global and local features effectively and integrates uncertainty estimation and pseudolabel generation to adapt the pretrained UWNet, improving underwater disparity predictions. In addition, a robust controller is designed, comprising an inner-loop position controller based on modified unknown system dynamics estimator and supertwisting sliding mode control to mitigate hydrodynamic disturbances and model uncertainties. An outer-loop variable admittance controller with adaptive feedback compensation ensures compliant interaction and force tracking. Experimental results show that, with reliable disparity estimation from the adapted UWNet, the system achieves a trajectory tracking RMSE of 1.44 mm at 1.0 m/s flow and a mean force error of 0.63 N during thickness measurement, while enabling stable surface cleaning under varying flow conditions, thereby facilitating hydraulic structure inspection and maintenance.