Centroidal Dynamics-based Posture Control for Inverted Ship-hull Walking of a Radial Underwater Quadruped Robot

This paper proposes a centroidal dynamics-based posture control algorithm for inverted ship-hull walking of a radial underwater quadruped robot with magnetic feet. The robot body is modeled as a single rigid body, and a quadratic programming (QP) optimization is formulated to distribute contact forces to the stance legs. Bidirectional normal force constraints are introduced so that compressive and tensile forces are naturally determined by external force conditions, enabling both upright and inverted walking within a unified framework without mode switching. Buoyancy is incorporated into the dynamics equation, and hydrodynamic drag compensation is applied as a feedforward term in the reference acceleration. Acceleration and jerk limits are also applied to suppress abrupt changes in contact forces and joint torques. The proposed algorithm is validated through Gazebo-based underwater simulations using the hull model of the FFG-821 Seoul-ham frigate. Results show that smoothing reduced the average joint torque by 23.02% and roll/pitch angular velocity root mean square (RMS) by 38.31%, and stable inverted walking was achieved under a current velocity of 0.3 m/s.