Abstract: Quadruped manipulators offer promising capabilities for complex tasks in unstructured environments by combining terrain traversability with dexterous manipulation. However, motion control is challenging because arm-base dynamic coupling can destabilize locomotion and degrade end-effector tracking, particularly when the manipulator mass is non-negligible. Meanwhile, coordinated motion planning is essential to ensure consistent behavior of both subsystems. This paper presents HiLoMa, a novel hierarchical loco-manipulation control framework with coordinated planning and decoupled execution. At the high level, a kinematic Model Predictive Control (MPC) module generates coordinated motion references for the manipulator joints and the quadruped base. At the low level, the manipulator is controlled via inverse dynamics with Proportional-Derivative (PD) feedback, while the quadruped employs a Nonlinear Model Predictive Control (NMPC) cascaded with a Whole-Body Control (WBC). The arm-induced wrench is estimated from the manipulator centroidal momentum dynamics and incorporated into the quadruped controller, enabling anticipatory compensation for coupling effects during execution. Comprehensive simulation and hardware experiments validate the feasibility and effectiveness of the proposed framework, demonstrating improved tracking accuracy, enhanced locomotion stability, and successful execution of practical loco-manipulation tasks.

Key words: Quadruped manipulator')">Quadruped manipulator, Legged robot, Model Predictive Control, Hierarchical control, Bionic robot