Abstract: Parallel-actuated mechanisms in humanoid robots, inspired by designs like Cassie’s legs, offer reduced leg inertia and improved impact absorption, but they introduce closed-chain kinematics that require complicated control. Traditional strategies either approximate nonlinear transmission effects as fixed gear ratios or rely on fully centralized joint-space controllers that distribute motor commands, thereby risking performance loss or instability in the event of communication failures. This paper proposes a novel decentralized control framework for parallel mechanisms, in which impedance control laws are executed locally at each motor, ensuring that the resulting joint-space impedance matches the desired target. This consistency is achieved through a Jacobian-based mapping between joint-space impedance and motor-level commands. The proposed method is implemented on a two-degree-of-freedom (2-dof) parallel ankle mechanism with dual motors arranged in a closed-loop differential configuration. Experimental results demonstrate improved joint torque tracking, stable behavior, and robust disturbance rejection during disturbance-sensitive motions, including dynamic walking and squatting, compared to a conventional control-law conversion approach. These results indicate that the proposed framework enables more reliable and higher-performance control of parallel-actuated joints in bipedal humanoid robots.