An asymmetric three-degree-of-freedom parallel mechanism is adopted in rehabilitation robots for assisting patients suffering from stroke or trauma in the hip. It is necessary to keep its kinematic singularity out of the workspace of human normal gait and increase the output power efficiency. Therefore, a novel method is proposed to optimize geometrical parameters of the mechanism. To describe the kinematic singularity in a better way, the improved force transmission indexes based on previous methods are proposed using the reciprocal product and mobility condition of the closed-loop mechanism. The indexes mainly represent the force transmission performance of unactuated parts of subchains and moving platform. Together with the driving force transmission indexes and geometrical constraints, the multiobjective optimization model is established. The differential evolution algorithm, which is widely applied to mechanism optimization, is used to achieve optimal results. The Jacobian matrix singularity and output power efficiency along giving trajectory before and after optimization are compared to verify the effectiveness of the method.