<p>Robotic polishing and grinding processes are often limited by insufficient stiffness, which compromises machining accuracy and surface quality. Traditional stiffness indices fail to capture the task-related deformation of the end-effector under constant-force conditions. To address this issue, this study introduces a novel task-oriented stiffness evaluation metric, the Polishing Deformation Index (PDI), which integrates both contact forces and gravitational loads to provide a comprehensive measure of stiffness performance. To extend the analysis to entire trajectories, the Comprehensive Polishing Deformation Index (CPDI) is proposed as a path-level indicator. Based on the CPDI, a pose-optimisation model for tool is formulated, with the objective of minimising deformation while controlling its variation. Practical constraints, including joint limits, kinematic feasibility, manipulability, and trajectory smoothness, are incorporated to ensure industrial applicability. The optimisation is solved using a Distributed Adaptive Genetic Algorithm (DAGA). Both simulations and physical experiments validate the proposed approach. The results show that variations in the PDI are consistent with measured deformation trends, while CPDI-based optimisation reduces end-effector deformation by nearly 70% along two test paths. These findings confirm the effectiveness of the proposed framework and demonstrate its potential for improving machining accuracy in robotic grinding and polishing.</p>