Aimed　to　achieve　quantitative　control　of　workpiece　surface　after　robotic　polishing　and　improve　polishing　efficiency,　a　two-step　processing　optimization　method　involves　artificial　intelligence　algorithms　is　investigated.　Firstly,　based　on　XGBoost　algorithm,　a　prediction　model　for　polished　workpiece　surface　depending　on　key　parameters　is　proposed,　and　the　accuracy　of　the　model　is　verified　by　experiments.　After　that,　by　using　the　above　model,　the　influence　of　each　parameter　on　the　roughness　was　evaluated　quantitatively.　Subsequently,　target　roughness-driven　optimization　of　processing　parameters　was　presented　by　combining　the　roughness　prediction　model　with　NSGA　II-TOPSIS　algorithm　based　on　the　influence　of　each　parameter　on　the　roughness.　To　verify　the　proposed　processing　optimization　method,　polishing　experiments　of　mold　steel　samples　were　conducted.　The　experimental　results　show　that　the　maximum　absolute　error　between　the　predicted　and　experimental　roughness　is　0.035　mu　m,　and　the　maximum　relative　error　is　＜9%.　At　the　same　time,　when　the　minimum　is　set　as　the　optimization　objective.　With　the　same　length　of　polishing　path,　the　feed　rate　is　increased　from　0.25　mm/s　to　0.37　mm/s,　and　the　efficiency　is　improved　to　48%.　The　NSGA　II-TOPSIS　algorithm　can　achieve　quantitative　control　of　mold　steel　surface　roughness　after　robotic　polishing　to　improve　polishing　efficiency,　and　provide　a　basis　for　reasonable　selection　of　processing　parameters,　which　have　certain　practical　value.