Toward Optimal Tabletop Rearrangement with Multiple Manipulation Primitives