In [16] a two-layered approach was proposed to compute a winning strategy for the game of billiards. AI tools as well as robust optimization routines for noisy environments were combined to plan the sequence of shots. We complete here the modelling by introducing significant developments for the high-level planner which guides the precise optimal controller to generate a plan given any random initial state. We will first resume the general model for this particular class of problems and then propose several domain-specific heuristics to guide our search and render the problem tractable. Several improvements to the optimal robust controller including refinements in the objective function will also be presented in order to improve single-shot optimization. Results are presented demonstrating the full potential of the methods proposed making it the state of the art in regards to the game of billiards.