In this chapter we present the DEFRAS architecture that was used in some of the ARS experiments, based on the integration of task simulation and execution. While in the previous chapters, simulation was used to verify the correctness of a plan or to generate data for machine learning, in Sect. 2 simulation is used as a nominal trace of the task execution to verify that the assumptions about the anatomy and the environment are correct. However, this approach highlighted that textbook knowledge and logical planning are not sufficient to deal with the uncertain and unpredictable anatomical environments. Surgeons in fact rely on their experience to interpret the textbook indications and to adapt them to the specific situations. In Sect. 3 we analyze what it would take to add common sense reasoning to an autonomous surgical robot.