When the simulation runs, the simulator sends a percept to the agent, which is supposed to return an action. This goes on until the simulation is done or something goes wrong.
