The Deliberative Level uses a symbolic world model to plan high level actions to achieve its goals. It responds to requests from the Sequencing Layer to perform deliberative computations. This may include computing when to perform actions, or it may not (which was the case in early implementations). It returns commands to the Sequencing Layer by placing them on a task queue.
Even if the Sequencer requests the Deliberator to compute when actions should be performed, the Sequencer is at liberty to ignore the result and perform differnt actions at different times. So in essence, the Deliberator is controlled by the Sequencer.
Because it is not responsible for the second-to-second low-level responses it has more time to plan. Also, it may plan at a higher level than the specification of all low-level responses.