MicroRaptor will not be running a fully generalized learning engine, at least not at first. In order to test walking and navigation, I will write a simple hierarchical goal system that will allow the robot to accomplish interesting tasks.
A leaf-level goal is simply (typically) a navigation destination and one or more sensor queries to assist that navigation. Other types of goals might include visual recognitions tasks, or manipulation tasks that attempt to change something in the robot's environment.
Goals, like everything else, will be stored in long-term memory, and will be accessible anytime by just instantiating a goal engine in a free slot in short-term memory.