Learning Symbolic Knowledge from Execution

		Karen Zita Haigh

I will describe my recent work on using Xavier's real execution
traces to extract relevant symbolic information for later use 
at planning time. In particular, I will describe the content
of the execution trace, and point out the differences between
that representation and the representations used by the path
planner and the task planner. I will talk about how I extract
learning examples from the execution traces, and how I correlate
them to easily-detectable features of the environment, and how I
then create situation-dependent rules which can be used to guide
the planning systems.