The following revisions have been made to my paper:

1. There is an extensive discussion of what is an action and on the
role of preconditions of actions.

2. There is a new "representational methodology" that specifies how
one should go about thinking about a domain to represent it in the
ICL. (This discussed both frame and ramification algorithms).

3. There is more discussion about the relationship to POMDPs, Bayesian
networks and to abductive formulations of sensing and action.

4. There is a detailed comparison with the work by Bachus, Halpern and
Levesque.

5. The paper has had some minor changes to make it more readable.

David