CS470/570 Intro to Artificial Intelligence Chapter 4: Beyond Classical Search Written Homework

Instructions: Answers to the questions below must be presented in hardcopy, on the due date noted on the course website. All submissions must:

• have solutions presented in the order listed here, and clearly labeled.
• A stapled submission with a neatly printing cover sheet is required.

Typed answers are preferable, but clear handwritten work is acceptable.

The Problems:

4.1 a-e. This is all about understanding the actual search behaviors of all of the various searches that we've talked about in Ch3 and Ch4. If you understand these, you can do reasoning like "Hmmm, well search-algoXX is really just based on a the more basic search-algoYY, but with the addition/modification of a, b, c." This question asks you to do something similar to this: think about the search behavior that would result when making the given tweaks to the given algos, then formulate that as "This would sort of like <algo name> search configured so that <whatever settings/assumptions>, in that <explain your reasoning, i.e., why the two search behaviors would be the same>". The focus is less on naming the algo it's like, but on your characterization of search behavior and how it could be seen as similar to some other one.

Detail: For 4.1c, you would obviously assumed that setting T=0 doesn't immediately terminate the algorithm, as would typically happen once Simulated Annealing's T value reaches zero. So assume it just keeps searching with T=0.

4.7. Searching in belief spaces. This question makes you think clearly about searching within believe state spaces, i.e., that it's really just like searching in physical spaces...only you have to think a bit differently about heuristics. You'll need to review our discussion of A* heuristics and their admissibility in order to translate that to this search in a simple belief space. Clarification: The last piece of the problem state (starting with 'Comment on the accuracy...") means "tell me what your heuristic would compute for each state, and then the order of exploration that would result in for your A*.

4.10. Defining belief state spaces for non-deterministic behaviors. This problem explores the interaction of non-determinism and belief state search, asking you to draw out the belief states when the "erratic vacuum cleaner" (section 4.3.1) is operated in a sensorless worlds, and must therefore plan its actions via belief state search. You are also asked to think about what it means for a problem to be solveable in such contexts. After you explain the unsolveability, add this: what is the minimum modification to the world that would be needed to be able to produce viable solutions?

4.12. Parts b,c. Contingencies plans in online search. Part (a), which you are NOT doing, ultimately illustrates that, without interleaving percepts, the belief state space is *enormous*...roughly 2^2^12. So what about when doing online search where I'm developing a contingency plan to operated based on all possible percepts that I might see as I move? When it says "describe the first few branches" it means: draw out the first two levels of the belief space possible based on percepts received at each step, starting at the states possible in the initial state (i.e. from part(b) ), when you turn on the robot in position (1,1).