CS5811: Homework 3

Planning

Due: Monday, Oct. 20, 2003 (beginning of class.) (Assigned: Monday, Oct. 6, 2003.)

Reminders:
  1. Submit your programs and documents both using "submit" and as a hardcopy. (Please submit the non-program documents too, your hardcopy package should be the same as the online package). For the hardcopies of programs and outputs, I prefer enscript with two columns to save paper:
       (enscript -2Gr -P<printer-name> <file-name>).
  2. Include a description of how you perform the test runs so that I can duplicate your work. Avoid printing long and unreadable search traces though---keep it concise and to-the-point.
  3. This is an individual assignment. All the work should be the author's in accordance with the university academic integrity policies.

Description:

In this assignment, you will compare the performance of two planning systems:
  1. Vhpop. It is a partial order planner that can use plan graph based heuristics.

    The Vhpop system can be accessed at: /classes/cs5811/common/VHPOP-2.0/. Information is available : here. Note that the version in this site is 2.1, and is not as compatible with our system software as 2.0.

    Try the following to get started:

Warm-up:

Run the examples provided with each system so that I will know if there are any problems with permissions etc. Both programs are in C++, and you will not need to modify the code. I've compiled them only under Linux on wopr. If you wish to install them somewhere else, you may copy just the executables and the examples, or the entire package.

The problem (The Housekeeper):

You are tired of making and executing plans for household duties and you would like to have an automatic planner generate these plans for you. Before adopting the planner though, you would like to explore its capabilities and efficiency. To do this, carry out the following steps:
  1. Code the dinner-date problem that we worked on in class with non-parametric actions. Have each planner solve the problem.
  2. Switch to parametric actions and code at least 4 variations of the same problem. The new problems should be different in some aspect (length of the solution plan, number of objects in the domain, and so on). For instance, you can add Be sure to code several problems rather than just one for each dimension of variation.
Compare the performance of each planner to itself (3 versions for VHPOP and 3 versions of Blackbox) and to the other one. In other words, describe for a total of 6 kinds of planner runs, show each planner's performance changes as the problem gets more complicated. Also compare their performance on individual problems.

Note that this assignment includes both experiment design and experiment runs. Write a report that clearly describes the domains you coded for each problem set. Then, highlight the differences between each set of problems. Compare the performance of each planner. Comment on the results, and the kind of problems that are hard(easy) for each. Comment on the limitations on expressivity and functionality that you encountered while coding the problems.