Artificial Intelligence
4 Instructions for Report
The report is in two parts, which will be submitted as part of the same document, but must be separated by the headings Part I and Part II, as these will be marked independently of each other. You should give your report a sensible title that identifies the domain you are intent on creating.
4.1 Part I: Designing a Planning Domain (14% Overall Module Mark)
(Maximum length 2 pages)
The first part of this coursework is about modelling a planning domain. You should devise an interesting problem that would seem prudent to address with Artificial Intelligence planning, and then model this problem using the Planning Domain Description Language (PDDL). For this segment of the report, you should:
Give an Introduction and motivation for your domain, an overview of what you are doing and what is inter- esting about it.
Include a Discussion of how creative, meaningful and unique your proposed domain is. There has to be some justification of why this is a domain for which planning may prove useful. Merely speculative works or domains that have little to no justification of why they were selected will struggle to receive marks. Furthermore, domains that are in essence those seen in class (or in other planning resources such as the IPC benchmarks) with the labels changed will score poorly.
• Note: Remember that copying existing work for academic credit is considered an academic o↵ense of plagiarism.
• You can take inspiration from existing domains encoded in PDDL, including both examples found on KEATS and other sources on the web, but you must clearly state how you modified and extended the domain for the purposes of this assignment.
Assess how Rich your domain is by discussing the range of information and actions available within it. A domain that involves just one type of action is not particularly interesting. While we do not require a line-by-line breakdown, you should take time to explain some of the more interesting design choices of your domain. This might relate to specific types, predicates or action conditions/e↵ects.
Identify which PDDL features are used in order to realise your domain model. This can include typing, metrics and durative actions. Do not simply add metrics and temporal features to your domain without justification. There is sometimes an expectation that by adding more of these features you will inherently earn more marks. But this all comes back to the domain itself. Does it make sense for these features to be employed? Have you given justification for their necessity in the report? It is the justification of these features that will translate into a good grade. Conversely, adding these features without any valid justification will work against you.
4.2
Part II: Experimental Analysis (14% Overall Module Mark)
(Maximum length 2 pages)
For the second task in this report, you will use an AI planner to test the planning domain designed in Part 1. You will take the domain you designed, build a set of problems for it and then test against the provided planners and discuss the results of the experimental analysis. For this segment you should:
Discuss the e cacy of your problem files designed to work with the domain file from Part I. It is important that you provide a variety of problem files that show the depth and scope of the domain you have designed.
• Note: Do not simply submit the domain with one or two problem files. Instead, show that you have really tested the range of possible problems you can create.
• Given the domain model you have established, do the problems su ciently represent it? Or are they merely a highlight of possible problems you could solve?
Prove those problems are solvable by running them against the domain file using the provided planner. This will enable you to understand whether a problem file can be solved successfully by the planner or if the problem (and potentially the domain file) needs work to perform e↵ectively.
Identify valid but insolvable problems that you have created. Can you theorise why the planner may face challenges in exploring the search space? Can you explain what features of your domain make it di cult for the planner to find a solution – or potentially never find one despite it being possible?
3