Develop, experiment with, and present a planning project evaluated on a Pass/Fail basis. The project requires defining a main classical planning model in PDDL and extending it with at least 2 advanced variants selected from the following topics:

• Probabilistic Planning (RDDL)

• Diverse Planning (generating sets of alternative plans)

• Temporal Planning (durative actions and temporal constraints)

• Numeric Planning (numeric fluents and resource allocation)

• GenAI for Planning (using Generative AI / LLMs for domain generation, plan synthesis, or heuristic support)

1. Domain Selection & Main Classical PDDL Modelling

   1. Select a realistic problem domain of comparable complexity to those discussed in lectures.

   2. Specify the planning domain and at least three planning problems in PDDL of increasing complexity, relying on standard STRIPS or extended features (ADL).

2. PDDL Solvers & Experimental Analysis

   1. Solve the core specification using standard PDDL planners from International Planning Competitions (e.g., Fast Downward, ENHSP, or planning.domains).

   2. Test your formulation using at least two different search heuristics studied in the course. If switching heuristics forces a downgrade from ADL features to a STRIPS-like formulation, document and explain this transition.

3. Variant Extension (Choose at least 2)

   1. Probabilistic Planning: Formulate the domain in RDDL to capture stochastic transition dynamics or probabilistic outcomes, evaluating it with compatible RDDL tools/solvers.

   2. Diverse Planning: Configure a planner or pipeline to produce a set of qualitatively different plans for the same instance, analyzing the diversity metrics and trade-offs.

   3. Temporal Planning: Extend the domain to include durative actions, action concurrency, and continuous time bounds using temporal PDDL features.

   4. Numeric Planning: Integrate numeric state variables, continuous resources, or numeric action effects, using a planner capable of numerical reasoning.

   5. GenAI: Use Generative AI / LLM tools to generate, repair, or translate domain/problem formulations, or to assist with planning heuristics. Critically evaluate the correctness, failure modes, and efficiency of the AI-generated artifacts versus manually authored ones.

4. Deliverables & Submission

   1. Prepare a slide deck covering the domain design, the PDDL experiments, and the implementation/analysis of your chosen variant(s).

   2. Submit all project material (code, domain/problem files, experimental results, and slides) to the dedicated Google Classroom assignment opened for your intended exam session.

5. Presentation

   1. Present the project live during an exam session.

   2. For each exam session, project dates will be communicated in due time.

Rules

1. Group Composition: Projects must be carried out in groups of min 2 – max 3 students.

2. Assessment: The project is evaluated as Pass/Fail.

3. No Pre-Approval Needed: There is no formal proposal submission or approval waiting period. Ensure your domain choice complies with the complexity guidelines and requirements outlined above.

4. Presentation Format: Presentations (10 minutes per group) take place on the same day as the written exam session (unless otherwise specified). Every group member must participate and present a part of the work.

5. Model Adaptations: Because different planning formalisms and variants serve different purposes and offer varying expressiveness, your main classical PDDL domain may need to be adapted for the selected variant(s). All structural differences or design trade-offs must be explicitly justified during the presentation.

Some student projects 2024-25 (PDDL + IndiGolog)

Some previous student projects (related only to PDDL planning)