CS 660: Topics in Artificial Intelligence: Planning Algorithms

Spring 2025: Mondays, Wednesdays, and Fridays from 2:00 PM - 2:50 PM in F. Paul Anderson Tower, Room 255

Contents

• Course Description
• Student Learning Outcomes
• Instructor
• Prerequisites
• Text
• Grading
• Policies
• Academic Integrity
• University of Kentucky Academic Policies
• Calendar

Jump to today's date on the calendar.

Course Description

This is a seminar-style course for computer science graduate students interested in planning, a branch of artificial intelligence research. Planning is thinking before acting. Given a (1) description of the world in some formal logic, (2) a set of actions with preconditions and effects, and (3) a goal, a planning algorithm's task is to find a sequence of those actions which achieves that goal. Planning is computationally intractable, but the research community has developed a number of interesting techniques to mitigate this high cost and solve an impressive array of problems. Throughout this course, students will read important research papers in the field as well as implement and test historically important planning algorithms with guidance from the instructor. The course concludes with a self-defined project that allows students to explore a planning topic of interest.

Topics covered will include:

• Partial order planning
• Causal links in planning
• Plan graphs
• Planning as satisfiability
• State-space planning
• Heuristic design
• Width-based planning

Student Learning Outcomes

After taking this class, it is expected that:

• Students will gain a deep understanding of planning problems, including their computational complexity, history, and various approaches to solving them.
• Students will understand the tradeoffs inherent in designing search algorithms for hard AI problems like planning.
• Students will increase their proficiency in reading, synthesizing, and presenting graduate-level research from top conference proceedings and journals.
• Students will gain practical experiencing in designing and programming scientific software for solving hard AI problems like planning.

Instructor

Prof. Stephen G. Ware, Ph.D.
Department of Computer Science
Davis Marksbury Building, Room 307
sgware@cs.uky.edu
Office Hours: Monday, Wednesday, and Thursday from 4:00 PM to 5:00 PM

Prerequisites

It is assumed that:

• You have completed CS 463G: Introduction to Artificial Intelligence or some equivalent course.
• You are a competent programmer with a solid foundation in data structures and algorithms.

Text

There is no required text for this course. The reading material will be research conference proceedings and journal articles provided to students as PDF files.

Grading

Grades for this class will be calculated based on the following assignments:

Assignment	Weight	Percentage
30 daily paper summaries	1% each	30% total
1 to 5 research paper presentations	varies	25% total
3 planning algorithm programming projects	15% each (1 dropped)	30% total
1 final project self-direct project	15%	15% total

There will be no exams in this course.

Readings and Paper Summaries

This is a seminar-style course, which means that most class meetings will be a discussion of an academic paper on planning algorithms. Critical discussions at the beginning of each unit will be led by the instructor, but other papers in each unit will be presented by a member of the class. Each person in the class will present between 1 and 5 papers over the course of the semester, depending on the number of people enrolled. Presentations will be graded according to this rubric.

Whether or not they are presenting, each person should read the day's paper before class. Each person must submit a short (about half a page) summary of the paper at the start of class. Each summary is worth 1% of the final grade, and they double as a measure of course attendance.

Programming Projects

This class is divided into several units. In units two, three, and four, we will discuss a related family of planning algorithms. By the end of each unit, each student must implement one of the algorithms in that family. The instructor will provide a library of helpful tools that for reading and processing benchmark planning problems; students will be responsible for filling in the "meat" of the algorithms--the interesting parts that affect their performance.

The instructor will provide a pre-configured IDE that includes a compiler, pre-written code libraries, and benchmark planning problems. By working in a common codebase, students will be able to make fair comparisons between their algorithms.

Each student's lowest programming project grade will be dropped at the end of the semester. (This policy does not apply to the final project.)

Click here to read the full details of each programming project.

The final project will be self-defined, and it will allow students to explore a topic of interest within planning algorithms. This project will require an initial proposal and a final presentation to communicate the results.

Policies

These policies are in place to maintain professionalism and mutual respect:

• Paper Summaries: Should be typed and submitted to Canvas before the start of each class.
• Attendance: Attendance is not required, but this is a seminar-style course, so in-class discussions are key. If a student is going to miss class, they should still submit the paper summary for that day.
• Due Dates: Programming projects are due on Canvas promptly at midnight on the day they are due. Submissions may still be submitted until 3 AM the morning after the deadline. Consider these three hours an unofficial grace period. At 3:01 AM, your assignment is not 1 minute late; it is 3 hours and 1 minute late, and it will not be accepted.
• Late Work: Late projects and paper summaries will not be accepted without a serious and documented excuse.
• Missed Presentations: If you need to miss an in-class presentation that you are scheduled to give, you must notify your instructor at least one week in advance to schedule a makeup. Missed presentations cannot be made up without a serious, documented excuse.
• E-mail Response Time: Please allow a minimum of 24 hours for e-mail responses.

Academic Integrity

All students are expected to follow UK's code of academic integrity. Violations of this policy will be dealt with according to UK's policies.

Please be aware that copying code from another student or from an online repository such as GitHub for a programming project is cheating and will be dealt with as such. Please do not post solutions you develop for this class to public repositories where others could copy it.

University of Kentucky Academic Policies

This course abides by the academic policies that all UK courses must follow. These policies include information on excused absences, accomodations for disabilities, and and Prep Week.

This course abides by UK's statement on diversity, equity, and inclusion.

Calendar

Introduction and Review

Topic	Assignments
Monday, January 13 Introduction and Syllabus	Definition of planning Class policies Project codebase and IDE
Wednesday, January 15 Uninformed Search	Breadth First Search Depth First Search Iterative Deepening Depth First Search Uniform Cost Search Bidirectional Search
Friday, January 17 Informed Search	Heuristics Greedy Search A* Search
Monday, January 20 Holiday	Dr. Martin Luther King Jr's Birthday
Wednesday, January 22 Logic	Propositional Logic Predicate Logic First Order Logic Boolean Logic Conjunctive and Disjunctive Normal Form
Friday, January 24 Logical Unification	Conjunctive and Disjunctive Normal Form Unification
Monday, January 27 GPS: The First Planner	Read GPS: The General Problem Solver by Newell, Shaw, and Simon. Focus: Understand what a planning problem is. What are the inputs and outputs to a planning algorithm? Presenter: Ware
Least Committment Planning
Wednesday, January 29 Partial Order Planning	Read The Non-Linear Nature of Plans by Sacerdoti. Focus: What is a total vs. partial order plan? Presenter: Damron
Friday, January 31 Causal Links in Plans	Read Systematic Non-Linear Planning by McAllester and Rosenblitt. Focus: What is a causal link and how does it ensure the correctness of a plan? When are they unsafe? Presenter: Ferry
Monday, February 3 Least Committment Planning	Read Introduction to Least-Commitment Planning by Weld, pages 27 - 43. Focus: The POP algorithm Presenter: Ware
Wednesday, February 5 The UCPOP Planner	Read UCPOP: a Sound, Complete, Partial-Order Planner for ADL by Penberthy and Weld. Focus: How does UCPOP handle conditional effects and quantifiers? Presenter: Ti
Friday, February 7 Optimizing Partial Order Planners	Read Accelerating Partial-Order Planners: Some Techniques for Effective Search Control and Pruning by Gerevini and Schubert (pages 95 - 112). Focus: What is a zero-committment choice and how are they helpful? Presenter: Butcher
Monday, February 10 Threat Resolution Strategies	Read Threat-Removal Strategies for Partial-Order Planning by Peot and Smith. Focus: How does the definition of a threatened causal link and order in which they are fixed effect efficiency? Presenter: Wu
Wednesday, February 12 Discussion and Review
Friday, February 14 Hierarchical Task Networks	Read UMCP: A Sound and Complete Procedure for Hierarchical Task-Network Planning by Erol, Hendler, and Nau. Focus: How is HTN planning different from classical planning, both in its knowledge representation and problem solving approach? Presenter: TBD Project 1 due on Canvas at midnight
Plan Graph Planning
Monday, February 17 Plan Graphs	Read Fast Planning Through Planning Graph Analysis by Blum and Furst. Focus: What is a plan graph and how does it approximate what is possible and when in a planning task? Presenter: Ware
Wednesday, February 19 Graphplan	Read Recent Advances in AI Planning by Weld, pages 93 - 102. Focus: The Graphplan algorithm Presenter: Ware
Friday, February 21 Boolean Satisfiability	Review SAT problems.
Monday, February 24 Planning as Satisfiability	Read Planning as Satisfiability by Kautz and Selman. Focus: How can a planning problem be translated into a SAT problem? Presenter: Racey
Wednesday, February 26 Planning as Satisfiability using Plan Graphs	Read Encoding Plans in Propositional Logic by Kautz, McAllester, and Selman. Focus: How can a plan graph be used to improve translating planning problems into SAT problems? Presenter: Butcher
Friday, February 28 Local Search on Plan Graphs	Read Planning through Stochastic Local Search and Temporal Action Graphs in LPG by Gerevini, Saetti, and Serina, pages 239 - 251. Focus: How does LPG use local search for solution extraction? Focus on the WalkSAT algorithm, and ignore details about other things like temporal actions and action costs. Presenter: Damron
Monday, March 3 Discussion and Review
Heuristic Planning
Wednesday, March 5 The HSP Planner	Read Planning as Heuristic Search by Bonet and Geffner. Focus: The HSP heuristic Presenter: Ware
Friday, March 7 The Fast Forward Planner	Read FF: The Fast-Forward Planning System by Jorg Hoffmann. Focus: How FF uses plan graphs in its heuristic. Also read about enforced hillclimbing search and helpful actions. Presenter: Ferry Project 2 due on Canvas at midnight
Monday, March 10 Multi-Valued Planning Tasks and Domain Transition Graphs	Read Concise Finite-Domain Representations for PDDL Planning Tasks by Helmert. Alternatively, read Translation (book chapter, same content, maybe easier to read). Focus: How can propositional planning tasks can be converted into multi-valued planning tasks? Ignore axioms, quantifiers, and conditional effects. Presenter: Egbert
Wednesday, March 12 The Fast Downward Planner	Read The Fast Downward Planning System by Malte Helmert. Focus: The Causal Graph heuristic Presenter: Egbert
Friday, March 14 Discussion and Review
Monday, March 17 Holiday	Spring Break
Wednesday, March 19 Holiday	Spring Break
Friday, March 21 Holiday	Spring Break
Monday, March 24 Project Deadline	Project 3 due on Canvas at midnight
Advanced Techniques
Monday, March 24 Landmarks	Read Landmarks Revisited by Richter, Helmert, and Westphal. Focus: What is a landmark? How can they be discovered, and how do they help with search? Presenter: Racey
Wednesday, March 26 Improvements to Fast Forward	Read Who Said We Need to Relax All Variables? by Katz, Hoffmann, and Domshlak. Focus: How does the Red-Black heuristic relax some but not all variables to improve heuristic estimates? Presenter: Wu
Friday, March 28 Improvements to Partial Order Planning	Read Reviving Partial Order Planning by Nguyen and Kambhampati. Focus: How can state-space planning heuristics be used to improve POP? Presenter: Ti
Monday, March 31 Width-Based Planning	Read Width and Serialization of Classical Planning Problems by Geffner and Lipovetzky. Focus: What is the width of a planning problem? How can the Iterated Width algorithm be used for effective, heuristic-free search? Presenter: Ferry
Wednesday, April 2 Portfolio Planning	Read Fast Downward Stone Soup: A Baseline for Building Planner Portfolios by Helmert, Roger, and Karpas. Focus: What is portfolio planning? Presenter: Racey
Friday, April 4 Discussion and Review	Final Project Proposal due on Canvas at midnight
Narrative Planning
Monday, April 7 Intentional Planning	Read Narrative Planning: Balancing Plot and Character by Riedl and Young. Focus: How does IPOCL use causal links and intention frames to ensure every agent's action appears intentional? Presenter: Ferry
Wednesday, April 9 Planning with Conflict	Read A Computational Model of Narrative Conflict at the Fabula Level by Ware, Young, Harrison, and Roberts. Focus: How does CPOCL allow plans to contain intended but non-executed steps? Presenter: Butcher
Friday, April 11 The Glaive Planner	Read Glaive: A State-Space Narrative Planner Supporting Intentionality and Conflict by Ware and Young. Focus: How does Glaive use state space search to ensure every action is explained for every agent? How does it use modern heuristics to speed up planning? Presenter: Damron
Monday, April 14 Beliefs	Read A Possible Worlds Model of Belief for State-Space Narrative Planning by Shirvani, Ware, and Farrell. Focus: How can infinitely deep theory of mind be represented in a finite graph? Presenter: Racey
Wednesday, April 16 Combining Beliefs and Intentions	Read Combining Intentionality and Belief: Revisiting Believable Character Plans by Shirvani, Farrell, and Ware. Focus: How do the beliefs and intentions of agents affect what stories people consider believable? Presenter: Ti
Friday, April 18 The Sabre Planner	Read Sabre: A Narrative Planner Supporting Intention and Deep Theory of Mind by Ware and Siler. Focus: How does the Sabre planning algorithm generate plans where character behavior makes sense according to both their intentions and beliefs? Presenter: Wu
Monday, April 21 Review and Discussion
Wednesday, April 23 Belief and Intention Recognition	Read Narrative Planning for Belief and Intention Recognition by Farrell and Ware. Focus: How can plan recognition in classical planning be extended to belief and intention recognition in narrative planning? Presenter: Egbert
Friday, April 25 Emotion	Read Personality and Emotion in Strong-Story Narrative Planning by Shirvani and Ware. Focus: How does relief enable new kinds of stories? Presenter: Butcher
Monday, April 28 Salience	Read Manipulating Narrative Salience in Interactive Stories Using Indexter’s Pairwise Event Salience Hypothesis by Farrell, Ware, and Baker. Focus: What is salience and how can it be used to predict what people remember and expect? Presenter: Ti
Wednesday, April 30 Narrative Planning and Large Language Models	Read Language Models as Narrative Planning Heuristics by Senanayake and Ware. Focus: How is an LLM used as a heuristic for planning? Presenter: Damron
Wednesday, May 7 Final Presentations	Deliver your final project presentation from 3:30 - 5:30 PM.
Thursday, May 8 Final Project Due	Final project deliverable due on Canvas at midnight