CAP6938-02: Special Topics in Plan, Activity, and Intent Recognition

Plan recognition is the process of making inferences about other agents based on observations, prior knowledge, and closed-world assumptions. This synergistic area of AI research combines techniques from human-computer interaction, autonomous and multi-agent systems, natural language understanding, machine learning, and computer vision.

Objectives:

• understand current research issues in plan/activity/intent recognition;
• learn the following major approaches and models: symbolic, probabilistic, graphical, and game-theoretic;
• implement a software application that uses plan/activity/intent recognition.

Assignments

• Nov 27th/29th: In-class presentations
• Dec 3rd: Final paper due (latex template, word template)

Outline

1. Introduction; Robocup Application Domain
   • Reading: H. Kautz, A Formal Theory of Plan Recognition and Its Implementation, 1991.
2. Elder-Care Application Domain; Event Hierarchy Circumscription
   • Homework: Written summary of final project (to be handed in Aug 30th)
   • Presentation: 5 minute summary of project (no PowerPoint required) to be delivered either Aug 28th or Aug 30th
   • Reading: M. Tambe and P. Rosenbloom, Event tracking in a dynamic multi-agent environment, Technical Report ISI/RR-94-393, USC/Information Sciences Institute, 1994.
   • Reading: D. Avrahami-Zilberbrand and G. Kaminka, Fast and complete symbolic plan recognition. In Proceedings of IJCAI, 2005
3. Event Tracking in SOAR; Plan Recognition for Adversarial Domains
4. Efficiency Improvements for Symbolic Plan Recognition
5. Probabilistic Grammars
   • Reading: D. Pynadath, Probabilistic Grammars for Plan Recognition, Ph.D. Thesis, University of Michigan, 1999
   • Abbreviated version: D. Pynadath and M. Wellman, Probabilistic State-Dependent Grammars for Plan Recognition, Proc. of Uncertainty in Artificial Intelligence, 2000
   • Homework: 2-page write-up of related work for project (Sept 6)
6. Graphical Models
• Reading: K. Murphy, Introduction to Graphical Models, 2001
• Homework: Project specification write-up (1-2 pages) (Sept 13)
7. Dynamic Bayes Networks (DBNs)
   • Reading: K. Murphy, Dynamic Bayesian Networks, pp.1-25
   • Homework: In-class demonstration of project (no writeup) (Sept 25)
8. Applications of Hidden Markov Models
   Reading: T. Starner and A. Pentland, Visual Recognition of American Sign Language Using Hidden Markov Models, Proceedings of Automatic Face and Gesture Recognition, 1995
9. Review
   • Exam: In-class Oct 4th
10. Exam Problems; Presentations; Questionnaire
11. Partial Observable Markov Decision Processes (POMDPs)
    • Homework: 1 page write-up summarizing project changes (Oct 18th)
    • Reading: L. Kaebling, M. Littman, and A. Cassandra, Planning and Acting in Partially Observable Stochastic Domains, Artificial Intelligence, Volume 101, pp. 99-134, 1998
12. Model Learning for POMDPs; Emotion Recognition
    • Reading: F. Doshi and N. Roy, Efficient Model Learning for Dialog Management Proceedings of Human-Robot Interaction (HRI 2007). March 2007.
    • Reading: F. Dellaert, T. Polzin, and A. Waibel, Recognizing Emotion in Speech, Proc of ICSLP, 1996
    • Homework: final-project write-up: 1 column write-up of either the Background or Method section of your paper (no in-class presentation) (latex template, word template)
13. Plan Recognition in Spoken Dialogue Corpora; Unsupervised Learning of Object Categories
    • Reading: N. Blaylock and J. Allen, Fast Hierarchical Goal Schema Recognition, Proc of AAAI, 2006 (Tues)
    • Reading: L. Fei-Fei, R. Fergus, and P. Perona, A Bayesian Approach to Unsupervised One-Shot Learning of Object Categories, Proc of ICCV, 2003 (Thurs)
    • Visitor (Nov 1): Dr. Rahul Sukthankar, Intel Research
    • Homework: continue final-project write-up: 1 column on the Method section of your paper (due Nov 1)
14. Human Motion Prediction; Tactical Game Play
    • Reading: A. Bruce and G. Gordon, Better Motion Prediction for People-tracking Proc of ICRA, 2004
    • C. Guestrin, D. Koller, C. Gearhart, and N. Kanodia Generalizing plans to new environments in relational MDPs, Proc of IJCAI, 2003 (skip proofs at end of paper)
    • Homework: continue final-project write-up: 1 column on Related Work section (due Nov 8th)
15. Undirected Graphical Models
    • Reading: D. Vail, M. Veloso, and J. Lafferty. Conditional random fields for activity recognition. In Proceedings of AAMAS, 2007.
    • Reading: W. Pentney, A.M. Popescu, S. Wang, H. Kautz, M. Philipose. Sensor-Based Understanding of Daily Life via Large-Scale Use of Common Sense In Proceedings of AAAI 2006. Boston, July 2006.
    • Homework: continue final-project write-up: 1 column on Results section (Nov 15th)
16. Learning by Observation
    • Reading: S. Chernova and M. Veloso. Interactive learning from demonstration using Gaussian mixture models. In Proceedings of AAMAS, 2007.
17. Final Presentations