AAAI 2011 Workshop on Generalized Planning


Full presentations are allotted 20 minutes and short presentations are allotted 10 minutes.

8:50 - 9:00 Opening Remarks
Session Chair: Paolo Traverso
09:00 - 09:50 Generalized Planning: Some Theory, Some Practice
Invited Talk by Hector Levesque (Slides)
09.50 - 10:30
  • A Generic Framework and Solver for Synthesizing Finite-State Controllers
    Toby Hu, Giuseppe DeGiacomo
  • An extension of knowledge-level planning to interval-valued functions
    Ronald P. A. Petrick
10:30 - 11:00 Coffee Break
Session Chair: Bob Givan
11:00 - 11:50 Life: Play and Win in 20 trillion Moves
Invited Talk by Stuart Russell (Slides)
11:50 - 12:30
  • Generalized Planning: Synthesizing Plans that Work for Multiple Environments
    Toby Hu, Giuseppe DeGiacomo
  • Qualitative Numeric Planning
    Siddharth Srivastava, Shlomo Zilberstein, Neil Immerman, Hector Geffner
12:30 - 01:40 Lunch
Session Chair: Siddharth Srivastava
01:40 - 02:30 A Symbolic Model Checking Approach to On-Board Autonomy
Invited Talk by Alessandro Cimatti (Slides)
02:30 - 03:30
  • Monitoring the Execution of Partial-Order Plans via Regression
    Christian Muise, Sheila A. McIlraith, J. Christopher Beck
  • From Software Services to a Future Internet of Services
    Marco Pistore, Paolo Traverso, Massimo Paolucci, Matthias Wagner
  • An Extensible Planning Architecture for Configurable Rich-World Actions [Short]
    Amit Kumar, Alexander Nareyek
  • Temporal Defeasible Logic-based POP for physically correct plans [Short]
    Pere Pardo
03:30 - 04:00 Coffee Break
Session Chair: Sheila McIlraith
04:00 - 05:20
  • Reasoning about Planning Domains [Short]
    Rajesh Kalyanam, Tanji Hu, Robert Givan
  • Flexible Plans for Adaptation by End-User
    Cyrille Martin, Humbert Fiorino, Gaelle Calvary
  • Stable Grounded Inference in Flexible Resource Scheduling
    Paul Morris, John Bresina, Javier Barreiro
  • Synthesizing Robust Plans under Incomplete Domain Models
    Tuan A. Nguyen, Subbarao Kambhampati, Minh B. Do
  • Planning over MDPs through HTNs [Short]
    Yuqing Tang, Felipe Meneguzzi, Katia Sycara, Simon Parsons
05:20 - 05:50 Closing Discussion

Abstracts of Invited Talks

Generalized Planning: Some Theory, Some Practice

Hector Levesque, Department of Computer Science, University of Toronto

While most of the research in automated planning within AI has focussed on methods for finding sequential (or straight-line) plans, there is growing interest in a more general account, where the plans may need to contain branches and loops. In this talk, I will present some recent work in this area, including some new theoretical results, as well as a description of a new planning system called FSAPLANNER and some of the generalized planning problems it can solve.

This is joint work with Toby Hu.

Life: Play and Win in 20 trillion Moves

Stuart Russell, Computer Science Division, University of California, Berkeley

How can we design systems that can achieve reasonable decision quality over long time scales? One approach is based on temporal abstraction, allowing deliberation over action choices of long duration. The talk will explore this idea first in the classical planning context, where the longstanding open problem of "downward refinement" is resolved. In the context of hierarchical reinforcement learning, the idea of partial programming provides a powerful and flexible method of specifying constraints on behavior, leaving unspecified those choices that the agent must learn to make on its own. Many avenues remain for further development of these ideas.

[Joint work with Ron Parr, David Andre, Bhaskara Marthi, Andy Zimdars, David Latham, Carlos Guestrin, Jason Wolfe]

A Symbolic Model Checking Approach to On-Board Autonomy

Alessandro Cimatti, IRST - Istituto per la Ricerca Scientifica e Tecnologica, FBK

Autonomous systems are typically very complex entities, and their design and operation pose substantial challenges. In addition to ensuring functional correctness, other steps may be required: safety analysis, the behavior is analyzed, and proved compliant to some requirements considering possible faulty behaviors; diagnosis and diagnosability are forms of reasoning on the run-time explanation of faulty behaviors; planning allows for the automated construction of suitable courses of actions.
Symbolic Model Checking (SMC) is a formal technique for ensuring functional correctness, that is achieving increasing industrial penetration. In this talk, we show how SMC can be used as a convenient framework for safety analysis, planning and plan validation, diagnosis and diagnosability, and monitoring. We also discuss how advanced model checking tools, based on Satisfiability Modulo Theory solvers, can be used to address the resulting computational challenges.