University of Massachusetts Amherst
Department of Computer Science

 

 

CS 791AP

Advanced Topics in Automated Planning

Fall 2010

 

Siddharth Srivastava and Shlomo Zilberstein

 


Seminar Information


Description: Planning is the hallmark of intelligence and one of the most fundamental problems in AI research. The goal of this seminar is to examine the most significant advances in planning, starting with the fundamentals of representation and leading to the state-of-the-art in domain-independent planning. Following a brief historical overview, we will cover modern planning techniques including partial-order planning, planning as satisfiability, planning graphs, planning as heuristic search, and contingent planning. The second half of the semester will focus on one of the hardest challenges in the field, namely the problem of finding program-like "generalized plans", which can solve many different problem instances. Such plans inherently require new, more compact representations, with a sophisticated flow of control. Finding and analyzing generalized plans therefore requires a synthesis of methods from AI and software model checking. We will cover different aspects of this problem, addressing questions such as: How must plan representations be extended to express such plans? How can we reason about large state spaces with unknown properties and quantities of objects? How can we efficiently find generalized plans and characterize the range of problems they solve? We will discuss applications and connections between these problems and program verification, automatic service composition, and robot task planning.

Credit: 1 or 3 units

Time: Wednesday 12:00-2:00 in CS 140

Instructors:

Seminar Format: Each seminar meeting will cover 1-2 topics. Each topic will have an associated required reading (typically one paper) plus optional related papers. In each meeting, the instructors will give an introduction to the topic being covered, placing it in the broader context of the seminar. This will be followed by a discussion led by participants. Each topic discussion will include an overview (using slides or the white board) of the assigned reading, followed by a critical examination of the contributions and their limitations.

Requirements:

1 credit

3 credits


2010 Siddharth Srivastava & Shlomo Zilberstein