EECS 492 Fall 2008 25-Feb-10 M. Wellman 1 GRAPH PLANNING EECS 492 Lecture 14 February 25 th , 2010 garb cleanH quiet carry dolly cook wrap garb ¬cleanH quiet ¬garb cleanH ¬quiet dinner present carry dolly cook wrap garb ¬cleanH quiet ¬garb cleanH ¬quiet dinner present State-Space Planning (Recap) Planning as state-space search Forward (progression) from initial state Backward (regression) from goal Action is fully specified at time of introduction to plan All variables bound to specific objects Ordering in sequence with existing actions
16
Embed
GRAPH PLANNING State-Space Planning (Recap)april.eecs.umich.edu/.../images/c/c3/L14_GraphPlan.pdf · The Grand Plan In principle, could anticipate all possible contingencies, and
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
EECS 492 Fall 2008 25-Feb-10
M. Wellman 1
GRAPH PLANNINGEECS 492 Lecture 14
February 25th, 2010
garb
cleanH
quiet
carry
dolly
cook
wrap
garb
¬cleanH
quiet
¬garb
cleanH
¬quiet
dinner
present
carry
dolly
cook
wrap
garb
¬cleanH
quiet
¬garb
cleanH
¬quiet
dinner
present
State-Space Planning (Recap)
Planning as state-space search
Forward (progression) from initial state
Backward (regression) from goal
Action is fully specified at time of introduction to plan
All variables bound to specific objects
Ordering in sequence with existing actions
EECS 492 Fall 2008 25-Feb-10
M. Wellman 2
GraphPlan
Blum & Furst, 1995.
Radical reformulation of plan search space
Applies to propositional encodings
Must propositionalize knowledge
Compacted state space with plan graph rep’n
Demonstrated dramatic performance improvements
Planning Graphs
Interleave levels of literals and actions, related by support, precondition, and mutex
0 t-1 t
…
…
…
…
…
…
…
…
…
…
Propositions true at time 0
Propositions true at time t-1
Propositions true at time t
Actions
EECS 492 Fall 2008 25-Feb-10
M. Wellman 3
GraphPlan Algorithm
zeroth level initial state
Repeat until solution or “level off”:
Extend planning graph by considering application of actions