80 likes | 142 Views
Probabalistic Grapgplan. CSE 574 April 29, 2003 Dan Weld. Administrivia. Leading class Your responsibility to schedule time with me Office hours tomorrow I’ll likely be late Projects Next tues: email me a status report / milestones 1. Goals 2. Progress so far
E N D
Probabalistic Grapgplan CSE 574 April 29, 2003 Dan Weld
Administrivia • Leading class • Your responsibility to schedule time with me • Office hours tomorrow • I’ll likely be late • Projects • Next tues: email me a status report / milestones 1. Goals • 2. Progress so far • 3. Milestones for each remaining week
What is prob planning? • Input? • Reward? • Output?
PGraphplan • Variant of? • Output? • Innovations?
Needed Nodes • Unary • Backward sweep from goal • Delete nodes with no path to goal • Tag remaining nodes with vector of reachable goals • Backtrack when in state where union of vectors G • Pairwise • P needs Q edges • Backtrack when state has P but not Q • How compute?
P Needs Q • P needs Q ifA with P prec(A) either • Q prec(A), or • A’, and P needs A’, and Q prec(A’) • A needs noop-P if • A doesn’t produce P yet all effects of A need P • Noop-P needs A if • Q, A is only producer of Q, and P needs Q
Admissible Heuristic T=2 Prop T=2 Act T=3 Prop T=3 Act T=4 Prop Backtrack if hval < time
TGraphplan • Output? • Variant of? • Innovations?