Lecture 27: PartialOrder Planning

1
Lecture 27Partial-Order Planning

• Andy Barto
• CMPSCI 383Spring 2001

2
Todays lecture

• The least commitment principle.
• Partial order planning.

3
Search in state space

• State-space search (like A) is not feasible in
  complex domains.
• Why not?
• Huge branching factor thousands or millions.
  (All the things you can do)
• Heuristic eval function cant eliminate actions
  completely
• Strips-like planners have problems as well.

4
State space for the blocks world
5
Search in plan space

• Following Earl Sacerdotis planner called NOAH
  1975.
• States represent partially specified plans.
• Edges represent plan-refinement operations (e.g.
  adding actions or ordering constraints).
• The initial state is the null plan.

6
Some Plan-Transforming Ops
7
Plan space for the blocks world

8
Solving the Sussman anomaly

• Using the principle of least commitment The
  planner avoids making decisions until there is a
  good reason to make a choice.

A
C
B
B
C
A
9
Graphical Rep of a Strips Rule
10
The initial plan for the Sussman anomaly contains
two rulesstart specifies the initial state and
finish specifies the goal.

11
The plan after adding an op to achieve one of the
goal conjuncts

Note least commitment principle the op is not
fully instantiated
12
The plan after adding another op. One of many
transformations that are possible.

Note that b lt a (where lt means comes before)
13
Now try to achieve the next goal conjunct add op
c

Note no commitment yet to order of c
with respect to a and b a partial order only
14
Add Threat Arcsif some ops areexecuted at the
wrong time they can undo anotherops
preconditions

drawn from an op node to precondition nodes that
are on that ops delete list (and are not
descendants of the op)
Discharge threats by constraining order of ops.
Here b lt c lt a
15
Advantages of partial-order planning

• It takes only a few steps to construct a plan
  that would take thousands of steps using a
  standard problem solving approach.
• The least commitment nature of the planner means
  it needs to search only in places where subplans
  interact with each other.
• The causal links allow the planner to recognize
  when to abandon a doomed plan without wasting a
  lot of time.