GPS Planning

1
GPS Planning
2
Goals, Plans and GPS

• Planning is about
• Robots
• Acting in the real world
• But robots and real worlds are hard
• Blocks world
• Tile world
• Virtual world?

3
A Tile World Drug world
4
Goals

• Partial State Descriptions
• Problems
• Hard to specify the partial states
• Are goals fundamental?
• For now, use a vector space representation
• 2,3, 2,

5
Operators

• Things the robot can do
• With information about outcomes (or
  postconditions)
• And preconditions
• Adding the drug dimension
• buyDrugs(x,y,d)-gtx,y,d1)
• Precondition 7,1,

6
Parameterizing operators

• Add difference to the parameters
• You might just view this as standing for a class
  of actions
• Lets add money to the mix

7
Some problems

• How will we know what the differences will be?
• How complex will the functions get?

8
Means-ends analysis

• OK, What does a planner do?
• Goals in, plans out
• Problems
• Knowing everything
• Opportunistic planning
• Newell Simons General Problem Solver (GPS)

9
Hillclimbing

• What should I do now?
• Choose the action that will get you closest to
  your goal.
• Difference function
• Problems
• Local minimum problem

10
Operator/difference table

• Consider everything?
• Index operators on the differences they make.

11
Subgoaling

• Goal of being at 8,8,5,
• Get to 8,8,,
• Get to ,,5,
• Use precondition information
• Difference functions may be hard to come by

12
Summary

• Operators are packages of results and
  preconditions)
• Operators as functions on preconditions results
• Operators should allow parameters
• Goals as partial descriptions
• States as vectors
• Means-end as hillclimbing
• Distance metrics are hard
• Index operators by dimensions of difference
• Subgoaling is important.

13
A Tile World Drug world
moveEast moveWest moveSouth moveNorth withDraw bu
yDrugs
Initial State 1,2,0,d Goal 1,2,5,