Announcements

 Assignment 4 due Thursday

 Midterm prep session Thursday

 Midterm in class next Tuesday, Feb. 8

 To take test at different time:
https://spreadsheets.google.com/a/google.com/gform?
key=teBgHp8uLYo4r-QZzIULP9w&hl=en#edit
CS 188: Artificial Intelligence
Fall 2010

Lecture 9: Logic and Planning

Peter Norvig and Sebastian Thrun

Slide credits: Stuart Russell, Rina Dechter,
Rao Kambhampati
 Finding Actions
 Search
 Over atomic states. Deterministic.
A*, etc.
 Policy finding
 Atomic or abstract (features). Stochastic.
MDP; POMDP
 Introduce today: Planning
 Features or structured states
 Plan can be hierarchical, partial order
 Other refinements can be added
State Representation
 Representation
 Atomic Representation
 s1 = s2; Result(s, a) = s′; GoalTest(s); h(s)
 Factored Representation
 Attribute(s) = val … (val numeric or Boolean)
 Result and GoalTest in terms of attributes
 Structured Representation
 All of above
 Objects
 Relations, Functions: Rel(a, b, c); F(a, b) = c
 Quantifiers (for all ∀, there exists ∃)
Factored State Variable Models
 World is made up of states which are defined in
terms of state variables
 Can be boolean (or multi-ary or continuous)
 States are complete assignments over state
variables (fully observable world)
 So, k boolean state variables can represent how
many states?
 Actions change the values of the state variables
 Applicability conditions of actions are also specified in
terms of partial assignments over state variables
 “Classical” Planning
 State: conjunction of Boolean state variables
 Action Schema:
 Action(Fly(p, from, to),
Precond: At(p, from) ∧ Plane(p)
∧ Airport(from) ∧ Airport(to)
Effect: ¬At(p, from) ∧ At(p, to))

Implicitly defines Actions(s) and Result(s, a)

Expressiveness of the language
 Advantages of the Language
 Natural to read, write, verify
 Abstract over similar actions
 Split task into parts (two plane flights)
 Easy to extend with more complex
syntax and semantics
 Can be compiled or interpreted
 Planning Algorithms
 Forward (progression) state-space search
 … it’s just search
 Backward (regresssion) state-space search
 Consider Goal: Own(0136042597)
Action(Buy(i), Pre: ISBN(i) Eff: Own(i))
 In general, may involve unbound variables
 Plan-space search
 Start with empty plan, add branches
Plan-space search
 Plan-space search
Start

Right
Shoe

Start Finish

Finish
Start

Left
Sock

Finish
Plan-space search
 Progression vs. Regression

 Progression has higher  Regression has lower

branching factor branching factor
 Progression searches in the  Regression searches in the
space of complete (and space of partial states
consistent) states  There are 3n partial states (as
against 2n complete states)
You can also do bidirectional search
A B stop when a (leaf) state in the progression tree
entails a (leaf) state (formula) in the regression tree A
holding(A)

Ontable(A) Pickup(A)
~Clear(A)
B
~Ontable(A)
Ontable(B), Ontable(B), ~clear(B) Putdown(A)
holding(A)
Clear(A) Clear(B) ~clear(B)
holding(B)
Clear(B) hand-empty
~handempty
hand-empty
~Clear(B) Stack(A,B)
Pickup(B) ~Ontable(B) holding(A)
Ontable(A), clear(B) Putdown(B)??
Clear(A)
~handempty
 Planning vs. Search
 Search assumes successor and goal-test functions which know how to
make sense of the states and generate new states
 Planning makes the additional assumption that the states can be
represented in terms of state variables and their values
 Initial and goal states are specified in terms of assignments over state variables
 Which means goal-test doesn’t have to be a blackbox procedure
 That the actions modify these state variable values
 The preconditions and effects of the actions are in terms of partial assignments over
state variables
 Given these assumptions certain generic goal-test and successor functions can
be written
 Specifically, we discussed one Successor called “Progression”, another called
“Regression” and a third called “Partial-order”
 Notice that the additional assumptions made by planning do not change the
search algorithms (A*, DFS, etc)—they only change the successor and
goal-test functions
 In particular, search still happens in terms of search nodes that have parent
pointers etc.
 The “state” part of the search node will correspond to
 “Complete state variable assignments” in the case of progression
 “Partial state variable assignments” in the case of regression
 “A collection of steps, orderings, causal commitments and open-conditions in the case of partial
order planning
 State of the art
 Annual planning competitions
 Best technique has varied over time
 Currently: mostly forward state-space
 Largely due to good heuristics (relaxed prob.)
 Heuristics for atomic (state search) problem
Can only come from outside analysis of domain
 Heuristics for factored (planning) problem
Can be domain-independent
8-puzzle state space
 8-puzzle action schema
 Action(Slide(t, a, b),
Pre: On(t, a)∧ Tile(t) ∧ Blank(b) ∧ Adjacent(a,b)
Eff: On(t, b) ∧Blank(a) ∧ ¬On(t, a) ∧¬Blank(b))
8-puzzle heuristics
Convex search: ignore del lists
Factored Rep allows control
Factored Rep allows control
 Planning Graphs
 Planning graphs are an efficient way to
create a representation of a planning
problem that can be used to
 Achieve better heuristic estimates
 Directly construct plans
 Planning graphs only work for
propositional problems
 Compile to propositional if necessary
 Planning Graphs
 Planning graphs consists of a seq of levels
that correspond to time steps in the plan.
 Level 0 is the initial state.
 Each level consists of a set of literals and a
set of actions that represent what might be
possible at that step in the plan
 Might be is the key to efficiency
 Records only a restricted subset of possible
negative interactions among actions.
 Planning Graphs
 Each level consists of
 Literals = all those that could be true at
that time step, depending upon the actions
executed at preceding time steps.
 Actions = all those actions that could
have their preconditions satisfied at that
time step, depending on which of the
literals actually hold.
Planning Graph Example
Init(Have(Cake))
Goal(Have(Cake)  Eaten(Cake))
Action(Eat(Cake),
PRECOND: Have(Cake)
EFFECT: ¬Have(Cake)  Eaten(Cake))
Action(Bake(Cake),
PRECOND: ¬ Have(Cake)
EFFECT: Have(Cake))
Planning Graph Example

Create level 0 from initial problem state.

Planning Graph Example

Add all applicable actions.

Add all effects to the next state.
Planning Graph Example

Add persistence actions (inaction = no-ops) to

map all literals in state Si to state Si+1.
Planning Graph Example

Identify mutual exclusions between actions and

literals based on potential conflicts.
 Mutual exclusion
 A mutex relation holds between two actions
when:
 Inconsistent effects: one action negates the effect of another.
 Interference: one of the effects of one action is the negation
of a precondition of the other.
 Competing needs: one of the preconditions of one action is
mutually exclusive with the precondition of the other.
 A mutex relation holds between two literals
when:
 one is the negation of the other OR
 each possible action pair that could achieve the
literals is mutex (inconsistent support).
 Cake example

 Level S1 contains all literals that could result from

picking any subset of actions in A0
 Conflicts between literals that can not occur together
(as a consequence of the selection action) are
represented by mutex links.
 S1 defines multiple states and the mutex links are the
constraints that define this set of states.
 Cake example

 Repeat process until graph levels off:

 two consecutive levels are identical
 PG and Heuristic Estimation
 PG’s provide information about the problem
 PG is a relaxed problem.
 A literal that does not appear in the final level
of the graph cannot be achieved by any plan.
 H(n) = ∞
 Level Cost: First level in which a goal appears
 Very low estimate, since several actions can occur
 Improvement: restrict to one action per level
using serial PG (add mutex links between every pair
of actions, except persistence actions).
 PG and Heuristic Estimation
 Cost of a conjunction of goals
 Max-level: maximum first level of any of
the goals
 Sum-level: sum of first levels of all the
goals
 Set-level: First level in which all goals
appear without being mutex
 The GRAPHPLAN Algorithm
 Extract a solution directly from the PG

function GRAPHPLAN(problem) return solution or failure

graph  INITIAL-PLANNING-GRAPH(problem)
goals  GOALS[problem]
loop do
if goals all non-mutex in last level of graph then do
solution  EXTRACT-SOLUTION(graph, goals,
LEN(graph))
if solution  failure then return solution
else if NO-SOLUTION-POSSIBLE(graph) then return
failure
graph  EXPAND-GRAPH(graph, problem)
 GRAPHPLAN example

 Initially the plan consist of 5 literals from the initial state (S0).
 Add actions whose preconditions are satisfied by EXPAND-GRAPH (A0)
 Also add persistence actions and mutex relations.
 Add the effects at level S1
 Repeat until goal is in level Si
 GRAPHPLAN example

 EXPAND-GRAPH also looks for mutex relations

 Inconsistent effects
 E.g. Remove(Spare, Trunk) and LeaveOverNight due to At(Spare,Ground) and not At(Spare, Ground)
 Interference
 E.g. Remove(Flat, Axle) and LeaveOverNight At(Flat, Axle) as PRECOND and not At(Flat,Axle) as
EFFECT
 Competing needs
 E.g. PutOn(Spare,Axle) and Remove(Flat, Axle) due to At(Flat.Axle) and not At(Flat, Axle)
 Inconsistent support
 E.g. in S2, At(Spare,Axle) and At(Flat,Axle)
 GRAPHPLAN example

 In S2, the goal literals exist and are not mutex with any other
 Solution might exist and EXTRACT-SOLUTION will try to find it
 EXTRACT-SOLUTION can search with:
 Initial state = last level of PG and goal goals of planning problem
 Actions = select any set of non-conflicting actions that cover the goals in the state
 Goal = reach level S0 such that all goals are satisfied
 Cost = 1 for each action.
 GRAPHPLAN Termination
 Termination of graph construction? YES
 PG are monotonically increasing or decreasing:
 Literals increase monotonically
 Actions increase monotonically
 Mutexes decrease monotonically
 Because of these properties and because
there is a finite number of actions and literals,
every PG will eventually level off
 Planning with Structured Rep

Situation Calculus

First-order Logic
 Situation Calculus
 Convenient to have more expressive lang.
“Move all the cargo from SFO to JFK”
 Use existing mechanisms for logical proof
 Strong foundation for studying planning
 Still, less used in practice than other
techniques
 Situation Calculus
 Possibility Axioms (for each action)
 SomeFormula(s) ⇒Poss(a, s)
 Alive(Agent, s) ∧ Have(Agent, Arrow, s) ⇒
Poss(Shoot, s)
 Successor-state Axiom (for each fluent)
 Poss(a, s) ⇒(fluent is true ⇔ a made it true
∨ it was true and a left it alone)
 Poss(a, s) ⇒ (Holding(Agent, g, Result(s, a)) ⇔
a = Grab(g) ∨
(Holding(Agent, g, s) ∧ a ≠ Release(g)))