Solving Problems by Searching

Chapter III
Part I
Outline
 Problem-solving agents

 Problem types

 Problem formulation

 Example problems
Problem solving
 We want:
 To automatically solve a problem

 We need:
 A representation of the problem
 Algorithms that use some strategy to
solve the problem defined in that
representation
States
 A problem is defined by its elements and their
relations.
 A state is a representation of those elements in a
given moment.
 In each instant of a problem, those elements have
specific descriptors (How to select them?) and
relations.
 Two special states are defined:
– Initial state (starting point)
– Final state (goal state)
State modification: successor function
 A successor function is needed to move
between different states.
 A successor function is a description of
possible actions, a set of operators. It is a
transformation function on a state
representation, which convert it into another
state.
 The successor function defines a relation of
accessibility among states.
State space
 The state space is the set of all states
reachable from the initial state.
 It forms a graph (or map) in which the
nodes are states and the arcs between
nodes are actions.
 A path in the state space is a sequence
of states connected by a sequence of
actions.
Problem solution
 A solution in the state space is a path from
the initial state to a goal state.

 Path/solution cost: function that assigns a

numeric cost to each path, the cost of
applying the operators to the states.

 Solution quality is measured by the path

cost function, and an optimal solution has
the lowest path cost among all solutions.
Problem solving agents
 Goal formulation, based on the current situation and

the agent’s performance measure, is the first step in

problem solving

 Goal is a set of states. The agent’s task is to find out

which sequence of actions will get it to a goal state

 Problem formulation is the process of deciding what

sorts of actions and states to consider, given a goal

Problem solving agents Cont’d..
 An agent can decide what to do by first
examining different possible sequences of
actions that lead to states of known value,
and then choosing the best sequence,
 Looking for such a sequence is called
search,
 A search algorithm takes a problem as input
and returns a solution in the form of action
sequence,
Problem solving agents Cont’d..
 After formulating a goal and a problem to
solve the agent calls a search procedure to
solve it,
 The basic algorithm for problem-solving
agents consists of 3 phases: “formulate,
search, execute” design for the agent,
 formulate the problem, search for a solution
and execute the solution.
Problem solving agents Cont’d..

 The aim of the problem-solving

agent is therefore to perform a
sequence of actions that change the
environment so that it ends up in
one of the goal states.
Once the solution has been
executed, the agent will
formulate a new goal.
Problem types
 Deterministic, fully observable  single-state
problem
 Agent knows exactly which state it will be in;

 Nondeterministic and/or partially observable 

contingency problem
 percepts provide new information about current state

 Non-observable  sensorless problem

 Agent may have no idea where it is;

 Unknown state space  exploration problem

 states and actions of the environment are initially
unknown
Example: Vacuum-Cleaner Env’t

 There is a vacuum-cleaner agent that can be

located in one of the two locations, A or B
 4 actions available to it: to move right, move
left, suck or NoOp, and its goal is to clean
both locations.
Deterministic, fully observable  single-state problem

 fully-observable: if the vacuum-

cleaner agent could perceive its own
location and whether or not there is dirt
in both locations.
 Deterministic: if the next state of the
environment could be predicted from
the current state and action.
Nondeterministic and/or partially observable
 contingency problem

 occur when the agent can obtain new

information from its sensors after acting.
 This can be the case because the
environment is partially-observable, or
because the environment is nondeterministic
(i.e. stochastic or strategic).
 Solutions to contingency problems are
harder to find.
Non-observable  sensorless problem

 if the vacuum-cleaner agent did not

receive any percepts from the
environment.
 the agent does not know what state the
environment is in
 use a belief state to indicate what set of
real ‘physical’ states it thinks it might
be in, based on actions it has taken.
Unknown state space  exploration problem

 states and actions of the

environment are initially unknown
 An extreme version of the
contingency problem
 The agent must therefore explore
the environment to discover them.
Problem Formulation in Single State
 Single-state problems can be formulated by specifying
four pieces of information:
 Initial state: what state are the environment/agent in to
begin with?
 Actions: the successor function specifies what actions
are possible in each state, and what their result would
be. It consists of a set of action-state pairs.
 Goal test: either an implicit or explicit statement of
when the agent’s goal has been achieved.
 Path cost: a step cost c(x,a,y) for each action ‘a’ that
takes the agent from state ‘x’ to state ‘y’ – the sum of
all step costs for a sequence of actions is the path cost.
Example: Romania
On holiday in Romania; currently in Arad.

 Formulate goal:
 be in Bucharest

 Formulate problem:
 states: various cities
 actions: drive between cities

 Find solution:
 sequence of cities, e.g., Arad, Sibiu, Fagaras, Bucharest
Example: Romania
Problem Formulation…. Cont’d

Initial state: the state in

which the agent starts in –
e.g. In(Arad)
Problem Formulation…. Cont’d
 Actions: A description of the possible

actions available to the agent

– Successor function – returns a set of

<action, successor> pairs

– e.g. {<Go(Sibiu),In(Sibiu)>,

<Go(Zerind), In(Zerind)>}
Problem Formulation…. Cont’d

 Goal test determines whether a given state

is a goal state – e.g.{In(Bucharest)}

Problem Formulation…. Cont’d
 Path cost function that assigns a
numeric cost to each path.
 The cost of a path can be described as the
sum of the costs of the individual actions
along the path –
 step cost – e.g. to go Bucharest
 c(Arad, AradZerind, Zerind) = 75, c(Arad,
AradSibiu, Sibiu) = 140,
 c(Arad, AradTimisoara, Timisoara) =
118, etc.
Example: The 8-puzzle

 states?
 actions?
 goal test?
 path cost?
Example: The 8-puzzle

 states? locations of tiles

 actions? move blank left, right, up, down
 goal test? = goal state (given)
 path cost? 1 per move
Vacuum world state space graph

 states?
 actions?
 goal test?
 path cost?
Vacuum world state space graph

 states? dirt and robot location

 actions? Left, Right, Suck, NoOp
 goal test? no dirt at all locations
 path cost? 1 per action
Exercise: Water Pouring
 Given a 4 gallon bucket and a 3 gallon
bucket, how can we measure exactly 2
gallons into one bucket?
 There are no markings on the bucket
 You must fill each bucket completely
Water Pouring
 Initial state:
 The buckets are empty
 Represented by the tuple ( 0 0 )

 Goal state:
 One of the buckets has two gallons of water in it
 Represented by either ( x 2 ) or ( 2 x )

 Path cost:
 1 per unit step
Water Pouring
 Actions and Successor Function
 Fill a bucket
 (x y) -> (3 y)
 (x y) -> (x 4)
 Empty a bucket
 (x y) -> (0 y)
 (x y) -> (x 0)
 Pour contents of one bucket into
another
Water Pouring
(0,0)

(4,0) (0,3)

(1,3) (4,3) (3,0)

(1,0) (0,1)
(3,3) (4,2)

(4,1)
(2,3)

(2,0) (0,2)
Exercise: Eight Queens
 States:
 Any arrangement of 0 to 8 queens on the
board
 Initial state:
 No queens on the board
 Successor function:
 Add a queen to an empty square
 Goal Test:
 8 queens on the board and none are
attacked
Solution for 8 Queens
Part two
Outline
 Basic search Strategies

 Uninformed Search Strategies

 BFS

 UCS

 DFS

 DLS
Search strategies
 A search strategy is defined by picking the order of
node expansion
 Strategies are evaluated along the following
dimensions:
 completeness: does it always find a solution if one
exists?
 time complexity: number of nodes generated
 space complexity: maximum number of nodes in
memory
 optimality: does it always find a least-cost solution?
 Time and space complexity are measured in terms
of
 b: maximum branching factor of the search tree
 d: depth of the least-cost solution
 m: maximum depth of the state space (may be ∞)
 Search Strategy Techniques
 Uninformed Search Strategy: a searching technique
which have no additional information about the distance
from current state to the goal.
 is a class of general purpose search algorithms that operate in
a brute-force way.

 Informed Search Strategy: is a another technique which

have additional information about the estimate distance
from the current state to the goal.
 uses prior knowledge about problem,hence very efficient
Uninformed search strategies
 Uninformed search strategies use only the
information available in the problem definition.

 Breadth-first search

 Depth-first search

 Depth Limited Search

 Uniform-cost search
Breadth-first search
 The root node is expanded first, then all the successors
of the root node, and their successors and so on
 In general, all the nodes are expanded at a given depth
in the search tree before any nodes at the next level are
expanded
 Expand shallowest unexpanded node
 Implementation:
– fringe is a FIFO queue,
– the nodes that are visited first will be expanded first
–All newly generated successors will be put at the end
of the queue
– Shallow nodes are expanded before deeper nodes
 Breadth-first Search
A

Successors: B,C,D D
B C Initial state

E F G H I J

K L M N O P Q R

Goal state
S T U
Visited:
The fringe is the data structure we use to store all of the
Fringe: A (FIFO) nodes that have been generated
 Breadth-first Search
A

Next node
D
Successors: E,F B C

E F G H I J

K L M N O P Q R

S T U Visited: A

Fringe: B,C,D (FIFO)

 Breadth-first Search
A

Successors: K,L
D
B C

Next node
E F G H I J

K L M N O P Q R

Visited: A, B, C, D
S T U

Fringe: E,F,G,H,I,J (FIFO)

 Breadth-first Search
A

Successors: M
D
B C
Next node

E F G H I J

K L M N O P Q R

Visited: A, B, C, D, E
S T U

Fringe: F,G,H,I,J,K,L (FIFO)

 Breadth-first Search
A

Successors: N
D
B C
Next node

E F G H I J

K N O P Q R
L M

Visited: A, B, C, D, E, F
S T U

Fringe: G,H,I,J,K,L,M (FIFO)

 Breadth-first Search
A

Successors: O
D
B C
Next node

E F G H I J

K L M N O P Q R

U Visited: A, B, C, D, E, F,
S T
G

Fringe: H,I,J,K,L,M,N (FIFO)

 Breadth-first Search
A

Successors: P,Q
D
B C
Next node

E F G H I J

K L M N O P Q R

S T U Visited: A, B, C, D, E, F,
G, H

Fringe: I,J,K,L,M,N,O (FIFO)

 Breadth-first Search
A

Successors: R
D
B C
Next node

E F G H I J

K L M N O P Q R

Visited: A, B, C, D, E, F,
S T U G, H, I

Fringe: J,K,L,M,N,O,P,Q (FIFO)

 Breadth-first Search
A

Successors: S
D
B C

E F G H I J
Next node

K L M N O P Q R

S T U Visited: A, B, C, D, E, F,
G, H, I, J

Fringe: K,L,M,N,O,P,Q,R (FIFO)

 Breadth-first Search
A

Successors: T
D
B C

E F G H I J

Next node

K L M N O P Q R

S T U Visited: A, B, C, D, E, F,
G, H, I, J, K

Fringe: L,M,N,O,P,Q,R,S (FIFO)

 Breadth-first Search
A

Successors: T
D
B C

E F G H I J

Next node

K L M N O P Q R

S T U Visited: A, B, C, D, E, F,
G, H, I, J, K

Fringe: L,M,N,O,P,Q,R,S (FIFO)

 Breadth-first Search
A

Successors: D
B C

E F G H I J

K L M N O P Q R

Next node
S T U Visited: A, B, C, D, E, F,
G, H, I, J, K, L

Fringe: M,N,O,P,Q,R,S,T (FIFO)

 Breadth-first Search
A
Goal state achieved
Successors: D
B C

E F G H I J

K L M N O P Q R

Next node
Visited: A, B, C, D, E, F,
S T U G, H, I, J, K, L, M

Fringe: N,O,P,Q,R,S,T (FIFO)

Properties of breadth-first search
 Complete? Yes (if b is finite)

 Time? 1+b+b2+b3+… +bd + b(bd-1) = O(bd+1)

 Space? O(bd+1) (keeps every node in memory)

 Optimal? Yes (if cost = 1 per step)

 Space is the bigger problem (more than time)

Uniform-cost search (UCS)
 Uniform cost search is a search algorithm used to
traverse, and find the shortest path in weighted
trees and graphs.
 Uniform Cost Search or UCS begins at a root node
and will continually expand nodes, taking the node
with the smallest total cost from the root until it
reaches the goal state.

 Uniform cost search doesn't care about how many

steps a path has, only the total cost of the path.
 UCS with all path costs equal to one is identical to
breadth first search.
 Uniform Cost Search

A
10
1

S 5 B 5 D
5
15
C

Similar to BFS except that it sorts (ascending order) the

nodes in the fringe according to the cost of the node.
where cost is the path cost.
 Uniform Cost Search
Fringe = [S0]
A Queue
10
1

S 5 B 5 D
5
15
C
Next Node=Head of Fringe=S, S is not
goal
Updated Fringe=[A1,B5,C15]
Queue Successor(S)={C,B,A}=expand(S) but
sort them according to path cost.
 Uniform Cost Search
Fringe = [S0]
A Queue
10
1

S 5 B 5 D
5
15
C
Next Node=Head of Fringe=S, S is not
goal
Updated Fringe=[A1,B5,C15]
Queue Successor(S)={C,B,A}=expand(S) but
sort them according to path cost.
 Uniform Cost Search
Fringe = [A1,B5,C15]
A
10
1

S 5 B 5 D
5
15
C
Next Node=Head of Fringe=A, A is not
Updated Fringe=[B5,D11,C15] goal
Successor(A)={D}=expand(A)
Sort the queue according to path cost.
 Uniform Cost Search
Fringe = [B5,D11,C15]
A
10
1

S 5 B 5 D
5
15
C
Next Node=Head of Fringe=B, B is not
goal
Updated Fringe=[D10,D11,C15]
Successor(B)={D}=expand(B)
Sort the queue according to path cost.
 Uniform Cost Search
Fringe = [D10,D11,C15]
A
10
1
S 5 B 5
D
5
15
C
Next Node=Head of Fringe=D,
Always finds the
D is a GOAL (cost 10 = 5+5)
cheapest solution
SBD
 Depth First Search - Method

 Expand Root Node First

 Explore one branch of the tree before
exploring another branch
 If a leaf node do not represent a goal
state, search backtracks up to the
next highest node that has an
unexplored path
DFS
 Depth-first search (DFS) is an
algorithm for traversing or
searching a tree, tree structure, or
graph.
 One starts at the root (selecting
some node as the root in the graph
case) and explores as far as
possible along each branch before
backtracking.
 Depth-First Search
A

Successors: B,C,D
D
B C Initial state

E F G H I J

K L M N O P Q R

S T Goal state U Visited:

Fringe: A (LIFO)
 Depth-First Search
A
Successors: E,F
D
B C

E F G H I J

K L M N O P Q R

S T U
Visited: A

Fringe: B,C,D (LIFO)

 Depth-First Search
A

Successors: K,L D
B C

E F G H I J

K L M N O P Q R

Visited: A, B
S T U

Fringe: E,F,C,D (LIFO)

 Depth-First Search
A

Successors: S D
B C

E F G H I J

K L M N O P Q R

S T U Visited: A, B, E

Fringe: K,L,F,C,D (LIFO)

 Depth-First Search
A

Successors: D
B C

E F G H I J

K L M N O P Q R

S T U Visited: A, B, E, K

Fringe: S,L,F,C,D (LIFO)

 Depth-First Search
A

Successors: T D
B C

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S
S T U
Backtracking
Fringe: L,F,C,D (LIFO)
 Depth-First Search
A

Successors: D
B C

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S, L
S T U

Fringe: T,F,C,D (LIFO)

 Depth-First Search
A

Successors: M D
B C

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S, L, T
S T U
Backtracking
Fringe: F,C,D (LIFO)
 Depth-First Search
A

D
Successors: B C

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S, L, T,
S T U F

Fringe: M,C,D (LIFO)

 Depth-First Search
A

Successors: G,H D
B C

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S, L, T,
F, M
S T U
Backtracking
Fringe: C,D (LIFO)
 Depth-First Search
A

Successors: N D
B C

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S, L, T,
F, M, C
S T U

Fringe: G,H,D (LIFO)

 Depth-First Search
A

Successors: D Goal state achieved

B C
Finished search

E F G H I J

K L M N O P Q R

Visited: A, B, E, K, S, L, T,
S T U F, M, C, G

Fringe: N,H,D (LIFO)

Properties of depth-first search
 Complete? No: fails in infinite-depth spaces,
spaces with loops
 Modify to avoid repeated states along path
 complete in finite spaces

 Time? O(bd): terrible if m is much larger than d

 Space? O(bm), i.e., linear space!
 Optimal? No
 Depth limited search
 Like Depth first search, but the search is limited to a
predefined depth.

 The depth of each state is recorded as it is generated. When

picking the next state to expand, only those with depth less or
equal than the current depth are expanded.

 Once all the nodes of a given depth are explored, the current
depth is incremented.
Properties of depth limited search

 Incomplete: if solution is below

cut-off depth l
 Optimal: if solution is above cut-
off depth l
 Time complexity: O(bl)
 Space complexity: O(bl)
 Summary of Uninformed Tree Search Algorithms

Criterion BFS UCS DFS DLS

Complete? Yes Yes No No
Optimal? Yes Yes No No
Time O(bd+1) O(b[C*/ε]) O(bm) O(bl)
Space O(bd+1) O(b[C*/ε]) O(bm) O(bl)
Where:
b = maximum branching factor m = maximum tree depth
d = depth of least-cost solution ε = minimum action cost
l = cut-off depth C* = path cost to solution
Informed (Heuristic) Search
 Best-first search
 Greedy best-first search (Assignment)
 A* search (Assignment)
 Hill-climbing search (Assignment)
 Simulated annealing search (Assignment)
?