Chapter 3

Solving Problems by
Searching
CS361 Artificial Intelligence
Dr. Khaled Wassif
Spring 2023

(This is the instructor’s notes, and the student must

read the textbook for complete material.)
 Chapter Outline
◼ Problem-solving agents
◼ Problem types
◼ Problem formulation
◼ Example problems
◼ Uninformed (Blind) search
◼ Informed (Heuristic) search
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 2
By Dr. Khaled Wassif
 Why Search ?
◼ Early works of AI was mainly towards
– Theorems proving
– Solving puzzles
– Playing games
◼ All AI is search!
– Not totally true (obviously) but more true than you might
think.
– All life is problem solving !!
– Finding a good/best solution to a problem between many
possible solutions.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 3
By Dr. Khaled Wassif
 Classic AI Search Problems
◼ Map searching (navigation)

◼ 6*3*3 Rubik’s Cube

◼ 8-Puzzle 2 1 3 1 2 3
4 7 6 4 5 6
◼ N-Queens 5 8 7 8

◼ Missionaries and cannibals River

boat
◼ The River Problem Farmer, Wolf,
Duck and Corn
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 4
By Dr. Khaled Wassif
 Map Searching (navigation)
◼ On touring holiday in Romania; currently in Arad.
◼ Flight leaves tomorrow from Bucharest
– Non-refundable ticket.
◼ Formulate goal:
– be in Bucharest on time
◼ Formulate problem:
– states: various cities
– actions: drive between cities
◼ Find solution (action sequences):
– sequence of cities, e.g., Arad, Sibiu, Fagaras, Bucharest
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 5
By Dr. Khaled Wassif
 Problem-Solving Agents
◼ A kind of goal-based agents that decide what to do by
finding sequences of actions that lead to desirable states.
◼ Goals help to organize behavior by limiting the
objectives that the agent is trying to achieve.
◼ Goal formulation is the first step in problem solving,
based on the current situation and the agent’s performance
measure.
– As well as formulating a goal, the agent may wish to decide
on some other factors that affect the desirability of different
ways of achieving the goal.
◼ Problem formulation is the process of deciding what
actions and states to consider, given the goal formulation.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 6
By Dr. Khaled Wassif
 Problem-Solving Agents (cont.)
◼ An agent with several immediate options of unknown
value 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 sequences.
– This process of looking for such a sequence is called search.
– A search algorithm takes a problem as input and returns a
solution in the form of an action sequence.
– Once a solution is found, the actions it recommends can be
carried out through the execution phase.
◼ Thus, the agent design is simply consists of “formulate,
search, execute” phases.
– Once the solution has been executed, the agent will formulate
a new goal.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 7
By Dr. Khaled Wassif
 Problem-Solving Agents

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 8

By Dr. Khaled Wassif
 Problem-Solving Agents (cont.)
◼ The agent design assumes that:
– The environment is static, because formulating and solving
the problem is done without paying attention to any changes
that might be occurring in the environment.
– The initial state is known; knowing it is easiest if the
environment is observable.
– The environment can be viewed as discrete to allow
enumerating “alternative courses of action”.
– Finally and most importantly, the environment is deterministic.
◼ Solutions to problems are single sequence of actions
and cannot handle any unexpected events.
◼ Solutions are executed without paying attention to the
percepts!!
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 9
By Dr. Khaled Wassif
 Problem Types
◼ There are four essentially different types of problems:
– Single-state problems
» The agent’s sensors provide enough information for the agent to tell
exactly which state it is in => Fully observable.
» The agent knows exactly the effects of all its actions => Deterministic.
» Then, the agent can calculate exactly which state it will be in after any
sequence of actions.
– Multiple-state (sensorless) problems
» The agent is not certain about which state in a given set of states it is in
(has limited access to the world state) => Non-observable.
» The agent knows exactly the effects of all its actions => Deterministic.
» the agent must reason about sets of states that it might get to, rather
than single states.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 10
By Dr. Khaled Wassif
 Problem Types
◼ There are four essentially different types of problems:
– Contingency problems
» Nondeterministic and/or partially observable
» Solving this problem requires sensing during the execution phase.
◼ Percepts provide new information about current state
» The agent must now calculate a whole tree of actions, rather than a
single action sequence.
◼ Each branch of the tree deals with a possible contingency that might arise.
– Exploration problems
» Unknown state space (as lost in a desert without a map)
» The agent must experiment, gradually discovering what its actions do
and what sorts of states exist.
» This is a kind of search, but a search in the real world rather than in a
model thereof.
» If it survives, the agent learns a "map" of the environment, which it can
then use to solve subsequent problems.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 11
By Dr. Khaled Wassif
 Example: Vacuum World
◼ Let the world contain just two
locations.
◼ Each location may or may not
contain dirt.
◼ the agent may be in one location
or the other.
◼ Then, there are 8 possible world
states.
◼ The agent has three possible
actions: Left, Right, and Suck.
◼ The goal is to clean up all the
dirt - set of goal states is{7,8}.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 12
By Dr. Khaled Wassif
 Example: Single-state Problem
◼ If its initial state is 5.
◼ The goal is to clean up all
the dirt - set of goal states is
{7,8}.
Solution?
◼ It can calculate that the
action sequence [Right, Suck]
will get to a goal state.

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 13

By Dr. Khaled Wassif
 Example: Multiple-state Problem
◼ If its initial state is one of
the set {1, 2, 3, 4, 5, 6, 7, 8}.
◼ It can calculate that the
action Right will cause it in
one of states {2, 4, 6, 8}.
◼ Set of goal states is{7,8}.
Solution?
◼ the agent can discover that the
action sequence [Right, Suck,
Left, Suck] is guarantee to
reach a goal state.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 14
By Dr. Khaled Wassif
 Example: Contingency Problem
◼ Nondeterministic: If Suck
may dirty a clean carpet.
◼ Partially observable:
location, dirt at current
location.
◼ Percept: [L, Clean], i.e.,
its initial state is one of
the set {5, 7}.
◼ Set of goal states is{7,8}.
Solution?
◼ [Right, if dirt then Suck].
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 15
By Dr. Khaled Wassif
Well-defined Problems and Solutions
◼ A problem is really a collection of information that the
agent will use to decide what to do.
◼ A problem can be defined formally by five components:
– The initial state that the agent starts in.
– A description of the possible actions available to the agent.
– A description of what each action does; named transition
model.
– The goal test, which determines whether a given state is a
goal state.
– A path cost function that assigns a numeric cost to each path.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 16
By Dr. Khaled Wassif
Well-defined Problems and Solutions
◼ A problem is really a collection of information that the
agent will use to decide what to do.
◼ A problem can be defined formally by five components:
– The initial state that the agent starts in.
– A description of the possible actions available to the agent.
» Given a particular state s, ACTIONS(s) returns the set of actions that
can be executed in s.
» We say that each of these actions is applicable in s.
– A description of what each action does; named transition
model.
– The goal test, which determines whether a given state is a
goal state.
– A path cost function that assigns a numeric cost to each path.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 17
By Dr. Khaled Wassif
Well-defined Problems and Solutions
◼ A problem is really a collection of information that the
agent will use to decide what to do.
◼ A problem can be defined formally by five components:
– The initial state that the agent starts in.
– A description of the possible actions available to the agent.
– A description of what each action does; named transition
model.
» specified by a function RESULT(s, a) that returns the state that results
from doing action a in state s.
» We also use the term successor to refer to any state reachable from a
given state by a single action.
– The goal test, which determines whether a given state is a
goal state.
– A path cost function that assigns a numeric cost to each path.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 18
By Dr. Khaled Wassif
Well-defined Problems and Solutions
◼ A problem is really a collection of information that the
agent will use to decide what to do.
◼ A problem can be defined formally by five components:
– The initial state that the agent starts in.
– A description of the possible actions available to the agent.
– A description of what each action does; named transition
model.
– The goal test, which determines whether a given state is a
goal state.
» Sometimes there is an explicit set of possible goal states, and the test
simply checks whether the given state is one of them.
» Sometimes the goal is specified by an abstract property.
– A path cost function that assigns a numeric cost to each path.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 19
By Dr. Khaled Wassif
 Single-state Problem Formulation
◼ A problem is defined by five items: ◼ A solution is:
– Initial state ̶ A sequence of
» e.g., In(Arad)
actions leading
from the initial
– Actions state to a goal state.
» e.g., from the state In(Arad), the applicable actions are {Go(Sibiu),
Go(Timisoara), Go(Zerind)}.
– Transition model
» e.g., we have RESULT(In(Arad), Go(Zerind)) = In(Zerind)
or successor function Succ(x) = set of action–state pairs
e.g., Succ(In(Arad)) = {<Go(Zerind), In(Zerind)>, … }
– Goal test
» can be explicit, e.g., x = In(Bucharest) or implicit, e.g., Checkmate(x)
– Path cost (additive)
» e.g., sum of distances, number of actions executed, etc.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 20
By Dr. Khaled Wassif
Well-defined Problems and Solutions
◼ The initial state, actions, and transition model implicitly
define the state space of the problem.
– A state space is the set of all states reachable from the initial
state by any sequence of actions.
– A state space forms a graph (as Romania 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.
◼ The problem-solving agent chooses a cost function that
reflects its own performance measure.
– Assuming that the cost of a path can be described as the sum
of the costs of the individual actions along the path.
◼ A solution to a problem is:
– A path from the initial state to a state that satisfies the goal
test and its quality is measured by the path cost function.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 21
By Dr. Khaled Wassif
 Selecting a State Space
◼ Real world is absurdly complex
– State space must be abstracted for problem solving
– Process of removing irrelevant detail from a representation is
called abstraction.
◼ (Abstract) state = set of real states
◼ (Abstract) action = complex combination of real actions
– e.g., “Arad → Zerind” represents a complex set of possible
routes, detours, rest stops, etc.
– For guaranteed realizability, any real state In(Arad) must
get to some real state In(Zerind)
◼ (Abstract) solution = set of real paths that are solutions
in the real world
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 22
By Dr. Khaled Wassif
 Vacuum World State Space Graph

◼ states?
◼ actions?
◼ goal test?
◼ path cost?
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 23
By Dr. Khaled Wassif
 Vacuum World State Space Graph

◼ states? figure dirt and robot locations (ignore dirt amounts)

◼ actions? Left, Right, Suck
◼ goal test? no dirt at all locations
◼ path cost? 1 per action
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 24
By Dr. Khaled Wassif
 Example: The 8-puzzle

◼ states?
◼ actions?
◼ goal test?
◼ path cost?

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 25

By Dr. Khaled Wassif
 Example: The 8-puzzle

◼ states? locations of tiles (ignore intermediate positions)

◼ actions? move blank left, right, up, down (ignore number
movement etc.)
◼ goal test? = goal state (given)
◼ path cost? 1 per move
(The 8-puzzle has 9!/2=181, 440 reachable states)
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 26
By Dr. Khaled Wassif
 Example: robotic assembly

◼ states?
◼ actions?
◼ goal test?
◼ path cost?
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 27
By Dr. Khaled Wassif
 Example: robotic assembly

◼ states?: real-valued coordinates of robot joint angles

parts of the object to be assembled
◼ actions?: continuous motions of robot joints
◼ goal test?: complete assembly
◼ path cost?: time to execute
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 28
By Dr. Khaled Wassif
 Real World Problems
◼ We have already seen how the route-finding problem
is defined in terms of specified locations and transitions
along links between them.
– This problem algorithms are used in a variety of applications,
such as in-car systems that provide driving directions, routing
video streams in computer networks, military operations
planning, and airline travel-planning systems.
◼ Touring problems (as traveling salesperson problem)
are related to route-finding problems, but with an
important difference.
◼ Robot navigation is a generalization of the route-
finding problem.
– But; rather than following a discrete set of routes, a robot can
move in a continuous space with an infinite set of possible
actions and states.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 29
By Dr. Khaled Wassif
 Real World Problems
◼ A VLSI layout problem requires positioning millions
of components and connections on a chip.
– The objectives are to minimize area, minimize circuit delays,
and maximize manufacturing yield.
◼ Automatic assembly sequencing of complex objects
by a robot.
– The aim is to find an order in which to assemble the parts of
some object.
– If the wrong order is chosen, there will be no way to add some
part later in the sequence without undoing some of the work
already done.
◼ Another important assembly problem is protein design
– The goal is to find a sequence of amino acids that will fold
into a three-dimensional protein with the right properties to
cure some disease.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 30
By Dr. Khaled Wassif
 Searching for Solutions
◼ The search process can be seen as building up a search
tree that is applied to search nodes over the state space.
– The root of the tree is a node corresponding to the initial state;
the branches are actions and the nodes correspond to states in
the state space of the problem.
– At each step, testing whether the current state a goal state.
– If it is not, expanding the current state; by applying each legal
action, to generate a new set of states.
– Selecting one state from the set of generated states to be the
current state and putting others aside for later, in case the
selected state does not lead to a solution.
– Continue testing, expanding, and selecting until either a
solution is found or there are no more states to expand.
◼ The set of all
choice of leaf nodes
which stateavailable forfirst
to expand expansion at any
is determined
given
by thepoint is called
search theoffrontier.
strategy each search algorithm.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 31
By Dr. Khaled Wassif
Partial Search Tree for Finding a Route
From Arad to Bucharest

The Initial State

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 32

By Dr. Khaled Wassif
Partial Search Tree for Finding a Route
From Arad to Bucharest

After Expanding Arad

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 33

By Dr. Khaled Wassif
Partial Search Tree for Finding a Route
From Arad to Bucharest

After Expanding Sibiu

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 34

By Dr. Khaled Wassif
Initial Description of Tree-search Algorithm

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 35

By Dr. Khaled Wassif
 Avoiding Repeated States
◼ Possibility of wasting time by expanding repeated states
that have already been encountered and expanded before
by loopy paths.
– For some problems, this possibility never comes up.
» The state space is a tree and there is only one path to each state.
– For other problems, repeated states are unavoidable.
» This includes all problems where the operators are reversible.
◼ Search trees for repeated states problems are infinite.
– But, some of its repeated states can be pruned to make the
search tree down to finite size.
◼ Repeated states can cause a solvable problem to become
unsolvable if the search algorithm does not detect them.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 36
By Dr. Khaled Wassif
 Avoiding Repeated States (cont.)
◼ Loopy paths are a special case of the more general
concept of redundant paths, which exist when there is
more than one way to get from one state to another.
– If we are concerned about reaching the goal, there’s never
any reason to keep more than one path to any given state.
◼ The way to avoid exploring redundant paths is to
remember where one has been.
– To do this, we enhance the TREE-SEARCH algorithm with a
data structure called the explored set (also known as the
closed list), which remembers every expanded node.
– Newly generated nodes that match previously generated
nodes can be discarded instead of being added to the frontier.
◼ The new algorithm is called GRAPH-SEARCH and it is
more efficient on problems with many repeated states.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 37
By Dr. Khaled Wassif
Initial Description of Graph-search Algorithm

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 38

By Dr. Khaled Wassif
 State Space vs. Search Tree
◼ Essential to distinguish between the state space and the
search tree.
◼ Important to remember the difference between the nodes
and the states.
– A node is a bookkeeping data structure used to represent the
search tree.
– A state corresponding to a configuration of the world.
◼ Thus, nodes are on particular paths, whereas states do not.
– Two different nodes can contain the same world state, if that
state is generated via two different search paths.
◼ Need to represent the collection of nodes that have been
generated but not yet expanded (called the frontier).
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 39
By Dr. Khaled Wassif
 State Space vs. Search Tree

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 40

By Dr. Khaled Wassif
 Needed Data Structures
◼ Search algorithms require a data structure to keep track
of the search tree that is being constructed.
◼ For each node of the tree, we have a structure that
contains four components:
– State: the state in the state space to which the node corresponds
– Parent: the node in the search tree that generated this node
– Action: the action that was applied to generate the node
– Path-cost: the cost of the path from the initial state to the node
◼ The frontier needs to be stored in such a way that the
search algorithm can easily choose the next node to
expand according to its preferred strategy.
– The appropriate data structure for this is a priority queue.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 41
By Dr. Khaled Wassif
 Search Strategies
◼ A search strategy is defined by the order of node expansion.
◼ Strategies are evaluated along the following dimensions:
– Completeness: is the strategy guaranteed to find a solution
when there is one?
– Optimality: does the strategy always find a least-cost solution?
– Time complexity: how long does it take to find a solution?
– Space complexity: how much memory is needed to perform
the search?
◼ Time and space complexity are measured in terms of:
– b: maximum branching factor of the search tree.
– d: depth of the least-cost solution (shallowest goal node).
– m: maximum length of any path in the state space (may be ∞).
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 42
By Dr. Khaled Wassif
 Search Strategies
◼ Uninformed (Blind) search:
– The search does not have additional information about states
beyond that provided in the problem definition.
– Include: breadth-first, uniform-cost, depth-first, depth limited,
iterative deepening, and bidirectional search
◼ Informed (Heuristic) search:
– The search is guided by an evaluation function:
– Include: Greedy best-first, A*, IDA*, and beam search
◼ Optimization
– The search is to find an optimal value of an objective function
– Include: Hill climbing, simulated annealing, genetic algorithms,
Ant Colony Optimization
◼ Game playing
– An adversarial search: Mini-max algorithm, alpha-beta pruning
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 43
By Dr. Khaled Wassif
Uninformed (Blind) Search Strategies
◼ They have no additional information about states beyond
that provided in the problem definition.
– All they can do is generate successors and distinguish a goal
state from a non-goal state.
– All they are distinguished by the order in which nodes are
expanded.
◼ The six uninformed search strategies are:
– Breadth-first search
– Uniform-cost search
– Depth-first search
– Depth-limited search
– Iterative deepening search
– Bidirectional search
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 44
By Dr. Khaled Wassif
 Breadth-first Search (BFS)
◼ The root node is expanded first, then all the successors
of the root node are expanded next, then their successors,
and so on.
– In general, all the nodes at depth d in the search tree are
expanded before the nodes at depth d + 1.
◼ An instance of the GRAPH-SEARCH algorithm in which
the shallowest unexpanded node is chosen for expansion.
– The nodes that are visited first will be expanded first.
– This is achieved simply by using a FIFO queue for the frontier.
– Thus, new nodes go to the back of the queue, and old nodes,
which are shallower than the new nodes, get expanded first.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 45
By Dr. Khaled Wassif
 Breadth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 46

By Dr. Khaled Wassif
 Breadth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 47

By Dr. Khaled Wassif
 Breadth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 48

By Dr. Khaled Wassif
 Breadth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 49

By Dr. Khaled Wassif
 Breadth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 50

By Dr. Khaled Wassif
 Breadth-first Search (cont.)
◼ BFS is complete:
– If there is a solution, BFS is guaranteed to find it.
– If there are several solutions, BFS will always find the
shallowest goal state (shortest path) first.
◼ BFS is optimal given the path cost is a non-decreasing
function of the depth of the node.
◼ Time complexity of BFS is:
1 + b + b2 + b3 +… + bd = O(bd)
◼ Space complexity of BFS is also O(bd)
– Because every generated node must remain in memory.
◼ The memory requirements are a bigger problem than
the execution time.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 51
By Dr. Khaled Wassif
Time and Memory Requirements for BFS

Assuming branching factor b = 10;

1 million nodes/second; 1000 bytes/node.

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 52

By Dr. Khaled Wassif
 Uniform-cost Search (UCS)
◼ BFS finds the shallowest goal state, but this may not
the least-cost solution for a general path cost function.
◼ Instead of expanding the shallowest node, UCS expands
the node n with the lowest path cost g(n).
– Done by storing the frontier as a priority queue ordered by g.
◼ Note: if all step costs are equal, UCS is identical to BFS.
◼ The cost of a path must never decrease as we go along
the path.
– If every step has a nonnegative cost, then the cost of a path
can never decrease as we go along the path, and can find the
cheapest path without exploring the whole search tree.
◼ Guarantee first founded solution is the cheapest solution.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 53
By Dr. Khaled Wassif
 Uniform-cost Search (cont.)

5 2

1 4 1 7

Goal state
4 5

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 54

By Dr. Khaled Wassif
 Uniform-cost Search (cont.)
[x] = g(n)
path cost of node n

5 2
[5] [2]

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 55

By Dr. Khaled Wassif
 Uniform-cost Search (cont.)
[x] = g(n)
path cost of node n

5 2
[5] [2]
1 7
[3] [9]

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 56

By Dr. Khaled Wassif
 Uniform-cost Search (cont.)
[x] = g(n)
path cost of node n

5 2
[5] [2]
1 7
[3] [9]
4 5

[7] [8]
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 57
By Dr. Khaled Wassif
 Uniform-cost Search (cont.)
[x] = g(n)
path cost of node n

5 2
[5] [2]
1 4 1 7
[6] [3] [9]
[9]
4 5

[7] [8]
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 58
By Dr. Khaled Wassif
 Uniform-cost Search (cont.)

5 2
[5] [2]
1 4 1 7
[6] [3] [9]
[9]
Goal state 4 5
path cost
g(n)=[6] [7] [8]
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 59
By Dr. Khaled Wassif
 Uniform-cost Search (cont.)

5 2
[5] [2]
1 4 1 7
[6] [3] [9]
[9]
Goal state 4 5
path cost
g(n)=[6] [7] [8]
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 60
By Dr. Khaled Wassif
 Uniform-cost Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 61

By Dr. Khaled Wassif
 Uniform-cost Search (cont.)
◼ Complete? Yes.
◼ Optimal? Yes
– Whenever UCS selects a node for expansion, the optimal
path to that node has been found.
– Because step costs are non-negative, paths never get shorter
as nodes are added.
– Hence, the first goal node selected for expansion must be the
optimal solution.
◼ Time Complexity: O(bd+1)
◼ Space Complexity: O(bd+1)
– Because UCS can explore large trees of small steps before
exploring paths involving large and perhaps useful steps.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 62
By Dr. Khaled Wassif
 Depth-first Search (DFS)
◼ Always expands the deepest node in the current frontier
of the search tree.
– The search proceeds immediately to the deepest level of the
search tree, where the nodes have no successors.
– As those nodes are expanded, they are dropped from the
frontier, so then the search “backs up” to the next deepest
node that still has unexplored successors.
– Backtracks if search hits a dead-end (a non-goal node with no
expansion).
◼ An instance of the GRAPH-SEARCH algorithm in which
the frontier is implemented by using a LIFO stack.
– The nodes that are visited last will be expanded first.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 63
By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 64

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 65

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 66

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 67

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 68

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 69

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 70

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 71

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 72

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 73

By Dr. Khaled Wassif
 Depth-first Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 74

By Dr. Khaled Wassif
 Depth-first Search (cont.)

Goal state
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 75
By Dr. Khaled Wassif
 Depth-first Search (cont.)
◼ DFS is incomplete:
– In tree-search version, it can make a wrong choice if left
subtree were of unbounded depth but contained no solutions,
it would never terminate.
– But, the graph-search version, which avoids repeated states
and redundant paths, is complete only in finite state spaces.
◼ DFS is not optimal:
– It will explore the entire left subtree even if a goal node near
the root.
◼ Time complexity of DFS is O(bm) in worst case.
– But if solutions are dense, may be much faster than BFS.
◼ Space complexity of DFS is O(bm), i.e., linear space!
– It needs to store only a single path from the root to a leaf
node, along with the remaining unexpanded sibling nodes for
each node on the path.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 76
By Dr. Khaled Wassif
 Depth-limited Search (DLS)
◼ Alleviate the problem of DFS in infinite state spaces by
supplying it with a predetermined depth limit l.
– Nodes at depth l are treated as if they have no successors.
◼ DLS can be implemented as a simple modification to
the general tree- or graph-search algorithm; or instead
implemented as a simple recursive algorithm.
– It can terminate with two kinds of failure: standard failure or
the cutoff value indicate no solution within the depth limit.
◼ DLS is incomplete when the shallowest goal is beyond
the depth limit (l < d).
◼ DLS also is not optimal if l > d.
◼ Time complexity of DLS is O(b l) in worst case.
◼ Space complexity of DLS is O(bl).
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 77
By Dr. Khaled Wassif
 Recursive Depth-limited Search

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 78

By Dr. Khaled Wassif
 Iterative Deepening Search (IDS)
◼ Applies DLS repeatedly with increasing depth and
terminates when a solution is found or no solutions exists.
– Occur when the depth limit reaches d, the depth of the
shallowest goal node.
◼ Combines the benefits of BFS and DFS:
– Like BFS, it is complete when the branching factor is finite.
– Like DFS, the memory requirements are very reasonable.
◼ Seem wasteful - many states are expanded multiple times.
– Almost all of the nodes are in the bottom level, so it does not
matter much that the upper levels are expanded multiple times.

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 79

By Dr. Khaled Wassif
Iterative Deepening Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 80

By Dr. Khaled Wassif
Iterative Deepening Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 81

By Dr. Khaled Wassif
Iterative Deepening Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 82

By Dr. Khaled Wassif
Iterative Deepening Search (cont.)

Goal
state

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 83

By Dr. Khaled Wassif
 Iterative Deepening Search (cont.)
◼ IDS is complete:
– If there is a solution, IDS is guaranteed to find it.
– If there are several solutions, IDS will always find the
shallowest goal state (shortest path) first.
◼ IDS is optimal given the path cost is a non-decreasing
function of the depth of the node.
◼ Time complexity of IDS is O(bd).
◼ Space complexity of IDS is O(bd).
◼ In general, IDS is the preferred uninformed search
method when the search space is large and the depth of
solution is not known.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 84
By Dr. Khaled Wassif
 Bi-directional Search (BDS)
◼ Run two simultaneous search:
– One forward from the initial state and the other backward
from the goal.
– Hoping that the two searches meet in the middle.
◼ Motivation: bd/2 + bd/2 is much less than bd.
◼ Implemented by replacing the goal test with a check to
see whether the frontiers of the two searches intersect
– If the intersection exist then a solution has been found.
◼ How do backward search? (using predecessors function).
– The predecessors of a state x to be all those states that have x
as a successor.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 85
By Dr. Khaled Wassif
 Bi-directional Search (cont.)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 86

By Dr. Khaled Wassif
 Bi-directional Search (cont.)
◼ BDS is complete and optimal:
– If both searches are BFS.
– Other search combinations may lose completeness or
optimality.

◼ Time complexity of BDS is O(bd/2).

◼ Space complexity of BDS is O(bd/2):
– At least one of the frontiers must be kept in memory so that
the intersection check can be done.

◼ What kind of search is going in each half?

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 87
By Dr. Khaled Wassif
Comparison of Uninformed Search Strategies

Breadth- Uniform- Depth- Depth- Iterative Bidirectional

Criterion
First Cost First Limited Deepening (if applicable)

Complete? Yes Yes No No Yes Yes

Time O(bd) O(bd+1) O(bm) O(b l) O(bd) O(bd/2)

Space O(bd) O(bd+1) O(bm) O(b l) O(bd) O(bd/2)

Optimal? Yes Yes No No Yes Yes

b: Branching factor d: Depth of solution

m: Maximum depth l : Depth Limit

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 3- 88

By Dr. Khaled Wassif
Informed (Heuristic) Search Strategies
◼ An informed search strategy can find solutions more
efficiently than an uninformed strategy.
– Use problem-specific knowledge beyond the definition of the
problem itself
– The search cost may be cut considerably if search is more
goal-directed.
– Goal-directedness is typically provided by an evaluation
function that assigns a number to each node indicating how
the node is closer to the goal.
◼ The general approach we consider is called best-first
search.
– An instance of the general TREE-SEARCH or GRAPH-SEARCH
algorithm in which a node is selected for expansion based on
an evaluation function, f(n).
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 89
By Dr. Khaled Wassif
 Best-first Search
◼ A class of search strategies that expands nodes in order
of desirability as indicated by some evaluation function.
– The evaluation function is construed as a cost estimate, so the
node with the lowest evaluation is expanded first.
– Identical to that for uniform-cost search, except for the use of
f instead of g to order the priority queue.
◼ The choice of f determines the search strategy.
– Most best-first algorithms include as a component of f a
heuristic function, denoted h(n):
h(n) = estimated cost of the cheapest path from the state at node n to a
goal state.
h(n) = 0 if n is represent a goal state.
◼ Heuristic functions are the most common form in
which additional knowledge of the problem is reported
to the search algorithm.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 90
By Dr. Khaled Wassif
 Greedy Best-first Search
◼ Try to expand the node that is closest to the goal, on
the base that this is likely to lead to a solution quickly.
– Evaluate nodes by using just the heuristic function:
f(n) = h(n)
– Select the node that appears to be closest to goal among all
possible choices.
◼ Need to choose a particular problem, because heuristic
functions are problem-specific.
– For route-finding problem in Romania; we use the straight-
line distance (SLD) heuristic:
hSLD(n) = straight-line distance from n to Bucharest
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 91
By Dr. Khaled Wassif
 Romania with
Straight-line distance to Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 92

By Dr. Khaled Wassif
Greedy Best-first Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 93

By Dr. Khaled Wassif
Greedy Best-first Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 94

By Dr. Khaled Wassif
Greedy Best-first Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 95

By Dr. Khaled Wassif
Greedy Best-first Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 96

By Dr. Khaled Wassif
 Greedy Best-first Search (cont.)
◼ hSLD heuristic leads to minimal search cost:
– It finds a solution without ever expanding a node that is
not on the solution path.
◼ However, heuristic is not optimal:
– The path it found via Sibiu and Fagaras to Bucharest is 32
kilometers longer than the path through Rimnicu Vilcea and
Pitesti.
◼ This strategy prefers to take the biggest bite possible
out of the remaining cost to reach the goal, without
worrying about whether this will be best in the long run.
hence the name "Greedy Search"
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 97
By Dr. Khaled Wassif
 Greedy Best-first Search (cont.)
◼ Greedy search looks like DFS in the way it prefers to
follow a single path all the way to the goal, but suffers
from the same defects.
– It is not optimal.
– It is incomplete without systematic checking of repeated
states.
» Can start down an infinite path and never return to try other possibilities.
– Time and Space Complexity are O(bm) in the worst case
» Where b is the branching factor and m is the maximum path length.
◼ With a good heuristic function, the space and time
complexity can be reduced substantially.
– Amount of reduction depends on the particular problem and
quality of the heuristic function.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 98
By Dr. Khaled Wassif
 A* (A Star) Search
◼ Greedy Search (GS) minimizes a heuristic h(n) which
is an estimated cost from a node n to the goal state.
– GS is efficient but not optimal and incomplete.
◼ Uniform Cost Search (UCS) minimizes the actual cost
g(n) from the initial state to n.
– UCS is optimal and complete but not efficient.
◼ A* search combine GS and UCS to get an efficient
algorithm which is complete and optimal.
Evaluation function ƒ(n) = g(n) + h(n)
where: g(n) = exact cost to reach the node n
h(n) = estimated cost from the node n to goal
ƒ(n) = estimated total cost of path through n to goal
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 99
By Dr. Khaled Wassif
 A* Search (cont.)
I

g(n)

ƒ(n) = g(n) + h(n) n

h(n)

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 100

By Dr. Khaled Wassif
 A* Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 101

By Dr. Khaled Wassif
 A* Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 102

By Dr. Khaled Wassif
 A* Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 103

By Dr. Khaled Wassif
 A* Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 104

By Dr. Khaled Wassif
 A* Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 105

By Dr. Khaled Wassif
 A* Search for Bucharest

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 106

By Dr. Khaled Wassif
 Conditions for Optimality
◼ 1st condition: we require for optimality is that h(n) be
an admissible heuristic.
– An admissible heuristic is one never overestimates the cost to
reach the goal.
– A heuristic h(n) is admissible if for every node n: h(n) ≤ h*(n)
where h*(n) is the actual cost to reach the goal state from n.
◼ An admissible heuristic is optimistic
– Because it think the cost of solving the problem is less than it
actually is.
– Such that g(n) is the actual cost to reach n then f(n) never
overestimates the total true cost of a solution along the
current path through n.
– Example: hSLD(n) never overestimates the actual road distance.
◼ Theorem: If h(n) is admissible, A* using TREE-SEARCH
is optimal.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 107
By Dr. Khaled Wassif
 Conditions for Optimality
◼ 2nd condition: required only for applications of A∗ to
graph search is called consistency (or monotonicity).
– A heuristic is consistent (monotonic) if for every node n,
every successor n' of n generated by any action a:
h(n) ≤ c(n,a,n') + h(n')
◼ It is easy to show that every consistent
heuristic is also admissible.
– Consistency is a stricter requirement than admissibility.
– Example: hSLD is a consistent heuristic (general triangle
inequality is satisfied when each side is measured by the
straight-line distance).
◼ Theorem: If h(n) is consistent, A* using GRAPH-
SEARCH is optimal.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 108
By Dr. Khaled Wassif
 Properties of A*
◼ Optimal? Yes
– A∗ is optimally efficient for any given consistent heuristic.
– No other optimal algorithm is guaranteed to expand fewer
nodes than A∗.

◼ Complete? Yes
– Unless there are infinitely many nodes with f ≤ f(G).

◼ Time complexity? Exponential

◼ Space complexity? Keeps all expanded nodes in memory

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 109

By Dr. Khaled Wassif
 Iterative Deepening A∗ (IDA∗)
◼ Simplest way to reduce memory requirements for A∗
is to adapt the idea of iterative deepening to the
heuristic search context.
– Main difference between IDA∗ and standard IDS is that the
cut-off used is the f-cost (g + h) rather than the depth.
– At each iteration, the cut-off value is smallest f-cost of any
node that exceeded the cut-off on the previous iteration.
◼ IDA∗ is practical for many problems with unit step
costs and avoids the substantial overhead associated
with keeping a sorted queue of nodes.
◼ But, IDA∗ suffers from the same difficulties with real
valued costs as does the iterative version of uniform-
cost search.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 110
By Dr. Khaled Wassif
Recursive Best-First Search (RBFS)
◼ A simple recursive algorithm that attempts to simulate
the operation of standard best-first search, but using
only linear space.
– Its structure is similar to that of a recursive DFS, but it uses
the f-limit variable to keep track of the f-value of the best
alternative path available from any ancestor of current node.
– If the current node exceeds this limit, the recursion returns
back to the alternative path.
– As the recursion returns, RBFS replaces the f-value of each
node along the path with the best f-value of its children.
– In this way, RBFS remembers the f-value of the best leaf in
the forgotten subtree and can therefore decide whether it’s
worth reexpanding the subtree at some later time.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 111
By Dr. Khaled Wassif
Recursive Best-First Search (RBFS)
◼ RBFS is somewhat more efficient than IDA∗, but still
suffers from excessive node regeneration.
– The mind can change because every time the current best
path is extended, its f-value is likely to increase.
– When this happens, the second-best path might become the
best path, so the search has to backtrack to follow it.
◼ RBFS is an optimal algorithm if the heuristic function
h(n) is admissible.
◼ Its space complexity is linear in the depth of the
deepest optimal solution.
◼ Its time complexity depends both on the accuracy of
the heuristic function and on how often the best path
changes as nodes are expanded.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 112
By Dr. Khaled Wassif
 Admissible Heuristics
◼ E.g., for the 8-puzzle:
– h1(n) = number of misplaced tiles
– h2(n) = total Manhattan distance
(i.e., sum of the distances of tiles from their goal positions)

– h1(S) = ?
– h2(S) = ?
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 113
By Dr. Khaled Wassif
 Admissible Heuristics
◼ E.g., for the 8-puzzle:
– h1(n) = number of misplaced tiles
– h2(n) = total Manhattan distance
(i.e., sum of the distances of tiles from their goal positions)

– h1(S) = 8
– h2(S) = 3+1+2+2+2+3+3+2 = 18
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 114
By Dr. Khaled Wassif
 Dominance Heuristics
◼ If h1 and h2 are both admissible and h2(n) ≥ h1(n) for
all n then h2 dominates h1 .
◼ Domination translates directly into efficiency
– h2 is better for search
◼ Typical search costs: (average number of nodes expanded)
– d=12
» IDS = 3,644,035 nodes
» A*(h1) = 227 nodes
» A*(h2) = 73 nodes
– d=24
» IDS = too many nodes
» A*(h1) = 39,135 nodes
» A*(h2) = 1,641 nodes
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 115
By Dr. Khaled Wassif
 Comparison of Search Costs for
∗
IDS and A Algorithms with h1, h2

AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 116

By Dr. Khaled Wassif
 Relaxed Problems
◼ h1 and h2 are estimates of the remaining path length
for the 8-puzzle, but they are also perfectly accurate
path lengths for simplified versions of the puzzle.
– If the rules of the 8-puzzle are relaxed so that a tile can
move anywhere, then h1(n) gives the shortest solution.
– If the rules are relaxed so that a tile can move to any
adjacent square, then h2(n) gives the shortest solution.
◼ A problem with fewer restrictions on the actions is
called a relaxed problem.
– The state space graph of the relaxed problem is a
supergraph of the original state space.
◼ The cost of an optimal solution to a relaxed problem
is an admissible heuristic for the original problem.
AI: A modern Approach © 2010 S. Russell and P. Norving Slide 4- 117
By Dr. Khaled Wassif