Planning and Decision

Making
Module 3
The Planning Problem Formulation provides a structured way to describe such
problems so that they can be solved using AI techniques.
Components of a Planning Problem
Components of a Planning Problem
No. Component Explanation (based on Grid Diagram)
Starting position of the agent.
1 Initial State
In diagram: (0,0) — bottom-left corner, where the robot starts.
Target or destination the agent must reach.
2 Goal State
In diagram: (4,4) — top-right corner marked by a flag.
Legal actions available to the agent.
3 Operations In grid: Move Up, Move Down, Move Left, Move Right (as long as not blocked
by a wall or out of bounds).
All possible positions the agent can occupy.
4 State Space
In a 5×5 grid: 25 cells – some are invalid (gray cells = walls).
A sequence of operations leading from the initial state to the goal.
5 Plan
Black path in the diagram shows the planned movement.
Execution of the plan step by step.
6 Plan Execution
Robot physically or virtually follows the path: (0,0) → (0,1) → … → (4,4).
Algorithm used to find the optimal path. Could be BFS, DFS, or A* (most likely,
7 Search Algorithm
since it avoids walls and finds shortest path).
Estimations used to guide search algorithms.
8 Heuristic Functions
Example: Manhattan Distance → h(n)=∣xgoal−xn∣+∣ygoal−yn∣
Plan Validation & Ensuring the plan is valid (no wall collisions), and modifying if necessary.
9
Refinement E.g., avoid wall at (1,1) and (3,2) by refining path.
Action Execution Monitoring whether each action is performed correctly.
10
Monitoring Ensures robot moves as planned, detects deviation.
Stores environment knowledge: walls, grid size, robot capabilities.
11 Knowledge Base
E.g., knowing (1,4) is a wall helps in planning.
 State Space
Representation
 State Space Representation
 State space representation is a formal model of a problem in Artificial Intelligence,
 where:
 The environment is described as a set of states.
 Actions cause transitions between these states.
 The goal is to find a sequence of actions that leads from an initial state to a goal state.
 It is the foundation for search-based problem solving in AI.
 Why Use State Space?
 To solve a problem using AI, you first define the problem clearly in a form that a machine
can understand. This is what state space does—it models the problem in terms of:
 What the world looks like (states)
 What the agent can do (actions)
 What it wants to achieve (goal)
 Elements of State Space Representation
Component Description
All possible situations/configurations of the
States (S)
environment or problem

Initial State (s₀) The state where the problem solving starts

Actions (A) Operations the agent can perform to change states

Transition
Describes how actions transform one state into another
Function (T)

Goal Test (G) A function that checks if a state is a goal state

Path Cost Numeric cost associated with a path; used in optimal

(optional) search problems
Formal Notation
 Example: Robot Navigation in a Grid

Component Example
States (S) Positions on grid: (x, y)
Initial State (0,0)
Goal State (4,4)
Move_Up, Move_Down,
Actions (A)
Move_Left, Move_Right
Transition Func. Result((x,y), Move_Up) = (x, y+1)
Example : Making Instant Noodles
Boil Water ⟶(WaterBoiled)⟶ Add Noodles
Add Noodles ⟶(NoodlesInPot)⟶ Cook Noodles
Add Spices ⟶(SpicesInPot)⟶ Cook Noodles
Cook Noodles ⟶(NoodlesCooked)⟶ Serve Noodles
Example :
Blocks World – Stack A on B on C
Example
Best Path: A-B-D-I
Example 2
Limitation
Alpha-Beta Pruning
 Utility Theory
 Utility Theory in Artificial Intelligence is a mathematical framework used to
help intelligent agents make rational decisions under uncertainty by evaluating
the outcomes of their actions based on utility values (preferences).
 It allows the agent to choose the best possible action by maximizing the expected
utility — a combination of how desirable an outcome is and how likely it is to
occur.
Example: Self-Driving Car Decision
A self-driving car approaches a traffic light. It must decide whether to:
•Action A: Speed up and cross before the light turns red
•Action B: Slow down and stop
Each action has uncertain outcomes and different utilities.
Possible Outcomes and Utilities:

Action (a) Outcome (s) Probability P(s/a) Utility U(s)

A (Speed up) Cross safely 0.7 100
Accident 0.3 0
B (Stop) Wait and go safely 1.0 80

Expected Utility : Maximum Expected Utility :

EU(a1) = 0.7*100 + 0.3*0 = 70 Since Action a2 has higher
EU(a2) = 1.0*80 = 80 expected utility (80) than
Action a1 (70), the self-driving
car should choose to stop.
Limitation of Utility Theory
Limitation Explanation Example
In healthcare AI, assigning a numeric value
Difficult to Define Utility Quantifying all preferences and outcomes is
to relive pain quickly with side effects or
Function complex and subjective
relive pain slowly with no side effects
Incomplete or Uncertain Probabilities of outcomes are often A robot navigating an unfamiliar building
Knowledge unknown or hard to estimate without a reliable map
Evaluating all possible actions and
Real-time game AI can't compute utilities
Computational Complexity outcomes can be time-consuming and
for every possible move
resource-intensive
Utilities and probabilities may change over Stock market prediction where trends and
Dynamic Environments
time values shift frequently
Humans do not always act logically or AI modeling consumer behavior fails
Irrational Human Behavior
consistently with utility maximization because people buy emotionally

Multi-Objective Decision Balancing conflicting goals (e.g., cost vs. Self-driving car needs to balance speed,
Making safety) is difficult passenger comfort, and fuel economy

Assigning utility to outcomes involving AI deciding between saving one person vs

Ethical and Moral Concerns
people may be ethically sensitive many in accident scenarios

Uncertainty in Probability Probabilities may be based on incomplete AI predicting rare diseases with limited
Estimation data or assumptions historical data
 Example 2: Robot Delivery Decision
 Scenario:
 A delivery robot must choose a route:
 Action A: Take the short route – fast, but may encounter traffic.
 Action B: Take the long route – slower, but still has uncertainty (may be
delayed due to construction).

Find Expected utility and Maximum Expected utility and state the
decision
Action A: Take the short route – fast, but may encounter traffic.
Action B: Take the long route – slower, but still has uncertainty (may be delayed
due to construction).

Action Outcome Probability Utility

A Arrive on time
Delayed (traffic)
B Arrive safely (slow)
Arrive late
Action Outcome Probability Utility
A Arrive on time 0.6 100
Delayed (traffic) 0.4 30
B Arrive safely (slow) 0.8 80
Arrive late 0.2 30
Decision Network
 Decision networks are graphical models used to represent and solve decision-making
problems.
Components of Decision Network
•Chance Nodes: Represent random variables and their possible values, capturing the
uncertainty in the decision-making process.

•Decision Nodes: Represent the choices available to the decision-maker.

•Utility Nodes: Represent the utility or value of the outcomes, helping to evaluate and
compare different decision paths.
 Arcs in the Graph
 Arcs to Decision Nodes: Represent information available when the decision is
made.
 Arcs to Chance Nodes: Represent probabilistic dependence.
 Arcs to Utility Nodes: Represent what the utility depends on.

Expected Utility :

Maximum Expected Utility (MEU)

Example of Representing a Decision Problem
 Consider a simple decision network for deciding
whether an agent should take an umbrella when going
out. The agent’s utility depends on the weather
(Sunny or Rainy)and whether he takes an umbrella.
Draw the Decision network and state the conclusion.
Example of Representing a Decision Problem
 Consider a simple decision network for deciding
whether an agent should take an umbrella when going
out. The agent’s utility depends on the weather and
whether he takes an umbrella.
 Utility table

Probability Table
Maximum Expected Utility

MEU = 70…
The Umbrella can be leave
 Problem:
 A student is deciding whether to study or party the night
before an exam. The exam could be easy or hard, which
affects their score and overall satisfaction. Draw the decision
Network and state the decision.
 Probabilities:

Exam Difficulty Probability

Easy 0.6

Hard 0.4

Utility Table:
Activity Difficulty Utility
Study Easy 80
Study Hard 70
Party Hard 30
Party Easy 90