1 of 10

ARTIFICIAL INTELLIGENCE

A* Search

Algorithm

Intelligent Pathfinding for Modern AI Systems

From Graph Theory to Autonomous Navigation

2 of 10

What is A* Search?

A* (pronounced "A-star") is a best-first search algorithm that finds the least-cost path from a start node to a goal node. It combines the strengths of Dijkstra's algorithm and greedy best-first search.

ORIGIN

Developed in 1968 by Peter Hart, Nils Nilsson, and Bertram Raphael

PROPERTY

Optimal and complete when using an admissible heuristic

APPROACH

Uses informed search with heuristics to guide exploration

Widely used in games, robotics, and navigation

3 of 10

The Core Formula

f(n) = g(n) + h(n)

f(n)

Total Estimated Cost

The evaluation function that determines which node to expand next

g(n)

Cost So Far

Actual cost from the start node to current node n

h(n)

Heuristic Estimate

Estimated cost from node n to the goal

Why This Works

A* always expands the node with the lowest f(n) value, balancing exploration of known paths with promising directions toward the goal.

4 of 10

Heuristic Functions

The heuristic h(n) guides the search. A good heuristic dramatically improves performance.

Manhattan

h = |x₁-x₂| + |y₁-y₂|

4-direction grid movement

✓ Tile puzzles, city maps

Euclidean

h = √[(Δx)² + (Δy)²]

Any-angle movement

✓ Robotics, drones

Diagonal

h = max(|Δx|, |Δy|)

8-direction grid

✓ Strategy games, chess

Admissibility: A heuristic is admissible if it never overestimates the true cost. This guarantees optimality.

5 of 10

How A* Works

Initialize

Add start node to OPEN list with f = h(start)

Select Best Node

Remove node with lowest f(n) from OPEN

Goal Check

If goal reached, reconstruct and return path

Expand Neighbors

For each neighbor, calculate f(n) = g(n) + h(n)

Update Lists

Add/update neighbors in OPEN, move current to CLOSED

Repeat

Continue from step 2 until goal or OPEN is empty

KEY DATA STRUCTURES

OPEN List

Priority queue of nodes to explore (sorted by f-value)

CLOSED List

Set of already explored nodes (prevents revisiting)

COMPLEXITY

Time

O(b^d)

Space

O(b^d)

b = branching factor, d = depth

6 of 10

A* in Action

Pathfinding on a 5×5 grid with obstacles

Search Process

A* expands nodes in order of their f-values, navigating around obstacles while minimizing total path cost.

KEY OBSERVATIONS

Only explores promising directions

Avoids exploring behind obstacles

Finds optimal path efficiently

Result

Path Length: 7

Nodes explored: 12 of 25

7 of 10

A* vs Other Algorithms

Algorithm

Optimal

Complete

Informed

A* Search

Dijkstra's

Greedy Best-First

BFS / DFS

WHY A* WINS

A* combines the optimality guarantee of Dijkstra's with the speed of greedy search by using heuristics to guide exploration toward the goal.

TRADE-OFF

A* requires more memory than DFS and computing heuristics adds overhead, but the dramatic reduction in nodes explored usually outweighs these costs.

BEST FOR

Single-source shortest path problems where a good heuristic is available and optimality is required.

8 of 10

AI Applications of A*

Robotics

Path planning for mobile robots, robotic arms, and autonomous vehicles navigating real-world environments.

3D navigation, obstacle avoidance

Video Games

NPC pathfinding, enemy AI, and real-time strategy games with thousands of units moving simultaneously.

Real-time, dynamic environments

GPS Navigation

Route planning in mapping applications, considering traffic, road types, and user preferences.

Multi-criteria optimization

Puzzle Solving

Solving 8-puzzles, Rubik's cubes, and other state-space search problems efficiently.

State space exploration

50+

YEARS IN USE

100+

VARIANTS DEVELOPED

#1

PATHFINDING ALGORITHM

9 of 10

Modern AI Integration

A* is being enhanced with modern AI techniques

Neural Network Heuristics

Deep learning models learn better heuristic functions from data, improving A* performance in complex domains.

Reinforcement Learning + A*

RL agents use A* for planning while learning value functions that serve as heuristics.

LLM-Guided Planning

Large language models propose actions while A* verifies and optimizes the resulting plans.

FEATURED: AUTONOMOUS VEHICLES

Perception

Neural networks detect obstacles

Planning

A* finds optimal trajectory

Control

RL refines execution in real-time

A* provides the backbone for safe, optimal navigation

💡 Key Insight: Classical algorithms like A* provide guarantees that pure neural approaches cannot match.

10 of 10

SUMMARY

Key Takeaways

Optimal & Complete

A* guarantees the shortest path when using an admissible heuristic

Heuristic is Key

The quality of h(n) determines search efficiency

Widely Applicable

Games, robotics, navigation, and AI planning systems

Combines Best of Both

Dijkstra's completeness + Greedy's speed

Evolving with AI

Neural heuristics and hybrid approaches extend its power

"A* remains the gold standard for informed search in AI"