From Classroom to Research: How 2025 Papers Are Revolutionizing A* Search and Agentic AI

Muhammad Aqsam Qureshi — Sat, 14 Mar 2026 09:57:35 +0000

Introduction

Hi! I'm M Aqsam Qureshi, a student at FAST University. This blog post is for my Artificial Intelligence course assignment, where I'm reviewing two research papers from 2025.

Paper 1: A* Algorithm with Adaptive Weights

What is the Goal of This Paper?

The traditional A* algorithm that we studied in class has three main problems:

Too Many Search Nodes - Wastes time and battery power
Excessive Turning Points - Robots cannot follow zig-zag paths
Local Optima Traps - Gets stuck in suboptimal routes

Goal of this paper: Solve these problems so robots can navigate efficiently.

Four Improvements:

Diagonal-Free Search - Near obstacles, avoid diagonal moves for safety
Adaptive Weights - Weight changes based on how many obstacles are nearby
Heuristic Reward Values - Helps escape local optima (like Simulated Annealing)
B-Spline Smoothing - Makes paths smooth so robots can follow easily

Results:

Search Nodes: 76.4% LESS
Turn Angle: 71.7% LESS
Computation Time: 10% FASTER

Paper 2: The Rise of Agentic AI

What is the Goal of This Paper?

Agentic AI = AI that can think and act independently.

Traditional AI vs Agentic AI:

Responds to commands vs Takes initiative
Waits for input vs Perceives environment
Handles single task vs Handles multiple goals
No memory vs Learns from experience

Five Patterns of Agentic AI:

Tool Use - AI can use calculators, APIs, search engines
Reflection - AI learns from its mistakes
Re Act - Reason and Act in cycles (think then do)
Planning - Break complex goals into smaller steps
Multi-Agent - Multiple AI systems work together

Seven Types of Agents:

Simple Reflex - Room thermostat
Model-Based - Self-driving car
Goal-Based - Rescue robot
Utility-Based - Stock trading bot
Learning Agent - Netflix recommendation system
Hierarchical - Factory automation robot
Multi-Agent - Swarm of drones

Course Connection (VERY IMPORTANT)

Connection to Part A: Rescue Robot Scenario

Partially observable environment Agentic AI works in such environments
Dynamic environment Yes
Stochastic environment Yes
Multi-agent environment Yes

Why we chose Utility-Based Agent? Paper 2 confirms that utility-based agents work best in complex environments like our flood scenario.

Connection to Part B: Search Algorithms

"A* searches too many nodes" → 76.4% reduction Yes
"Paths have too many sharp turns" → 71.7% reduction Yes
"Different terrains need different costs" → Adaptive weights based on obstacle density Yes

Connection to Part C: Simulated Annealing

Remember this formula from Part C? P(accept) = e^(-ΔE / T)

Paper 1 uses the SAME concept! When the algorithm gets stuck in a local optimum, it temporarily accepts "worse" paths to explore better options - exactly like Simulated Annealing!

Connection to CSPs

Paper 1's Grey Wolf Optimizer treats weight adjustment as an optimization problem - exactly like we formulated survivor prioritization in Part C.

Personal Insight: Manual Reading vs NotebookLM

Phase 1: Manual Reading (Confusion)

When I first read these papers by myself:

The mathematics of Grey Wolf Optimizer was difficult to understand
Technical terms like "Re Act" and "Reflection" were confusing
I felt overwhelmed by the complexity

Honest feeling: "Will I ever be able to understand research papers?"

Phase 2: NotebookLM (Clarity)

I uploaded both papers to Google NotebookLM (notebooklm.google.com) and asked:

"Explain Grey Wolf Optimizer in simple terms with an example"
"Create a table comparing all 7 agent types"
"Summarize the five operational patterns in simple words"
"Show me the results in a bullet list"

What NotebookLM gave me:

Simple explanations with everyday examples
Clean tables and easy comparisons
Clear summaries of complex sections
Citations back to the original paper

Example: NotebookLM explained Grey Wolf Optimizer as:
"Imagine a pack of wolves hunting. The alpha wolf leads, beta and delta help, and omega follows. They search (explore), surround (exploit), and attack (converge). GWO mimics this behavior."

So easy to understand!

Phase 3: Back to the Paper (Understanding)

With NotebookLM's explanations, I read the papers again. Everything became clear!

Key Realization:

NotebookLM helped me understand 30% of the paper
I understood the remaining 70% by reading myself
AI didn't replace learning - it accelerated learning

What I Found Most Interesting

From Paper 1: Grey Wolf Optimizer is inspired by wolf hunting behavior! Nature + algorithms = beautiful combination.

From Paper 2: Self-improving AI agents that get better without human intervention - science fiction is becoming reality!

Both Papers: They take what we learned in class and show how real research improves it. This makes me feel confident about my choice of field.

Video Walkthrough

I made video explaining these papers. Watch here:

[https://youtu.be/794eacYVp_U]

Comparison Table: Traditional vs Improved A*

Heuristic Weight: Fixed vs Adaptive (Grey Wolf Optimizer)
Movement: 8-direction vs 5-direction near obstacles
Local Optima: Gets stuck vs Reward values for escape
Path Smoothness: Sharp turns vs B-spline curves
Search Nodes: High vs 76.4% less
Turn Angle: High vs 71.7% less

Why This Matters for AI Students

We often think that what we learn in class is the final truth. These papers prove:

Algorithms constantly improve - A* was created in 1968, still improving in 2025!
Research builds on fundamentals - What we study is the foundation
Interdisciplinary thinking works - Wolf hunting behavior inspired an algorithm!
Course material is essential - Without basics, you cannot understand research

Conclusion

Both papers showed me the depth of AI:

Paper 1 showed that even classic algorithms like A* have room for 76% improvement when you think creatively.

Paper 2 showed that agentic AI - which sounds like science fiction - is already here with clear patterns, types, and frameworks.

Course connection proved that what we are learning isn't outdated theory - it's the foundation for cutting-edge research.

About Me

Name: M Aqsam Qureshi
University: FAST University
Course: Artificial Intelligence

Follow me:

Hashnode:https://hashnode.com/@aqsamqureshi
Dev.to: [https://dev.to/muaqqu]

Tagging @raqeeb_26(dev.to) and @raqeebr (Hashnode) as per assignment requirement.

DEV Community: Muhammad Aqsam Qureshi

From Classroom to Research: How 2025 Papers Are Revolutionizing A* Search and Agentic AI