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.
Top comments (0)