Two Friends Meeting in Saudi Arabia 🇸🇦

Multi-Agent Search Problem with Real-World Context

Coordination, heuristic analysis, and graph theory application

Problem Statement

Two friends live in different cities on a map of Saudi Arabia and want to meet as quickly as possible.

R
—
M
—
D


J
—
K
—
A

R=Riyadh, M=Mecca, D=Dammam, J=Jeddah, A=Abha, K=Khobar
Rules
  • Simultaneous movement: Each turn, both move to neighboring cities
  • Turn time: Max of both travel times (slower sets pace)
  • Goal: Meet in same city as quickly as possible
Learning Outcomes

By completing this exercise, you will be able to:

  • Formulate multi-agent problems with states, actions, and coordination costs
  • Analyze heuristic admissibility using lower bounds and mathematical reasoning
  • Apply graph theory to determine solution existence and revisit requirements
How to Use This Exercise

Each question provides two levels of explanation:

  • 🔵 Simple Response: Easy-to-understand explanation using everyday language and examples
  • 🔧 Technical Solution: Detailed mathematical analysis with formal proofs and step-by-step reasoning
Choose the level that matches your comfort with technical details!
  • Question: Formulate as a search problem: state, actions, transition model, step cost, goal test?
Simple Response

Think of this like planning a meetup with GPS tracking:

  • State = Where are both friends right now? Like "(Ahmed in Riyadh, Kamal in Jeddah)"
  • Actions = Where can they go next? Each friend picks a neighboring city to drive to
  • Cost = How long does each round take? Wait for the slower friend (if Ahmed takes 2 hours, Kamal takes 3 hours, then this round takes 3 hours)
  • Goal = When are they done? When both friends are in the same city
  • Problem size = All possible combinations If there are 10 cities, there are 10×10 = 100 possible "where both friends are" situations
Technical Solution - Search Formulation
1 State: $(a, b)$ where $a, b$ = current cities of friends A & B
2 Actions: $(a', b')$ where $a' \in N(a), b' \in N(b)$ (joint moves to neighbors)
3 Step cost: $c = \max\{d(a,a'), d(b,b')\}$ (slower traveler sets turn time)
4 Goal test: $a = b$ (same city) | State space: $|V|^2$ states
Explanation - Multi-Agent Problem Formulation

Think of this as coordinating two people with GPS tracking:

🏠 State like a "snapshot":
  • State (Riyadh, Jeddah): "Ahmed is in Riyadh, Kamal is in Jeddah right now"
  • State (Mecca, Mecca): "Both are in Mecca - they've met!" (Goal achieved)
  • Why pairs?: You need to know where BOTH friends are to plan the meeting
🚗 Actions like "travel plans":
  • Action (Mecca, Dammam): "Ahmed goes to Mecca, Kamal goes to Dammam"
  • Neighbors only: You can only drive to directly connected cities
  • Joint decision: Both friends decide their next city together
⏱️ Why "max" cost makes sense:
Example: Ahmed takes 2 hours (Riyadh→Mecca), Kamal takes 3 hours (Jeddah→Taif). They agreed to coordinate, so Ahmed waits 1 hour in Mecca until Kamal arrives in Taif. Total turn time = 3 hours (the slower one).
🎯 How to solve similar problems:
1. State = all information needed to make decisions
2. Actions = what can change in one step
3. Cost = what you're trying to minimize (here: meeting time)
4. Goal = when you're done (here: same location)
  • Question: Given straight-line distance $D(i,j)$, which are admissible: $h_1 = D(i,j)$, $h_2 = 2D(i,j)$, $h_3 = D(i,j)/2$?
Simple Response

Think of heuristics like GPS time estimates - they should never lie by saying "too easy":

  • Best case scenario: Friends run directly toward each other in a straight line
  • Meeting time: They meet halfway, so each travels half the distance
  • Why h₁ = D fails: Says "it takes the FULL distance time" but friends help each other by moving toward each other
  • Why h₂ = 2D fails: Even worse - says it takes TWICE the full distance (way too pessimistic!)
  • Why h₃ = D/2 works: Says "at least half-distance time" which is true - they might take longer due to roads, but never less than this

Answer: Only h₃ = D/2 is admissible because it never underestimates the meeting time.

Technical Solution - Admissibility Analysis
1 Admissibility requirement: A heuristic $h(s)$ is admissible if $h(s) \leq h^*(s)$ for all states $s$
Where $h^*(s)$ = true optimal cost from state $s$ to goal
2 Physical lower bound derivation:
Consider friends at cities $i$ and $j$ with straight-line distance $D(i,j)$
Best case scenario: Both move directly toward each other at maximum speed
Meeting point: Halfway between $i$ and $j$
Distance per friend: $D(i,j)/2$
Therefore: $h^*(i,j) \geq D(i,j)/2$ (fundamental physical constraint)
3 Analysis of $h_1 = D(i,j)$:
Claim: Not admissible
Strategy: To prove non-admissibility, we need to show there exists at least one case where $h_1 > h^*$ (the heuristic overestimates the true cost).
Proof by counterexample:
• What we have: $h_1(i,j) = D(i,j)$ and we established $h^*(i,j) \geq D(i,j)/2$
• What we need for admissibility: $h_1 \leq h^*$, which means $D(i,j) \leq h^*(i,j)$
• The problem: Our lower bound tells us $h^*(i,j)$ could be as small as $D(i,j)/2$
• Counterexample scenario: Consider the ideal case where friends meet exactly halfway
  - True optimal cost: $h^*(i,j) = D(i,j)/2$ (each friend travels half the distance)
  - Our heuristic predicts: $h_1 = D(i,j) = 2 \cdot D(i,j)/2 = 2 \cdot h^*(i,j)$
• Mathematical violation: $h_1 = D(i,j) > D(i,j)/2 = h^*(i,j)$
• Conclusion: Since we found a case where $h_1 > h^*$, heuristic $h_1$ is not admissible
4 Analysis of $h_2 = 2D(i,j)$:
Claim: Not admissible
Intuition: If $h_1 = D(i,j)$ already overestimates, then $h_2 = 2D(i,j)$ (twice as large) will overestimate even more severely.
Proof by direct comparison:
• Given: $h_2(i,j) = 2D(i,j)$ and we know $h^*(i,j) \geq D(i,j)/2$
• Admissibility requires: $h_2 \leq h^*$, which means $2D(i,j) \leq h^*(i,j)$
• The gap analysis: We need to check if this inequality can ever be violated
• Critical observation: Our physical constraint only guarantees $h^*(i,j) \geq D(i,j)/2$
• Worst-case scenario: When friends meet optimally at the halfway point
  - True optimal cost: $h^*(i,j) = D(i,j)/2$
  - Heuristic prediction: $h_2 = 2D(i,j)$
  - Ratio comparison: $h_2/h^* = 2D(i,j) / (D(i,j)/2) = 4$
• Mathematical violation: $h_2 = 2D(i,j) = 4 \cdot D(i,j)/2 = 4 \cdot h^*(i,j) > h^*(i,j)$
• Conclusion: $h_2$ overestimates by a factor of 4, making it severely non-admissible
5 Analysis of $h_3 = D(i,j)/2$:
Claim: Admissible
Strategy: To prove admissibility, we must show $h_3 \leq h^*$ holds for ALL possible states. We'll use our established lower bound.
Proof by universal inequality:
• What we established in Step 2: $h^*(i,j) \geq D(i,j)/2$ (this is our fundamental physical constraint)
• What our heuristic predicts: $h_3(i,j) = D(i,j)/2$
• The key insight: Our heuristic exactly matches the lower bound we derived
• Mathematical relationship: $h_3(i,j) = D(i,j)/2 \leq h^*(i,j)$ for all states $(i,j)$
• Why this works:
  - In the best case: $h^*(i,j) = D(i,j)/2$ exactly → $h_3 = h^*$ (perfect estimate)
  - In realistic cases: $h^*(i,j) > D(i,j)/2$ → $h_3 < h^*$ (conservative estimate)
  - Never happens: $h^*(i,j) < D(i,j)/2$ → impossible due to physics
• Admissibility verification: Since $h_3 \leq h^*$ always holds, $h_3$ satisfies the admissibility condition
• Conclusion: $h_3$ is a valid admissible heuristic that provides a safe lower bound
6 Final Summary and Comparison:
Remember: Admissibility means the heuristic never overestimates. Even one counterexample breaks admissibility!
Results overview:
• $h_1 = D(i,j)$: ❌ Not admissible
  - Problem: Overestimates by factor of 2 in optimal scenarios
  - Why it fails: Assumes one friend travels the full distance (ignores simultaneous movement)
• $h_2 = 2D(i,j)$: ❌ Not admissible
  - Problem: Overestimates by factor of 4 in optimal scenarios
  - Why it fails: Even worse than $h_1$, completely ignores coordination benefits
• $h_3 = D(i,j)/2$: ✅ Admissible
  - Success: Matches the fundamental physical lower bound
  - Why it works: Represents the best-case scenario (perfect coordination)
Key insight: The only admissible heuristic is the one that equals our derived lower bound: $h_3 = D(i,j)/2$
h₁ = D(i,j)
❌ Not Admissible
Often overestimates
h₂ = 2D(i,j)
❌ Not Admissible
Even worse overestimate
h₃ = D(i,j)/2
✅ Admissible
Safe lower bound
Explanation - Understanding Heuristic Admissibility

Think of admissibility as "never lying about how easy the problem is":

🏃‍♂️ The physics argument for D/2:
  • Best case scenario: Friends run directly toward each other in a straight line
  • Meeting point: They meet exactly halfway between their starting points
  • Time needed: Each travels distance D/2, so meeting time = D/2
  • Reality check: Roads aren't straight lines, so real time ≥ D/2
❌ Why h₁ = D fails:
  • Problem: Says meeting takes D time (full distance)
  • Reality: Both friends move toward each other simultaneously
  • Example: Riyadh-Jeddah distance = 1000km. h₁ = 1000km, but they can meet in Mecca (500km each) much faster
🧭 Navigation GPS analogy:
A good heuristic is like GPS estimated time: it might say "at least 30 minutes" and you actually take 35 minutes (fine!). But if GPS said "at least 60 minutes" when you actually take 35, you'd be frustrated - that's an overestimate like h₁ and h₂.
🎯 How to design admissible heuristics:
1. Find physical constraints: What's the absolute best case?
2. Use straight-line distances: Reality can only be equal or worse
3. Consider simultaneous movement: Don't forget both agents move together
4. Test with examples: Can you find any case where h > actual cost?
  • Question: Could there be a connected road map where no meeting is possible?
Simple Response

Think of this like asking "Can you drive between any two places?"

  • Connected road map: You can drive from anywhere to anywhere else (maybe not directly, but through other cities)
  • Can friends always meet? YES! Here's why:
    • Pick any city as the meeting spot (say, Riyadh)
    • Since roads connect everything, both friends can drive to Riyadh
    • They coordinate to arrive at the same time - meeting successful!
  • When is meeting impossible? Only when roads are broken/missing between friend locations
  • Example: Ahmed in mainland Saudi, Kamal on an island with no bridge - they literally can't reach each other

Answer: No, there cannot be a connected road map where meeting is impossible. Connected = meeting always possible!

Technical Solution - Graph Connectivity Analysis
1 Setup: Graph $G = (V, E)$ where $V$ = cities, $E$ = roads
Question: Given connected $G$, can friends always meet?
2 Claim: If $G$ is connected, meeting is always possible
Constructive proof:
• Choose any rendezvous city $r \in V$
• Since $G$ connected: paths $P_A: a_0 \leadsto r$ and $P_B: b_0 \leadsto r$ exist
• Friends coordinate to both reach $r$ (possibly with detours for synchronization)
• ∴ Solution exists
3 When meeting is impossible:
Disconnected case:
• Friends in different components $C_i, C_j$ where $C_i \cap C_j = \emptyset$
• No path between components → No common reachable city
• ∴ Meeting impossible
4 Answer: No - connected graphs always allow meetings
Verification: Run BFS from $a_0$. If $b_0$ reachable → meeting possible. Complexity: $O(|V| + |E|)$
Explanation - Graph Connectivity and Solution Existence

Think of connectivity as "can you drive from anywhere to anywhere else?":

🗺️ Connected graph = guaranteed meeting:
  • Proof by example: Pick any city as meeting point (say, Riyadh)
  • Friend A's path: Since graph is connected, there's a route from A's city → Riyadh
  • Friend B's path: Since graph is connected, there's a route from B's city → Riyadh
  • Coordination: They can synchronize to both arrive at Riyadh
🏝️ Disconnected = impossible meeting:
  • Island scenario: Ahmed in mainland Saudi Arabia, Kamal on an island with no bridge
  • Different components: No path exists between their locations
  • Result: No matter how they move, they can never reach the same city
🔍 How to check connectivity:
Algorithm: Start from Friend A's location, use BFS/DFS to explore all reachable cities. If you can reach Friend B's location, they're in the same connected component → meeting is possible!
🎯 Real-world examples:
• Saudi Arabia: Highly connected road network → meetings always possible
• Island nations: Without ferries/bridges, some meetings impossible
• War zones: Destroyed roads can create disconnected regions
• Internet routing: Same principle applies to data packet delivery
  • Question: Must one friend visit the same city twice before meeting?
Simple Response

Think of this like coordinating two people who must move at the same time:

  • The setup: Hub city H with 3 side streets (L₁, L₂, L₃). Ahmed starts at H, Kamal starts at L₁
  • The goal: Both arrive at the same place after the same number of moves
  • The problem: Kamal needs 1 move (L₁→H), but Ahmed can't just wait at H!
  • The solution: Ahmed must "waste time" by going H→L₂→H (2 moves) while Kamal goes L₁→H→somewhere→H (2 moves)
  • The result: Ahmed visited H twice (start and end)!
  • Why this happens: When you can't wait and need to synchronize timing, sometimes you must backtrack
  • Real-world: People can usually wait, so this isn't common in real life

Answer: Yes, sometimes one friend must revisit cities - but only when "waiting" isn't allowed and timing needs to be synchronized.

Technical Solution - City Revisit Analysis
1 Key modeling choice: Answer depends on whether "wait" actions are allowed
• Model A: Wait action allowed
• Model B: No wait (must move every turn)
We'll prove Model B can require revisits
2 Counterexample construction:
Star graph: Hub $H$ connected to leaves $L_1, L_2, L_3$
Initial positions: Friend A at $H$, Friend B at $L_1$
Constraint: Both must move simultaneously with equal total moves
3 Movement sequence forcing revisit:
Strategy: Both make 2 moves to meet at hub $H$
• Solution sequence:
  - A: $H \rightarrow L_2 \rightarrow H$ (2 moves)
  - B: $L_1 \rightarrow H \rightarrow H$ (but can't wait, so needs alternative)
• Result: A visits $H$ twice (start and end) to enable synchronization
4 General condition: Revisits required when:
$|d_A - d_B| > 0 \land \text{no wait} \land \text{limited connectivity}$
Model comparison: With waiting: A stays at $H$, B moves $L_1 \rightarrow H$ (no revisits). Without waiting: synchronization forces detours.
Answer: Yes - revisits necessary under restrictive action models
H

L₁
—
L₂
—
L₃

Star: requires timing coordination
Explanation - The Synchronization Dilemma

Think of this as a timing coordination problem:

⏰ The timing mismatch problem:
  • Friend B needs 1 move: L₁ → H (arrives at Hub in 1 turn)
  • Friend A is already at H: If A just waits, they could meet immediately
  • But "wait" isn't allowed: A must move every turn (game rules)
  • A's solution: Go H → L₂ → H (takes 2 turns, same as B)
🔄 Why revisit happens:
  • Turn 1: A: H→L₂, B: L₁→H (A visits L₂, B reaches H)
  • Turn 2: A: L₂→H, B: already at H (A revisits H, they meet!)
  • Result: A visited H twice (start + end)
🎭 Dancing analogy:
Imagine two dancers who must move to the same beat. If one dancer needs 1 step to reach the center and the other is already there, the one at the center must "mark time" by stepping away and back, so they both arrive simultaneously.
🔧 Modeling choice impact:
• With "wait" action: A stays at H, B comes to H → No revisits needed
• Without "wait" action: Must move every turn → Revisits for synchronization
• Real world: People can usually wait, so revisits are often unnecessary
• Game/robotics: Sometimes continuous movement is required
📝 Problem-solving tip:
When analyzing multi-agent problems, always consider: Can agents wait? Can they move at different speeds? These modeling choices drastically affect solution complexity.

Continue Learning

Robot Navigation Heuristics A* Search Maze Exercise