Download Lecture 15

the
gamedesigninitiative
at cornell university
Lecture 15:
Artificial Intelligence
Assignment 9 due tonight
 Submit through CMS
2
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Second Prototype Due
 Thursday, October 23rd
 In-class playtesting
3
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Logging release later today
 Will send out an email
 Needed for first release (Nov. 4th)
4
Artificial Intelligence
the
gamedesigninitiative
at cornell university
 Scripts
 Planning
 Pathfinding
5
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Scripts
1. IF <condition is true> THEN <perform action>
2. IF <condition is true> THEN <perform action>
3. IF <condition is true> THEN <perform action>
6
Artificial Intelligence
the
gamedesigninitiative
at cornell university
First boss in Final Fantasy VII
7
Artificial Intelligence
the
gamedesigninitiative
at cornell university
First boss in Final Fantasy VII
 If (Count == 0 OR Count == 2) Then
 SelectedTarget = random opponent
 Use Search Scope on SelectedTarget
 Count = Count + 1
 If (Count == 1 OR Count == 3) Then
 With probability 2/3:
 If Self HP < (Self Max HP / 2) Then
 Use Scorpion Tail on SelectedTarget
 Else
 Use Rifle on SelectedTarget
 With 1/3 Chance:
 Use Scorpion Tail on SelectedTarget
 Count = Count + 1
8
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Example: Tic Tac Toe
1. If I can win, win
2. If opponent can win,
prevent it
3. If center is available,
take it
4. If corner is
available, take it
9
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Advantages/Disadvantages
Easy to implement
Can express complex behaviors
“Smart” behavior can be very complex
Not so scalable
No natural way to vary difficulty
10
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Adversarial Search
Now
I move
Opponent
moves
I11move
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Importance of search depth in Chess
 Novice: ~4
 Master: ~8
 Grandmaster: ~12
 Deep Blue: 6 – 40
12
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Advantages/Disadvantages
Potentially much smarter
Natural way to vary difficulty
State-space explosion
Game must have certain properties
Unclear what to do if you can’t “see” the end
13
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Heuristics: Chess
 Pawn: 1 point
 Knight & Bishop: 3 points
 Rook: 5 points
 Queen: 11 points
14
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Heuristic Planning
Now
I move
Opponent
moves
15
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Heuristic Planning
Now
I move
Opponent
moves
16
11
10
5
14
Artificial Intelligence
2
20
3
16
the
gamedesigninitiative
at cornell university
Heuristic Planning
Now
I move
Opponent
moves
17
5
11
3
2
10
5
14
Artificial Intelligence
2
20
3
16
the
gamedesigninitiative
at cornell university
Heuristic Planning
Now
5
I move
Opponent
moves
18
5
11
3
2
10
5
14
Artificial Intelligence
2
20
3
16
the
gamedesigninitiative
at cornell university
Minimax
MAX
5
MIN
5
11
19
3
2
10
5
14
Artificial Intelligence
2
20
3
16
the
gamedesigninitiative
at cornell university
Advantages/Disadvantages
Scalable
Can be rational without “seeing” the endgame
Strength depends a lot on the heuristic
Still only works for some games
20
Artificial Intelligence
the
gamedesigninitiative
at cornell university
21
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Simultaneous Actions
1
3
22
1
3
Artificial Intelligence
1
3
the
gamedesigninitiative
at cornell university
23
Artificial Intelligence
the
gamedesigninitiative
at cornell university
right
left
right
left
24
Artificial Intelligence
the
gamedesigninitiative
at cornell university
right
left
right
left
25
Artificial Intelligence
the
gamedesigninitiative
at cornell university
right
left
26
right
left
1
-1
-1
1
Artificial Intelligence
the
gamedesigninitiative
at cornell university
right
left
27
right
left
1
-1
-1
1
Artificial Intelligence
the
gamedesigninitiative
at cornell university
right
left
28
right
left
1
-1
-1
1
Artificial Intelligence
the
gamedesigninitiative
at cornell university
What Is an Optimal Policy?
Agent Policy
29
Artificial Intelligence
2
3
right
1
3
left
the
gamedesigninitiative
at cornell university
What Is an Optimal Policy?
0
30
1
1
3
2
3
-1
3
1
3
Artificial Intelligence
the
gamedesigninitiative
at cornell university
What Is an Optimal Policy?
Agent Policy
2
3
1
3
31
Opponent Response
0 1
-1
3
Expected Reward
Artificial Intelligence
the
gamedesigninitiative
at cornell university
What Is an Optimal Policy?
32
Artificial Intelligence
the
gamedesigninitiative
at cornell university
What Is an Optimal Policy?
1
2
0
1
2
33
Artificial Intelligence
the
gamedesigninitiative
at cornell university
1
2
1
2
1
2
1
2
Agent’s Expected Reward = 0
34
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Expected Reward
𝑇
𝑣 = 𝜋𝑜 𝑹𝜋𝑎
35
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Calculating Reward
max 𝑣 such that
𝑣,𝜋
𝜋𝑖 = 1
𝑖
𝜋≥𝟎
𝑣 ≤𝑹𝜋
36
Artificial Intelligence
the
gamedesigninitiative
at cornell university
37
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding
A
38
B
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Make a grid!
A
39
B
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: Depth-First
A
40
B
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: Breadth-First
28
24
24
14
20
10
24
14
A
20
24
41
10
14
10
10
14
28
24
20
24
38
34
30
34
X
X
X
X
B
X
54
58
34
44
54
38
48
54
X
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: Breadth-First
28
24
24
14
20
10
24
14
A
20
24
42
10
14
10
10
14
28
24
20
24
38
34
30
34
X
X
X
X
B
X
54
58
34
44
54
38
48
54
X
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Breadth-First is Slow!
A
B
Idea: use heuristics
43
Artificial Intelligence
the
gamedesigninitiative
at cornell university
A* Algorithm
 Score f = g + h
 g: distance on best path
 h: naïve distance to goal
B
A
Manhattan distance = 30+20 = 50
44
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: A* Algorithm
f: 74
f: 60
f: 54
g:14
f: 60
h:60 g:10
h:50 g:14
f: 40
h:40
A
45
g:10
f: 74
h:50
g:14
B
h:30
f: 60
g:10
f: 54
h:60 g:10
h:50 g:14
h:40
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: A* Algorithm
f: 74
f: 60
f: 54
g = 20
g = 24
f=60
f=74
g:14 h:60 g:10 h:50 g:14 h:40
f: 60
f: 40
X
A
g:10
f: 74
h:50
f: 60
g:10
f: 54
B
h:30
g = 20
g = 24
f=60
f=74
g:14 h:60 g:10 h:50 g:14 h:40
46
X
X
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: A* Algorithm
f: 74
f: 60
f: 54
g:14
f: 60
h:60 g:10
h:50 g:14
f: 40
h:40
X
A
g:10
f: 74
h:50
f: 60
g:10
f: 54
h:30
g = 24
f=74
g:14 h:60 g:10 h:50 g:14 h:40
f: 88
B
X
f: 74
X
g:28
47
h:60 g:24
h:50
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: A* Algorithm
f: 94
f: 80
88
f: 74
h:70 g:20
g:28
f: 60
h:60 g:24
f: 54
h:50
f: 94
g:24
f: 74
g:24
f: 80
h:70 g:14
f: 60
h:60 g:10
h:50 g:14
f: 40
h:40
A
48
g:20
f: 94
h:60 g:10
f: 74
h:50
g:24
B
h:30
f: 60
g:10
f: 54
h:70 g:14
f: 94
h:60 g:10
f: 80
88
h:50 g:14
f: 74
h:40
g:24
h:70 g:20
g:28
h:60 g:24
h:50
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: A* Algorithm
f: 94
f: 80
88
f: 74
h:70 g:20
g:28
f: 60
h:60 g:24
f: 54
h:50
f: 94
g:24
f: 74
g:24
f: 80
h:70 g:14
f: 60
h:60 g:10
h:50 g:14
f: 40
h:40
In case of tie, use
most recently added
A
g:20
f: 94
h:60 g:10
f: 74
h:50
g:24
h:70 g:14
f: 94
g:24
h:30
f: 60
g:10
f: 54
h:60 g:10
f: 80
88
h:50 g:14
f: 74
h:40
h:70 g:20
g:28 h:60 g:24
f:108
f: 94
g:38
49
B
h:70 g:34
f: 74
f: 68
h:20 g:58
f: 74
h:10
f: 74
g:54
f: 74
h:50 g:34
f: 88
h:40 g:44
f: 88
h:30 g:54
f: 88
h:20
h:60 g:38
h:50 g:48
h:40 g:58
h:30
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Pathfinding: A* Algorithm
f: 94
f: 80
88
f: 74
h:70 g:20
g:28
f: 60
h:60 g:24
f: 54
h:50
f: 94
g:24
f: 74
g:24
f: 80
h:70 g:14
f: 60
h:60 g:10
h:50 g:14
f: 40
h:40
A
g:20
f: 94
h:60 g:10
f: 74
h:50
g:24
h:70 g:14
f: 94
g:24
h:30
f: 60
g:10
f: 54
h:60 g:10
f: 80
88
h:50 g:14
f: 74
h:40
h:70 g:20
g:28 h:60 g:24
f:108
f: 94
g:38
50
B
h:70 g:34
f: 74
f: 68
h:20 g:58
f: 74
h:10
f: 74
g:54
f: 74
h:50 g:34
f: 88
h:40 g:44
f: 88
h:30 g:54
f: 88
h:20
h:60 g:38
h:50 g:48
h:40 g:58
h:30
Artificial Intelligence
the
gamedesigninitiative
at cornell university
A* Mario
51
Artificial Intelligence
the
gamedesigninitiative
at cornell university
Group Activity: AI++
 IF your game has no AI




That’s How We Roll
Box
Thermo
THEN write a script that solves a level of your game
 IF your game has minimal AI




Sleep Walker
Pyrokid
Epic’s Epic Epic
THEN improve the AI script
 IF your game has heavy AI
 Zombify
 Nameless
 THEN discuss how you would add planning, heuristics, or intelligent randomness
52
Artificial Intelligence
the
gamedesigninitiative
at cornell university