Ropossum is a framework that lets you play the beloved Cut The Rope game as much as you want and the levels will keep coming. You can design your own levels, check your designed levels for playability at real time, ask it to complete your unfinished designs according to your own preferences, or even suggest endless playable design variations according to your initial level design.
Optimizing AI for immediate response in Smart CCTV
Ropossum: A Game That Generates Itself
1.
2.
3.
4. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability
–Authoring tool
•Results and analysis
•Future Perspectives
5. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability
–Authoring tool
•Results and analysis
•Future Perspectives
16. 241,920,000 logins per month
4000 logins per second after 1.0 launch
2,000,000,000 files download by the launcher
11,000 skin downloads (in game) per second
18. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability
–Authoring tool
•Results and analysis
•Future Perspectives
19. -Popular physics based game
-Can generate endless levels
-All levels should be playable
-Opens the imagination of all players to design, test, modifytheir own levels and helpthem achieve that.
?
20.
21. Cut the Rope
Sequels 2010, 2011, 2012 and 2013
WWDC 2011,won anApple Design Awardfor the iOSplatform
100 million times
1.6 million users/daily
24. “Automatic Generation and Analysis of Physics-Based Puzzle Games”
IEEE Conference on Computational Intelligence and Games, CIG 2013. Nominated for best paper award.
CIG 13
“Evolving Playable Content for Cut the Rope through a Simulation-Based Approach”
AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, AIIDE 2013
AIIDE 13
“Ropossum: An Authoring Tool for Designing, optimizing and Solving Cut the Rope Levels”
AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, AIIDE 2013
AIIDE 13
25. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability
–Authoring tool
•Results and analysis
•Future Perspectives
31. should be placed higher than Om Nom when the level contains no component to elevate the candy, such as bubbles or water.
Candy
should be placed under the closest rope when there is no water,
bubbles or rockets.
Om Nom
should be placed close to the end of a rope, inside a predefined circle surrounding the end of it.
Blower
Should be placed in a position where it does intersect with at least one of the ropes.
Bubble
rockets should aim at Om Nom
Rocket
Components shouldn’t intersect, 3 block size apart
Comps
25%
10%
10%
10%
20%
25%
38. “Automatic Generation and Analysis of Physics-Based Puzzle Games”
IEEE Conference on Computational Intelligence and Games, CIG 2013
39. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability: The Model
–Authoring tool
•Results and analysis
•Future Perspectives
47. Explore each action sub tree
CRUST Engine 2.0
Prolog Engine
Engine State
as Facts
Set of ordered potential actions
Actions tree branch
New
Engine State
Inference Engine
Playability Model Closed Loop
51. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability: The Performance
–Authoring tool
•Results and analysis
•Future Perspectives
58. Presentation Timeline
•The problem, the solution and what others are doing
•Ropossum; the game and the research
•Ropossum; the framework
–Content Generation
–Playability
•The playability model
•Performance enhancement
•Content generation with playability check
•Other playability models
–Authoring tool
•Results and analysis
•Future Perspectives
60. “Evolving Playable Content for Cut the Rope through a Simulation-Based Approach”
AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, AIIDE 2013
61. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability: Other Models?
–Authoring tool
•Results and analysis
•Future Perspectives
62. Output
Input
Previous State
StateHidden
Weights V
Weights U
Weights W
SRN
Output
Input
Previous State
State/ Hidden
Weights V
Weights U
Copy
(Delayed)
Weights W
Elman
Output
Input
Previous State
State/ Hidden
Weights V
Weights U
Weights W
Copy
(Delayed)
Jordan
Recurrent
Neural
Networks
63. Output
Input
Previous State
Hidden
Weights V
Weights U
State
Weights W
Standard Recurrent Network (SRN)
65. Output
Input
Previous State
State/ Hidden
Weights V
Weights U
Copy
(Delayed)
Weights W
Elman RNN
66. Output
Input
Previous State
State/ Hidden
Weights V
Weights U
Copy
(Delayed)
Weights W
Jordan RNN
67. Output
Input
Previous State
State/ Hidden
Weights V
Weights U
Copy
(Delayed)
Weights W
Jordan RNN
68. Experimental setup
<rope> <rope> <air_cushion> <bumper> <bubble> <rocket> <candy> <Om_Nom>
Grid
XY
Candy distance
Om Nom distance
Input Representation
69. 0
400
200
300
100
Rope
Cuts(1)
Rope
Cuts(2)
Rocket
Presses
Air- Cushion
Presses
Bubble
Bursts
Void
Actions
Presented actions in the 100 playable levels test set altogether
Problem, Un-balanced Data
70. Balancing
COI
Data Representation
Training
levels #
Network
Type
Setup
Hidden
Neurons #
TestId
Balanced
N
All
100
Elman
10 networks
50traininglevels
0.0001 error
10 runs
37
Test 1
Balanced
Y
All
100
Elman
N
Y
XY
10, 50, 100
Elman
6 networks
50traininglevels
0.0001 error
6 runs
37
Test 2
4
Y
XY
10, 50, 100
Elman
N
Y
XY
150
Jordan
8 networks
50traininglevels
0.01 error
8 runs
5, 10, 15, 20,25, 30, 35, 40
Test 3
4
Y
XY
50
Jordan
4
Y
XY
50
Elman
N
Y
XY
10, 50, 100
Jordan
8 networks
50traininglevels
0.02 error
8 runs
5, 10, 15, 20,25, 30, 35, 40
Test 4
2
Y
XY
10, 50, 100
Jordan
N
N
Cookie distance
10, 50, 100
Elman
8 networks
50traininglevels
0.02 error
5 runs
5
Test 5
N
N
XY
10, 50, 100
Elman
71.
72. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability: Other Models?
–Authoring tool
•Results and analysis
•Future Perspectives
78. “Ropossum: An Authoring Tool for Designing, optimizing and Solving Cut the Rope Levels”
AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, AIIDE 2013
79. Presentation Timeline
•The problem, the solution and what others are doing
•The game and the research
•Ropossum: the framework
–Content generation
–Playability: Other Models?
–Authoring tool
•Results and analysis
•Future Perspectives
80. Selected InterestingResults
•35% of the playable levels haven’t been branch cut [direct solution]
•72.3% is the percentage of void action in the generated levels [relatively short levels?]
•29.8 ∓58.3 sec for checking a playable level.
•210% ∓167.6 for checking a non-playable level
–115 and 638 nodes is the average number of explored nodes for generated playable and non-playable levels [more explored nodes]
•10 min is the average time for generating a playable level
81. Average and standard deviation values of the four components
extracted from 500 generated levels
Frequency Analysis
87. Usedvs. Presentedcomponents by the playability agent over 100 generated playable levels
0
2
1
1.5
0.5
Ropes
Rockets
Air- Cushions
Bubbles
Bumpers
# activatedcomponents
88. Presentation Timeline
•The problem, the solution and what others are doing
•Ropossum; the game and the research
•Ropossum; the framework
–Content generation
–Playability
–Authoring tool
•Results and analysis
•Future Perspectives
89. Marry someoneand break your addiction
Employedor NOT
Employed people are 2Xmore likely to feel
anxiety and 2.25Xlikely to experience anger
Singlevs. Married
You are X2likely to be addicted on angry birds if
you are single than if you are married
90. 6%
Very
anxious
17%
Somewhat anxious
22%
Neither
32%
Somewhat relaxed
23%
Very
relaxed
Player Emotion