6. アンチャーテッド4 の人工知能
• Authored vs. Systemic: Finding a Balance for
Combat AI in 'Uncharted 4'
• Matthew Gallant
• http://www.gdcvault.com/play/1024467/Auth
ored-vs-Systemic-Finding-a
42. 3 第二次AIブーム(1980年代)
IF (A) then B
IF (C) then D
IF (E) then F
IF (G) then H
IF ( I ) then J
シンボルによる人工知能
(記号主義)
ニューラルネットによる人工知能
(コネクショニズム)
ルールベース
新しい学習法=
逆伝搬法
59. Deep Q-Learning
Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves,
Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller (DeepMind Technologies)
Playing Atari with Deep Reinforcement Learning
http://www.cs.toronto.edu/~vmnih/docs/dqn.pdf
画面を入力
操作はあらかじめ教える
スコアによる強化学習
61. 学習過程解析
Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves,
Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller (DeepMind Technologies)
Playing Atari with Deep Reinforcement Learning
http://www.cs.toronto.edu/~vmnih/docs/dqn.pdf
62. • Pπ ロールアウトポリシー(ロールアウトで討つ手を決める。
Pπ(a|s) sという状態でaを討つ確率)
• Pσ Supervised Learning Network プロの討つ手からその
手を討つ確率を決める。Pσ(a|s)sという状態でaを討つ確
率。
• Pρ 強化学習ネットワーク。Pρ(学習済み)に初期化。
• Vθ(s’) 局面の状態 S’ を見たときに、勝敗の確率を予測
する関数。つまり、勝つか、負けるかを返します。
Mastering the game of Go with deep neural networks and tree search
http://www.nature.com/nature/journal/v529/n7587/full/nature16961.html
https://deepmind.com/research/alphago/
63. Mastering the game of Go with deep neural networks and tree search
http://www.nature.com/nature/journal/v529/n7587/full/nature16961.html
https://deepmind.com/research/alphago/
170. メタAI Left 4 Dead の事例
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and
Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
今回は Left 4 Dead の事例を見てみる。
172. メタAI(=AI Director)によるユーザーのリラックス度に応じた敵出現度
ユーザーの緊張度
実際の敵出現数
計算によって
求められた
理想的な敵出現数
Build Up …プレイヤーの緊張度が目標値を超えるまで
敵を出現させ続ける。
Sustain Peak … 緊張度のピークを3-5秒維持するために、
敵の数を維持する。
Peak Fade … 敵の数を最小限へ減少していく。
Relax … プレイヤーたちが安全な領域へ行くまで、30-45秒間、
敵の出現を最小限に維持する。
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
より具体的なアルゴリズム
174. メタAIが作用を行う領域
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
メタAIが作用(敵の生成・
消滅)を行う領域を、
AAS(= Active Area Set) と
言う。
175. メタAIが作用を行う領域
(AAS=Active Area Set)
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
176. メタAIが作用を行う領域
(AAS=Active Area Set)
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
177. 安全な領域までの道のり(Flow Distance)
メタAIはプレイヤー群の経路を
トレースし予測する。
- どこへ来るか
- どこが背面になるか
- どこに向かうか
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
179. 敵出現領域
背後 前方
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
前方と背後のプレイヤー群から見えてない部屋に、
モンスターを発生させる。
181. メタAI Left 4 Dead の事例
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and
Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
今回は Left 4 Dead の事例を見てみる。
182. メタAI(=AI Director)によるユーザーのリラックス度に応じた敵出現度
ユーザーの緊張度
実際の敵出現数
計算によって
求められた
理想的な敵出現数
Build Up …プレイヤーの緊張度が目標値を超えるまで
敵を出現させ続ける。
Sustain Peak … 緊張度のピークを3-5秒維持するために、
敵の数を維持する。
Peak Fade … 敵の数を最小限へ減少していく。
Relax … プレイヤーたちが安全な領域へ行くまで、30-45秒間、
敵の出現を最小限に維持する。
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
より具体的なアルゴリズム
183. 安全な領域までの道のり(Flow Distance)
メタAIはプレイヤー群の経路を
トレースし予測する。
- どこへ来るか
- どこが背面になるか
- どこに向かうか
Michael Booth, "The AI Systems of Left 4 Dead," Artificial Intelligence and Interactive Digital Entertainment Conference at Stanford.
http://www.valvesoftware.com/publications.html
229. The Sims シリーズのAIの作り方
人をダイナミクス(力学系、動的な数値の仕組み )として動かす。
世界を動かす PeerAI(=キャラクターAI) を構築。
Sub
Peer
Meta
Meta
Peer
Sub
[原則] 周囲の対象に対する、あらゆる可能な行動から、
ムード(幸せ) 係数を最大化する行動を選択する。
Sims (not under direct player control) choose what to do by selecting, from all of the
possible behaviors in all of the objects, the behavior that maximizes their current happiness.
Will Wright, AI: A Design Perspective (AIIDE 2005)
http://www.aaai.org/Papers/AIIDE/2005/AIIDE05-041.ppt
Kenneth Forbus, Will Wright, “Some notes on programming objects in The Sims – Example”
http://www.qrg.cs.northwestern.edu/papers/Files/Programming_Objects_in_The_Sims.pdf
230. The Sims における「モチーフ・エンジン」
Ken Forbus, “Simulation and Modeling: Under the hood of The Sims” (NorthWerstern University)
http://www.cs.northwestern.edu/%7Eforbus/c95-gd/lectures/The_Sims_Under_the_Hood_files/frame.htm
Data
- Needs
- Personality
- Skills
- Relationships Sloppy - Neat
Shy - Outgoing
Serious - Playful
Lazy - Active
Mean - Nice
Physical
- Hunger
- Comfort
- Hygiene
- Bladder
Mental
- Energy
- Fun
- Social
- Room
Motive Engine
Cooking
Mechanical
Logic
Body
Etc.
AIの人格モデル
255. The Game
• Spec Ops: The Line – AAA Third-Person Shooter for Xbox 360, PS3, PC
• We are Daniel Kollmann and Jörg Reisig ( YAGER Development / Berlin )
256. Behavior Trees in SOTL
Summary:
• No separation between Evaluation and Execution
• Parallel processing of nodes
• Loops and decorators
• Synchronization points for groups
• Events drive the evaluation, reactive
257. Behavior Trees in SOTL
• No separation between Evaluation and Execution
Evaluation aborts currently executed action
No Evaluation while an action is being executed
Composite
Condition
Actions
259. Behavior Trees in SOTL
• Parallel processing of nodes
Stack can have multiple top nodes
All children of a parallel are updated one after another Run in parallel
261. Behavior Trees in SOTL
• Loops
Loops keep triggering their child until a certain
condition is met.
• Decorators
Decorators are active as long as their child
is running.
Keep pushing the child
Apply effect/buff
265. • Events and impulses
Any node in the stack can be the target of an event. When the event fires the
tree jumps back to that node.
An impulse checks if a certain event was fired not long ago. If an impulse is not
active, it is completely ignored and does not return any result.
Behavior Trees in SOTL
Jump here
Do this
267. Behavior Trees in SOTL
• Events and impulses
Events can also be blocked
in a branch.
Jump here
Do not react to any following need for cover
Do this
269. Behavior Trees in SOTL
Final conclusion ....
• The event system always aborts active nodes even when not doing
anything. Being able to specify some conditions would be nice.
• Parallels sort of allow you to create your own behaviour nodes, instead
have no parallels and build biggers nodes.
• Gates are really useful if used in the right places. For more complex stuff
rather do a dedicated node.
271. 強化学習
(例)格闘ゲームTaoFeng におけるキャラクター学習
Ralf Herbrich, Thore Graepel, Joaquin Quiñonero Candela Applied Games Group,Microsoft Research Cambridge
"Forza, Halo, Xbox Live The Magic of Research in Microsoft Products"
http://research.microsoft.com/en-us/projects/drivatar/ukstudentday.pptx
Microsoft Research Playing Machines: Machine Learning Applications in Computer Games
http://research.microsoft.com/en-us/projects/mlgames2008/
Video Games and Artificial Intelligence
http://research.microsoft.com/en-us/projects/ijcaiigames/
272. CORE Layer は、Physical Laryer 、Mission Layer のうちで、
どの認識を生成するかを決定するコマンドを投げる。
CERA-CRANIUM認識モデル
Arrabales, R. Ledezma, A. and Sanchis, A. "Towards the Generation of Visual Qualia
in Artificial Cognitive Architectures". (2010)
http://www.conscious-robots.com/raul/papers/Arrabales_BICS2010.pdf
285. 戦略位置検出システム
Matthew Jack, Mika Vehkala
Spaces in the Sandbox: Tactical Awareness in Open World Games (GDC2013)
http://www.gdcvault.com/play/1018038/Spaces-in-the-Sandbox-Tactical
286. 戦略位置検出システム
Matthew Jack, Mika Vehkala
Spaces in the Sandbox: Tactical Awareness in Open World Games (GDC2013)
http://www.gdcvault.com/play/1018038/Spaces-in-the-Sandbox-Tactical
287. 戦略位置検出システム
Matthew Jack, Mika Vehkala
Spaces in the Sandbox: Tactical Awareness in Open World Games (GDC2013)
http://www.gdcvault.com/play/1018038/Spaces-in-the-Sandbox-Tactical
288. 戦略位置検出システム
Matthew Jack, Mika Vehkala
Spaces in the Sandbox: Tactical Awareness in Open World Games (GDC2013)
http://www.gdcvault.com/play/1018038/Spaces-in-the-Sandbox-Tactical
289. 戦略位置検出システム
Matthew Jack, Mika Vehkala
Spaces in the Sandbox: Tactical Awareness in Open World Games (GDC2013)
http://www.gdcvault.com/play/1018038/Spaces-in-the-Sandbox-Tactical
290. 戦略位置検出システム
Matthew Jack, Mika Vehkala
Spaces in the Sandbox: Tactical Awareness in Open World Games (GDC2013)
http://www.gdcvault.com/play/1018038/Spaces-in-the-Sandbox-Tactical