Date post: | 27-Jul-2015 |
Category: |
Presentations & Public Speaking |
Upload: | aist |
View: | 324 times |
Download: | 0 times |
Imita&on of Human Behavior
in 3D-‐Shooter Game
Makarov Ilya Tokmakov Mikhail Tokmakova Lada
Can't Kill Progress
Фон
• Brief Review of Exis<ng 3D-‐shooters • Geometry • Visual Recogni<on • Decision Making Model • Process of Gaming • Conclusion
AIST'2015, Yekaterinburg 2
Фон
Brief Review of Exis<ng 3D-‐shooters
AIST'2015, Yekaterinburg 3
Wolfenstein Call of duty: Ghosts
Idea
AIST'2015, Yekaterinburg 4
The main goal of our project is to replace commonly used automa<c target recogni<on and target aiming by using visual recogni<on methods and to test obtained model
We create an automa<c decision making module to search for enemies in a 3-‐D maze
Geometry
• Types of the objects: – walls, boxes – columns, doorways
• Processing the queue: – finding element with maximal priority
– recalcula<ng dangerous zones – comparing the first K queue elements
AIST'2015, Yekaterinburg 5
Example
AIST'2015, Yekaterinburg 6
The density func<ons represent the process of visual recogni<on of objects when we search for an enemy but have to spend some <me on processing object’s shapes.
Visual Recogni<on Screenshot with filled holes
AIST'2015, Yekaterinburg 7
Filtering of the screenshot
Yflgbcmyflj pfrhsnm Yflgbcmyflj pfrhsnm Yflgbcmyflj pfrhsnm
Yflgbcmyflj pfrhsnm
Decision Making Model 1) Walking
a) Bypassing and Search b) Moving toward some point in the maze c) Finding cover
2) Shoo<ng a) Sigh<ng b) Opening fire c) Correc<on rela<ve to recoil and motor reflexes
3) Visual recogni<on a) Recogni<on of objects b) Recogni<on of dangerous zones c) Recogni<on of enemy
(shoo<ng aeer enemy recogni<on)
AIST'2015, Yekaterinburg 8
Process of Gaming • The rela<ve error depends on: – the angle between BOT’s rota<on movements and direc<on on the target
– X-‐speed – Y-‐speed
AIST'2015, Yekaterinburg 9
Conclusion
In the current state we evaluate the parameters to iden<fy the dangerous zones and to sight on enemy. We proceed with the comparison of the methods of visual recogni<on to improve our model.
AIST'2015, Yekaterinburg 10
Thank you for aien<on!
Department of Data Analysis and Ar<ficial Intelligence Na<onal Research University Higher School of Economics
Makarov Ilya: [email protected]
Tokmakov Mikhail: [email protected] Tokmakova Lada: [email protected]
© hip://www.square-‐enix.com
AIST'2015, Yekaterinburg 11