This talk will showcase some of the latest practices where technology is used in order to create more believable results with non-player characters and situations. Apart from game worlds, we will look into the usage of content created by bots which are displayed and executed in real-time experiences and interactive media. The goal of the presentation is to discuss the tools and approaches in storytelling for both tech and non-tech enthusiasts which can help them not only to automate but to widen their (creative) output.
2. Introduction
● Creative technologist
● Specialized in new media and real-time content design and development
● Early adopter of game engines as a non-game development creation platform
● Focused on immersive technologies and interactive audio-visual experiences
● PhD candidate
● Assistant Professor @ Metropolitan University Belgrade
3. Introduction
But, most importantly
● Creative technologist
● Specialized in new media and real-time content design and development
● Early adopter of game engines as a non-game development creation platform
● Focused on immersive technologies and interactive audio-visual experiences
● PhD candidate
● Assistant Professor @ Metropolitan University Belgrade
Passionate about intersection
of art and technology.
4. Cast (in order of appearance):
● Roommate with five pets.
Supica (Soup) Knedla (Dumpling) Kari (Curry) Kim & Zrno
(Caraway & Grain)
8. in its core, game AI
IS ABOUT PERCEPTION*
* in context of the game world
9. Practical use of Artificial intelligence in games is, largely, oriented towards
implementing several approaches for controlling Non-player characters (NPCs).
Most importantly, these techniques tell the engine exactly how a NPC should interact
with the game world in your scene (including player, being peer or opponent).
Some of these aspects are commonly taken care of through
pathfinding or state machines.
01 AI & Games
10. ☹️
Also, in world of real-time graphics, we are in a never-ending pursuit of optimization, so
many of commonly used techniques for AI are still in experimental phase.
11. ☺️
Also, in world of real-time graphics, we are in a never-ending pursuit of optimization, so
many of commonly used techniques for AI are still in experimental phase.
But we believe it will change.
12. Game engines are now actively
using machine learning to optimize and improve
RENDERING PIPELINES
And there are interesting experiments with style transfer post-processing effects.
13. Pathfinding
is the shortest route between two points.
01 AI & Games
NAVIGATION MESH
Tells agents where they can walk (climb, jump etc.),
and are generated at runtime.
22. But, in order to have believable animation changes, it is necessary to use motion
matching and animation blending.
Often from different animation segments that are masked on a single skeletal mesh (f.e,
when talking about bipeds)
01 AI & Games
23. But, in order to have believable animation changes, it is necessary to use motion
matching and animation blending.
Often from different animation segments that are masked on a single skeletal mesh (f.e,
when talking about bipeds)
01 AI & Games
● Smooth animation changes
● Believable motion
● Accurate reactions
24. But, in order to have believable animation changes, it is necessary to use motion
matching and animation blending.
Often from different animation segments that are masked on a single skeletal mesh (f.e,
when talking about bipeds)
01 AI & Games
● Smooth animation changes
● Believable motion
● Accurate reactions
● Production scalability
● Control over animation
● Animation file size
26. State machines will not be replaced soon, but
in terms of complexity might be augmented with use of
reinforcement learning
02 Digital experiences and machine learning
27. State machines will not be replaced soon, but
in terms of complexity might be augmented with use of
reinforcement learning
02 Digital experiences and machine learning
Since our intention as players is driven by winning, it could make sense to create
agents that are supposed to play against humans with reinforcement learning as
they are equally and fundamentally goal-directed.
28. 02 Digital experiences and machine learning
Get
current state
Initialize
Predict
best action
Perform
action
Get
the reward
Update
Q-table
29. So,
it takes some time :)
Rome wasn’t built in a day (because there was no pre-trained model)
30. In order to achieve realistic interaction with agents in digital environment, we need to
identify challenges, which may be, but are not limited to:
● Exploration
● Dialogue
● Object or environment handling
02 Digital experiences and machine learning
31. In order to achieve realistic interaction with agents in digital environment, we need to
identify challenges, which may be, but are not limited to:
● Exploration
● Dialogue
● Object or environment handling
02 Digital experiences and machine learning
For exploration, we could use pathfinding.
When it comes to dialogue - natural language processing.
Interacting with scene sounds like good task for reinforced learning.
32. As real-time interactive content creators,
it is our goal to define what the agent’s observation
space must include so it can define
strategy to earn a
reward.
33.
34. AI tools for
CONTENT GENERATION
In the end, although they use very advanced and cutting edge tech, our agents are part of the content.
03
35. One of the recent questions
with these rising tools is
“are they going to
replace the artists?”
And, of course, the answer is no.
03 AI tools for content generation
36. Content creators actually got an extra role. Maybe it can be called
creative prompt designer?
realistic action hero mix of cyborg and wolf cinematic style
prompt:
37. Paradigm shift is that now artist can become curator. Precisely, we have to further develop the ability to
articulate our ideas to the machine.