Transcript Wanching_Ho_Hangzhou_event

A New Kind of
Episodic Information and Memory
for NPC Design
in Computer Games
Wan Ching Ho
[email protected]
Adaptive Systems Research Group
University of Hertfordshire
Overview
•
Introduction
–
–
•
•
Episodic information (memory) for NPCs
Own research work
–
•
Typical techniques used in designing &
developing NPCs
Nouvelle AI concepts applying to the design
and development process of computer games
Narrative autobiographic agents
Looking into the games (case studies)
–
–
Black & White
The Sims
Introduction (1)
• Typical techniques used in
designing & developing NPCs:
– Finite State Machine (FSM)
– Scripting
Introduction (2)
• How FSM works in controlling the
behaviours (states) of NPCs:
– A limited number of behaviours, or states, are
pre-defined by the game designer. The actual
state of the NPC is made to switch with
seeming intelligence, depending on criteria
also pre-defined by the designer.
– To add unpredictability to NPCs, many
designers incorporate fuzzy logic or random
weighting into the decision-making process.
Health > 10% &
Player not in view
Victory
Dance
Search
Win
Wait &
Charge
Health > 10% &
Player in view
Attack
Health < 10% &
Player in view
Health < 10% &
Player not in view
Escape
Health = 0%
Death
Introduction (3)
• Scripting – a technique in which the
control of game events and NPCs is not
hard coded into the game, but defined
using a high-level language.
– A fixed sequence of actions or dialogs
will be triggered when the condition
has been matched, e.g. NPC answering
a question selected by the player
– Good to craft a illusion of AI
Introduction (4)
• “Most game programmers…don’t know
how to take techniques from academic AI
research and make them work in a
practical way.” (Rabin, AI Game
Programming Wisdom)
• “Currently the mood of the game industry
is pragmatic, rather than rush to new
technologies… developers were mostly
focused on digesting what they had,”
(Woodcock, 2002)
• Imagine the problem of moving a
patrolling guard through a number of
rooms in a castle. The designer can
sidestep the problem of teaching the
guard how to work out the best path
between rooms by simply drawing an
invisible track for the guard to follow
through the castle.
• But, if something (or the player) is
standing on that invisible track?...
Introduction (5)
• Nouvelle AI concepts can be applied to
the design and development process of
computer games:
– Embodied agents that are actually situated in
realistic worlds.
– Behaviour-based control architecture.
– Learning techniques used for the simulation
of adaptive behaviors produce creatures with
intelligent capabilities.
– Neural networks and genetic algorithms might
seem to be useful…
Introduction (6)
• Agent with embodiment and situatedness
– Embodiment: accurately simulating the body of
creatures, notably their interaction with the
environment.
– Situatedness: Embodied systems are mostly
affected by their immediate surroundings. Using
senses, only local information is gathered, similar
to the way humans or animals interact with their
environment.
Introduction (7)
• Behaviour-based control architectures
– Suitable for intelligent control because of
their reliability and simplicity, often
providing a foundation for more
elaborate techniques.
– E.g. subsumption control architecture
(Brooks, 1985)
Introduction (8)
• Agents with AI techniques of learning:
– Teaching involves humans providing a set of
examples that help the agent to behave until it's
managed to understand what to do.
– Imitation allows the agent to copy another
player, who is usually human. It can thereby
learn its behavior from a third-party experience.
– Shaping sets up successive trials from which
the agent can learn. After the agent learns to
accomplish simple tasks, more complex ones
are presented.
– Trial and error places the agent in its
environment and expects it to learn by trying
out all the different approaches on its own.
Introduction (9)
• Neural networks – attempts to model how
neurons behave in the brain, agents can
learn from experience. It allows agents to
respond to players and accommodate
changes in their behaviour without the
need for the designer to explicitly program
every possibility into the agent.
• Genetic algorithm –uses digital evolution
and selection to develop a solution to a
problem by trial and error
Introduction (10)
• But when “something misbehaves with one
of these technologies, it’s not easy to fix.
You can’t exclude the one thing that’s
broken without destroying all of the other
beautiful things in there. It’s all or nothing,
which is a very difficult situation when
deadlines approach,”
• “The hardest thing in game AI is just
making sure that the game never looks
dumb. You’d be better off having an AI that
was just above average all the time, rather
than one that was brilliant 98 percent of the
time and stupid 2 percent of the time.”
(Rabin, 2002)
Episodic Information
• How to increase the intelligence and
believability of the characters?
– NPC’s behaviour looks unnatural since there
is no connection or simply players cannot see
the reason when the NPC switches from one
behaviour to another one. This is described as
"schizophrenic" (Sengers, 2001)
– Episodic information can be the historical
grounding of the agent
• AUTOBIOGRAPHIC MEMORY
Autobiographic Agents (1)
Psychology[1]
Autobiographic memory is a specific kind of episodic memory,
may develop in human childhood.
Minimal Artificial Life Agents[2]
An agent possesses an autobiographic memory if it can create and
access information about sequences of actions which it experienced
during its lifetime. Autobiographic memory relates to meaningful
events.
Autobiographic Agents[2]
Autobiographic agents are agents which are embodied and situated in a
particular environment (including other agents), and which dynamically
reconstruct their individual history (autobiography) during their lifetimes.
Katherine Nelson (1993) “The Psychological and Social Origins of Autobiographic memory”
Kerstin Dautenhahn (1996) “Embodiment in Animals and Artifacts”
Autobiographic Agents (2)
The first previous work[1] showed how a single agent's
survival can benefit from autobiographic memory.
[2]
1. Wan Ching Ho, Kerstin Dautenhahn, Chrystopher L. Nehaniv (2003) “Comparing different control architectures
for autobiographic agents in static virtual environments” in Intelligent Virtual Agents Workshop (IVA2003).
2. VRML Model from www.atom.co.jp
Autobiographic Agents (3)
•
The second previous work[3] showed that autobiographic agents effectively
extend their lifespan by embedding an Event-based memory as compared
to a Purely Reactive subsumption control architecture, both in single-agent
and multi-agent experiments. Multi-agent environmental interference
dynamics result in a decreasing average lifespan of agents.
3. Wan Ching Ho, Kerstin Dautenhahn, Chrystopher L. Nehaniv, Rene te Boekhorst (2004) “Sharing
Memories: An Experimental Investigation with Multiple Autonomous Autobiographic Agents” in The 8th
Conference on Intelligent Autonomous Systems (IAS-8)
Autobiographic Agents for NPC
Design in PC Games (1)
• Nowadays only few behaviour simulation
systems have made explicit use of episodic
memory as a learning mechanism; where
learning means individual adaptation processes
that occur throughout a character’s brain (Isla &
Blumberg, 2002).
• The advantage of using episodic memory for
learning, compared to other mechanisms such
as reinforcement, neural networks or genetic
algorithms, is speed, as the game character can
form usable hypotheses for making decisions or
selecting behaviours to execute in the future,
after just one observation of users or other
agents.
Autobiographic Agents for NPC
Design in PC Games (2)
• Story Telling and Memories
– Story telling provides ‘empathic resonance’ through
transferring own experiences or receiving experiences
between autobiographic agents, that is a degree of
empathic re-experiencing of (the internal state of)
others since agents who received experiences may also
reconstruct aspects of individual’s history of others
(Dautenhahn, 1997).
• Recognizing individuals
– Although the receiver (autobiographic agent) can
selectively choose the information to store into its own
autobiographic memory from the sender, however, for the
internal states of the receiver, a certain level of trust to a
specific sender (another autobiographic agent) can be
built up.
– For example, the receiver had received many experiences
from that sender, which brought benefits (such as
avoiding dangers) to the receiver.
– Consequently, story telling not only brings changes of the
agents’ own memory but also the attitude that towards
others, this is the impact of communicating with others.
– In addition to this, the usefulness of stories which the
agent received from others will be known only after the
agent re-experienced them later; it is the same idea with
the Evaluation element in the narrative structure (Linde,
1993), but applying in other agents’ stories.
• Understanding a story from another agent
– Not simply matching the explicit contents of the
story to the agents’ own stories in its memory
– Not using high-level cognitive process from the
perspective of human understanding.
– The agent has to employ a low-level abstraction
process for recognizing the meaning and the
contents of this story from its own accumulated
stories.
– Then the agent is not going to directly merge this
story with other old stories and store it into its
own memory, but to shape it to the agent’s own
story figure, i.e. using its own memory schemata
to remember the story.
Narrative Influenced by AI
in PC Games
• Top-down design: Embedded
narrative with fixed story line & AI
cannot make much influence.
• Bottom-up design (like The Sims and
Black & White): Emergent narrative
with open-ended story & AI can be
highly involved when there are much
more chances for the players
interacting with the characters.
Case Studies (1)
• The Sims
– Characters always behave unexpectedly
– The intelligence to interact with any object is
not built into the Sims themselves. Rather,
they are equipped with a few basic needs, for
food, say, or entertainment. Objects within the
game advertise their ability to satisfy some of
these needs to any Sims who wander nearby.
• The Sims series:
– Need strategies to play
– Open-ended
– Game play soon goes beyond any
developer’s capacity to pre-program
situations
– Have the greatest need for advanced AI
technologies
Case Studies (2)
• Black & White
– Giving the player an in-game
representative - a Creature, which acts
autonomously, but it can be trained by
a system of punishments and rewards.
– It watches the player’s actions and
attempts to divine the intent behind
them, a technique called empathic
learning & imitation.
• The next big challenge for game AI
may be getting a game’s cast of
characters better at learning and
social interaction. (Cass, 2002)
End …
Thanks for your attention.