Sall goal Impact(event,goal)

Matthias Scheutz
Articial Intelligence and Robotics Laboratory
Department of Computer Science and Engineering
University of Notre Dame, Notre Dame, IN 46556, USA
[email protected]
Useful Roles of Emotions in Artificial Agents: A
Case Study from Artificial Life
In Proceedings of AAAI 2004 , AAAI Press
Lior Shamir
Applications
• Human/Robot Interaction
• User Interface Design
• Natural Language Processing
• Agent Modelling
•Video Games & Entertainment
input
External Environment
Emotions
action
Agent
Roles For Emotions In Artificial Agents
action selection (e.g., what to do next based on the current emotional state)
adaptation (e.g., short or long-term changes in behavior due to the emotional states)
social regulation (e.g., communicating or exchanging information with others via
emotional expressions)
sensory integration (e.g., emotional filtering of data or blocking of integration)
alarm mechanisms (e.g., fast re ex-like reactions in critical situations that interrupt
other processes)
motivation (e.g., creating motives as part of an emotional coping mechanism)
goal management (e.g., creation of new goals or reprioritization of existing ones)
learning (e.g., emotional evaluations as Q-values in reinforcement learning)
attentional focus (e.g., selection of data to be processed based on emotional
evaluation)
memory control (e.g., emotional bias on memory access and retrieval as well as
decay rate of memory items)
strategic processing (e.g., selection of dierent search strategies based on overall
emotional state)
self model (e.g., emotions as representations of what a situation is like for the agent)
AIBO
An Ethological and Emotional Basis for Human-Robot Interactio
Ronald C. Arkin*, Masahiro Fujita**, Tsuyoshi Takagi**, Rika Hasegawa**
3 Objects
WATER
FOOD
MASTER
6 Emotions
Hunger
Thirst
Elimination
Tiredness
Curiosity
Affection
ParleE: An Adaptive Plan Based Event Appraisal Model of
Emotions
Duy Bui, Dirk Heylen, Mannes Poel, and Anton Nijholt
Emotion impulse
at time t
Emotion at
time t
Emotion at
time t-1
Decay Function
Event
EIV
(Emotion Impulse Vector)
ESV
(Emotion State Vector)
Impact(event,goal) = Pafter(goal) - Pbefore(goal)
Hope
Hope = Sall goal Import(goal) × (P(goal) - 0.5)
Fear
Fear = Sall goal Import(goal) × (0.5 - P(goal))
Happiness
Sall goal Impact(event,goal) × Import(goal) × (1 - P(event))1/2
Sadness
Sall goal Impact(event,goal) × Import(goal) × (1 - P(event))2
Liking
Desirability(event) = Sall goal Import(goal) × Impact(event,goal)
LikingLevel t+1=max(1.0, min(1.0,LikingLevel t+0.1* Desirability(eve
Happy-For
Happy for = LikingLevel(another agent) × (another agent’s happine
Obie
FLAME - Fuzzy Logic Adaptive Model of Emotions
Magy Seif El-Nasr, John Yen, Thomas R. Ioerger
Desirability of events
Degree of importance
Degree of impact:
IF Impact(G1,E) is A1
AND Impact(G2,E) is A2
…..
AND Impact(Gk,E) is Ak
AND Importance(G1) is B1
AND Importance(G2) is B2
….
AND Importance(Gk) is Bk
THEN Desirability(E) is C
Example:
IF Impact(prevent starvation, food dish taken away) is HighlyNeg
AND Importance(prevent starvation) is ExtremelyImportant
THEN Desirability(food dish taken away) is HighlyUndesired
Hope
Hope = (1.7 * expectation 0.5 ) + (- 0.7 * desirability)
Useful Roles of Emotions in Artificial Agents: A Case Study from Artificial Life
Rules of the Game
1800 x 1800 grid.
Initially 50 resources. Each cycle adds another resource.
Each resource = 800 energy points.
Speed is between 1 and 4.
Movement cost = distancespeed2
Below 400 energy units – speed = 1
Agents compete for resources and can fight or flee.
Fighting cost 50 units. Fleeing is in the speed of 7 for 5 to
10 cycles.
Probability of fighting is public for each agent.
D = S Gr*resource(n) + S Ga*agent(m)
resource(n) – Vector from the position of A to the Nth resource
agent(m) – Vector from the position of A to the Mth agent
Social
Agents
Fight only if their tendency is
higher than the oponents
tendency
Vs.
Asocial
Agents
Based only on their own
tendency
Adaptive
Agents
Lower their action tendenacy
if they win.
Vs.
Non-Adaptive
Agents
Don’t change action tendency
based on the results
r – basic action tendency
m – current action tendency
Ar(m)+ =
Ar(m)- =
{
m + (1-m)/2
2m
r + (2m –r)*(1 – r)/2r
m>=r
m<=r/2
r/2<m<r
{
m - (1-m)
m/2
r/2 + r*(m –r)*(1 – r)
m>=r + (1-r)/2
m<=r
r<m<r + (1-r)/2
Emotional Agents
ga = action_tendency*100 - 50
Conflict Tendancy
>0
<0
Action Tendency
> 0.5
< 0.5
Emotional State
angry
fearful
Asocial Emotional
Asocial Non-Adaptive
Social Non-Adaptive
Asocial Adaptive
Social Adaptive