affectTED

Exploring single-agent personality and emotions in the Gridworld architecture

Written by Santiago Loane and Nathaniel Potrepka

The Basics

The architecture of our emotional agent, TED, is twofold. First, we define a “personality” for the agent based on the Five Factor Model, which is a five-trait model of personality often given the acronym OCEAN for Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. Each personality trait is in the range [-1, 1]. Second, we define just one dimension of emotion for the agent, positive/negative valence, within a range of [-5, 5]. Given more time, we would implement high/low arousal and possibly open/closed stance.

Implementation

The personality traits for TED are first defined in the hash table *traits*, defined in gridworld-definitions.lisp. Later on in our world, our-world.lisp, we use the values of this hash table to set facts for TED, is_open_to_degree, is_conscientious_to_degree, is_extroverted_to_degree, etc., so that the personality traits can be later referenced by actions. A fact is also given to TED concerning his emotional valence, namely is_happy_to_degree.

Many actions were then rewritten (or newly written) to depend on the personality traits and to influence the emotional state. For example,

(setq ask+whether
	(make-op :name 'ask+whether :pars '(?x ?y ?z)
	:preconds '( (is_at AG ?z)
				 (is_at ?x ?z)
				 (can_talk ?x)
				 (knows ?x (whether ?y))
				 (not (knows AG (whether ?y))) )
	:effects '( (knows AG (whether ?y)) )
	:time-required 1
	:value 5
	)
)

becomes

(setq ask+whether
	(make-op :name 'ask+whether :pars '(?x ?y ?z ?h ?e)
	:preconds '( (is_alive TED)
               (is_extroverted_to_degree TED ?e)
      				 (is_at TED ?z)
      				 (is_at ?x ?z)
      				 (can_talk ?x)
      				 (knows ?x (whether ?y))
      				 (not (knows TED (whether ?y)))
               (is_happy_to_degree TED ?h) )
	:effects '( (knows TED (whether ?y))
              (is_happy_to_degree TED (min 5.0 (+ ?e ?h)))
              (not (is_happy_to_degree TED ?h)) )
	:time-required 1
	:value '(* 5.0 (+ 1.0 ?e))
	)
)

A full list of actions for TED is as follows:

answer_user_ynq ;for utility purposes, these two
answer_user_whq ;are not dependent on personality
die
walk
eat
sleep
drink
ask+whether
play
play+with
talk+with
heal

Additionally, we have added the possibility of death. Death is implemented as an exogenous event, which required a bit of extra effort to get working right, but essentially, for parameters (?u ?h ?f), the action die has preconditions

(is_alive TED)
(is_tired_to_degree TED ?f)
(is_thirsty_to_degree TED ?h)
(is_hungry_to_degree AH ?u)
(>= (+ ?f ?h ?u) 30.0)

and effects

(is_dead TED)
(not (is_alive TED))

In the agent's version of die (as opposed to the actual version), a very low value is given to die, specifically -999, in order to ensure that TED avoids death at all costs.

Compile & Run

To compile and run the code, call:

> (load "init")

and to run a step of the agent, simply call:

> (go!)

Additionally, some question answering is possible:

> (listen!)
((ask-yn (TED is_bored))
 (ask-wh (TED is_happy_to_degree ?h)))

It is also possible to run some basic tests using:

> (load "test")

References

Liu, D. (2011). Gridworld Framework Manual. Retrieved April 28, 2017, from https://www.cs.rochester.edu/~schubert/191-291/gridworld-code/Manual/GridworldManual.pdf