DOC PREVIEW
USC CSCI 534 - Canamero-Homones

This preview shows page 1-2-3 out of 8 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 8 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 8 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 8 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 8 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Hormonal Modulation of Perception in Motivation-BasedAction Selection ArchitecturesOrlando Avila-Garc´ıaAdaptive Systems Research GroupDepartment of Computer ScienceUniversity of HertfordshireCollege Lane, Hatfield, Herts AL10 9AB, [email protected] Ca˜namero††Adaptive Systems Research GroupDepartment of Computer ScienceUniversity of HertfordshireCollege Lane, Hatfield, Herts AL10 9AB, [email protected] animat approach to artificial intelligence proposes biologically-inspired control mechanisms forautonomous robots. One of the related subproblems is action selection or “what to do next”. Manyaction selection architectures have been proposed. Motivation-based architectures implement a com-bination between internal and external stimuli to choose the appropriate behavior. Recent studies havepointed out that a second order mechanism to control motivation-based architectures would improvedramatically their performance. Drawing on the notion of biological hormones we have modeled twoof the functionalities ascribed to them in order to improve the adaptivity of motivation-based architec-tures. We have tested our “hormone-like” mechanisms in dynamic and unpredictable robotic scenarios.We analyze the results in terms of interesting behavioral phenomena that emerge from the interactionof these artificial hormones with the rest of architectural elements.1 IntroductionWithin the “behavior-based” (Brooks, 1986; Steelsand Brooks, 1995) or animat approach (Wilson,1985; Meyer, 1995) to AI, the ultimate goal of anautonomous agent is survival in a given dynamic,unpredictable and possibly threatening environment.Following inspiration from models in biology, neuro-science and cybernetics, animat’s survival needs arecommonly represented as internal essential variables.In order to remain “alive” the animat must main-tain homeostasis, i.e., keep the level of those essen-tial variables within certain ranges of viable values(Ashby, 1952; Aubin, 2000). Since different coursesof action can be taken to maintain homeostasis, oneof the related subproblems is action selection (Maes,1995), i.e., making a decision as to what behavior toexecute in order to guarantee survival in a given envi-ronment and situation.Many action selection architectures have been pro-posed (see Tyrrell (1993) or Guillot and Meyer (1994)for an overview). Following the behavior-based ap-proach to robotics, architectures started to be es-sentially reactive. Later on it became apparentthat some internal stimuli—e.g., the level of the es-sential variables—-were necessary in order to keepthose internal variables within their ranges (Arkin,1992). Following inspiration from ethology (Timber-gen, 1951; McFarland, 1999) and motivational sys-tems (McFarland, 1974; Toates, 1986), different waysof combining internal and external factors started toappear, proposing integration of those factors at dif-ferent levels of the action selection process (Maes,1991; Tyrrell, 1993; Blumberg, 1994; Spier and Mc-Farland, 1997). In motivation-based architectures(Ca˜namero, 1997), motivations constitute tendenciesto maintain homeostasis as a consequence of internaland external factors.In previous studies, we compared differentmotivation-based action selection architectures(Avila-Garc´ıa and Ca˜namero, 2002), and we sug-gested that the cyclic fashion in which motivationsare satisfied greatly influences the performanceof the agent (Avila-Garc´ıa et al., 2003). In thispaper, we show that the same action selectionarchitecture, appropriate in certain static environ-ments, does not perform viable activity cycles inenvironments with added dynamic complexities. Wepropose “hormone-like” mechanisms to adapt actionselection architectures to changing and dynamicenvironmental circumstances. Such mechanismsmodulate the sensory input of motivation-basedarchitectures in order to adapt its decisions tothe changing environmental circumstances. Ourmodulatory mechanisms are based on the functionalproperties of hormones (Levitan and Kaczmarek,1997).Section 2 describes the architectural elements ofour action selection architectures. Section 3 showshow hormonal modulation of the perception of ex-ternal stimuli (exteroceptors) adapt the action selec-tion architecture to a competitive scenario. Sec-tion 4 shows hormonal modulation of the percep-tion of one internal essential variable (interoceptor)to adapt the architecture to a dynamic prey-predatorscenario. Section 5 draws some conclusions.2 Action Selection ArchitecturesOur motivation-based architectures consist of twolayers—motivational and behavioral—linked througha synthetic physiology, leading to a two-step compu-tation of intensity. This computation is parallel withineach layer, but motivational intensity must be com-puted prior to the calculation of behavioral intensity,since the latter depends on the former. The moti-vational layer is made of motivational states that setthe goals of the system—the tendency to satisfy bod-ily (physiological) or internal needs. The behaviorallayer implements different ways in which those bod-ily needs can be satisfied. This distinction betweenmotivations and behaviors is essential when imple-menting more than one behavior satisfying the samemotivation (Toates, 1986).The Physiology consists of a number of survival-related, homeostatically controlled essential vari-ables—abstractions representing the level of internalresources that the agent needs in order to survive.They must be kept within a range of values for therobot to stay “alive,” thus defining a physiologicalspace (Sibly and McFarland, 1974) or viability zone(Ashby, 1952; Meyer, 1995) within which survival(continued existence) is guaranteed, whereas trans-gression of these boundaries leads to “death.”Motivations are abstractions representing tenden-cies to behave in particular ways as a consequence ofinternal and external factors (Toates, 1986). Internalfactors are mainly (but not only) bodily or physiolog-ical deficits or needs (0 ≤ di≤ 1), also traditionallyknown as “drives,” that set urges to action to main-tain the state of the controlled physiological variableswithin the viability zone. External factors are envi-ronmental stimuli or incentive cues (0 ≤ ci≤ 1)that allow to execute (consummatory) behaviors andhence to satisfy bodily needs. In our implementation,each motivation performs homeostatic control of onephysiological


View Full Document

USC CSCI 534 - Canamero-Homones

Download Canamero-Homones
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Canamero-Homones and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Canamero-Homones 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?