Pop11 Experiments in Building Emotional Agents

Abbott2 and Gridland

When combined with the SIM_AGENT toolkit, Pop11 provides an attractive environment for investigating cognitively adequate agent architectures. This web page descibes a series of experiments developed as part of research into "concern processing in Autonomous Agents". This is very much on-going research (with many more ideas described in the thesis still to be tested through the "design-based" approch).

Abbott2 is an extension of work by Dolores Cañamero (incorporating ideas from the Cognition and Affect Project at Buirmingham University). Abbotts are constructed within a subsumption style framework, where more complex behaviours are implemented by adding agents to the society without modifying existing society members. Abbotts are endowed with primitive motivational states – impulses to action based on bodily needs – and “emotions” – peripheral and cognitive responses triggered by the recognition of a significant event. “Emotions” perform an alarm/meta-management function in Abbott, releasing chemicals (hormones, neuromodulators, and neurotransmitters) to alter both the perception of external stimuli and the activation levels of Abbott’s internal motivational drives. The code for Abbott2 is given below.
 

Abbott3

The main difference between the design of Abbott3 and its predecessors (Abbott and Abbott2), is the absence of a clearly demarcated emotion system (Cañamero’s [97] original design called for emotion agents to act as proto-specialists in a similar style to motivation agents). As we argue in chapters 4 and 5 of Allen [00], we believe that “emotions” are emergent mental states caused by the interaction of a variable number of intricately connected cognitive systems (i.e. systems that mediate arousal, attention, perception, concepts, memories, and physiological change) operating at different information-processing levels of the brain. Our approach towards elucidating emotions in Abbott3, is to replicate some of these systems at the information-level, and then explore the possible pathways through which emotional states can emerge. The code for Abbott3 is given below.

We have also performed a number of simple experiments using our Abbott3 design and the Gridland toolkit. These experiments are described in chapter 8 of Allen [00]. The raw results, and instructions for reproducing these results, can be found here.

Papers

Allen, S. (2000). A Concern-Centric Society-Of-Mind Approach To Mind Design. In Proceedings of the AISB'00 Symposium on How To Design A Functioning Mind. April 2000, Birmingham, England. Summary PDF (127K), Summary Postscript (34K), Poster StarOffice A0 (400K), Poster PDF A4 (320K), Poster Postscript A4 (230K)

Allen, S. (2001). Concern Processing in Autonomous Agents. Submitted PhD Thesis, School of Computer Science, University of Birmingham.

Cañamero, D. (1997). Modeling Motivations and Emotions as a Basis for Intelligent Behavior. In Proceedings of the First International Symposium on Autonomous Agents, AA'97, Marina del Rey, CA, February 5-8, The ACM Press.
(http://www.ai.mit.edu/people/lola/aa97-online.ps)
 

Images

 
Gridland Gridland Toolkit
Gridland world (large image 10K)
Gridland Toolkit (large image 20K)
Abbott2 architecture (large image 20K)
Abbott3 architecture (large image 28K)

 

Code

Requires Pop11, OpenMotif (or Motif), and the Birmingham Pop11 installation (rclib, simagent, poprulebase)
 
To Be Updated - gzipped tar file Code for Experiment 1 - see section 6.2.2 of thesis
(Motivated Control of Behaviour)
To Be Updated - gzipped tar file Code for Experiment 2 - see section 6.2.3 of thesis
(Affective Control of Motivation)
Abbott3 (14/01/2001) - gzipped tar file Code for Abbott3 - see sections 7 and 8 of thesis

Last updated by Steve Allen 14 January '01