Autonomy in Virtual Agents: Integrating Perception and Action on Functionally Grounded Representations

Autonomy is a fundamental property for an intelligent virtual agent. The problem in the design of an autonomous IVA is that the respective models approach the interactive, environmental and representational aspects of the agent as separate to each other, while the situation in biological agents is quite different. A theoretical framework indicating the fundamental properties and characteristics of an autonomous biological agent is briefly presented and the interactivist model of representations combined with the concept of a semiotic process are used as a way to provide a detailed architecture of an autonomous agent and its fundamental characteristics. A part of the architecture is implemented as a case study and the results are critically discussed showing that such architecture may provide grounded representational structures, while issues of scaling are more difficult to be tackled.