Approaches to learning for hybrid dynamical cognitive agents

Abstract

As a foundation for goal-directed behavior, a hybrid agent’s reactive and deliberative systems can share a single, unifying representation of intention. In this paper, we summarize a framework for hybrid dynamical cognitive agents (HDCAs) that incorporates a single representation of dynamical intention into both reactive and deliberative structures of a hybrid dynamical system agent model, and we introduce the first proposed approaches to learning for such intention-guided agents. The HDCA framework is based on ideas from spreading activation models and belief-desire-intention (BDI) models: Intentions and other cognitive elements are represented as interconnected, continuously varying quantities, employed for goaldirected intelligence by both reactive and deliberative processes. Methods for learning that modify interconnections among an HDCA’s cognitive elements —such as Hebbian associations based on co-active elements, and belief-intention learning of task-specific relationships— can therefore improve goal-directed performance without additional reliance on deliberation. We also present simple demonstrations of agents that learned geographic and domain-specific task relationships in a virtual grid world, and we discuss limitations and potential extensions of our approaches to HDCA learning.