Towards modelling spatial cognition for intelligent agents

When people participate in face-to-face interactions they position themselves intuitively, and throughout the course of interaction, they try to maintain a common focus of attention with co-present individuals by readjusting their position as and when necessary. Kendon's (1990) F-formation system shows people's spatial positions and orientations as key in accounting for their behavioral patterns in social interactions. By taking insights from such human spatial cognition abilities, we aim to design for spatial behaviors in intelligent agents. For example, in a multi-agent system, we would expect agents to approach one another from right directions, maintain appropriate interpersonal distances, and during interactions, maintain a common focus of attention with co-present agents. In this paper, we present AgentMax, a software agent with a qualitative view of its spatial environment, capable of positioning and orienting itself appropriately with respect to the objects and events in its simulated environment called the Grid World. With the growing need for autonomous agents to exhibit appropriate social behaviors, we argue that our approach, which accounts for some fundamental behavioral abilities of software agents at the interactional level, is a necessary foundation of intelligent agent design.