A multi-strategy spatial navigation model in a text-based environment

Cognitive maps have been used to explain the cognitive processes underlying human spatial memory for several decades. Numerous theories have offered explanations regarding the structure of cognitive maps from the perspectives of spatial cognition, visual perception, and neuroscience. Within this literature, two representations of cognitive maps (map-like representations and route-based representations) have generally been used to model the learning of spatial information. Map-like representations involve spatial geometries, with landmarks acting as strong references. Alternatively, route-based representations consist of sequences of positions, orientations, and landmarks. Studies suggest that humans use both of these representations, and that the representation (and thus the strategy) used is often a function of the agent's familiarity with the environment. In this paper, we propose an ACT-R model that integrates both these representations and their associated strategies, using a route-based representation when learning new environments and a map-based representation to refine the agent's understanding. The results show that the map-based representation could improve an agent's route-following abilities in at least two cases: (1) searching the route when the agent is in an off-route position, and (2) for taking novel shortcuts in the route. Our model's behavioral output is consistent with the literature (Foo, Duchon, Warren, & Tarr, 2005; Foo, Warren, Duchon, & Tarr, 2007), in that our agents like human beings rely primarily on route-based memory when navigating while using map-based memory as a secondary reference.