Biologically Inspired Approach to Robot Intelligence: Spatial Language Learning in Virtual Environment

This paper proposes the framework for biologically inspired approach to the robot intelligence for learning the spatial language in virtual environment. Current research works model the spatial language using statistical models and 2D geometry information of the entities. In this paper, a new direction is taken to model the spatial language in deterministic way using entities properties and their interactions. The spatial words such as on, in, over, under, above, below, inside, outside, behind, in front of, back, front, left, right, beside, between, among, near and far are modeled and learned. The spatial concepts are learned through modeling the entities and their interactions. In contrast, current works learn the spatial concepts through modeling the words by statistical methods like hidden Markov model. It is seen in the experiments that only twenty five different types of entities are needed to learn the above spatial concept.