Commonsense Reasoning about Moving Objects: an Elusive Goal

Leo Joskowicz

In this paper, we identify and characterize some of the sources of difficulty and propose an explanation for the limited success in formalizing commonsense spatial reasoning. First, we evaluate four general approaches to spatial reasoning about moving objects: simulation, analytical, axiomatic, and analogical reasoning. We characterize their expected input and output, identify their underlying assuml)tions, and evaluate their strengths and weaknesses. We observe a wide divergence of goals and techniques, analyze their possibility of fully capturing commonsense spatial reasoning, and question the feasibility of a general purpose reasoning engine. We conclude by suggesting that, more than any other commonsense reasoning task, spatial reasoning heavily relies on sensory-motor information and previous experience. This directly challenges the accepted wisdom in formalizing commonsense reasoning, which assumes that scnsory processes preceed mental processes and can thus be sequentially decoupled. We propose revising the assumptions made about the input and output of a reasoner.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.