Grad research talk: A Computational Model for Affordance Perception

Abstract

An "affordance" represents the possibility of an action on an object. For example, a chair affords us the possibility of sitting. Being able to infer affordances is an important cognitive ability and central to commonsense reasoning, tool use and creative problem solving in artificial agents. However, modeling affordance perception is challenging because it can be uncertain and can depend on social, cultural, moral and mental context. For example, a chair-like object on display in a museum might not afford sitting. What the suitable representations, algorithms and architectures are for perceiving this broader class of affordances remains somewhat of a mystery. In this talk I will discuss our ongoing work in developing a novel computational model comprising a flexible probabilistic rules-based logical representation coupled with a computational framework to reason about affordances in a more general manner. I will share some results and videos showing that an agent, using our approach, can reason through situations that involve a tight interplay between various social and functional factors.