Dynamic Ontology Driven Learning and Control of Robotic Tool Using Behavior

Abstract

One of the most interesting and rich fields of recent artificial intelligence (AI) research has come from examining embodied agents, the creation of which, poses interesting challenges and opportunities. Many traditional AI approaches which have previously proven successful quickly fail in the face of the unique challenges facing embodied agents. There is extensive multidisciplinary research into solving these problems, employing ideas and theory from not just computer science, but cognitive science, psychology, philosophy, neuroscience, as well as a range of other fields. Although the nature of embodied intelligence has risen to prominence in AI research relatively recently, animal behaviorists have been examining it for decades. One of the most explored areas of research into the nature of natural embodied intelligent agents (animals) involves their use of tools. We believe that the creation of artificial tool using behaviors yields insights into the nature of intelligence.

The proposed research will survey animal tool using behaviors and argue that some form of imitation may serve as an integral part of most animal tool using behavior. This claim, for the significance of imitation in tool use, will be supported with results from ethology, psychology and neuroscience. We will present a system based on multidisciplinary research that employs action ontologies to enable robotic imitation. We will demonstrate with this research that if mechanisms for imitative behaviors are implemented on a robotic platform, these imitative mechanisms may then be employed to enable tool using behaviors. While the achievement of tool using behaviors through this type of imitative mechanism is a novel and significant technical achievement in and of itself, it’s success also provides insight into how tool using behaviors may have first arisen in animals.