An Object Manipulation System Architecture for Humanoid Robots Based on Primate Cognition

Abstract

Currently one of the most important challenges is to bring robots out of factory floors to work alongside humans. Because these environments are characterized by a very large variety of objects, a key factor is to provide them with better adaptive object manipulation skills. This means that robots are required to connect, in a meaningful way, a high level task to the robot body movements. Understanding the objects at a physical level can give a robot a connecting mechanism to the higher level system. A previous experiment showed that a robot can skillfully manipulate an object if it is provided with the right mathematical models and controllers [1]. We want to expand this experiment by creating a system that can generalize this type of object manipulation capabilities to many more objects and tasks. In this paper we propose an architecture that helps bridge this gap by using insights from primate cognition. This system enables robots to handle more objects, deal better with tools, and facilitate the process of reasoning about actions and their expected outcomes. We exercised our implementation with some simple testing object models, and were able to corroborate its proper behavior under the proposed circumstances.