Computer Model of Early Human Cognitive Development - Adam Sadilek
Hometown: Prague, Czech Republic
Class Year: 2008
Major: Computer Science
Research Advisor: Prof. Brendan Burns
Project Description:
We have addressed the problem of developing a general autonomous learning mechanism to allow an agent to achieve high accuracy in its prediction of the unknown features of objects. Being able to asses “what can I do with this object” and “what would be the effect of this action” is essential for the emergence of intelligent behavior and efficient planning. Ultimately, only interaction with objects can truly show how they behave when acted upon. Additionally, objects have a substantial set of attributes - such as mass, attachment, friction properties, or softness - whose values are virtually impossible to estimate based solely upon visual inspection. Without a good mental model, revealing these “hidden” attributes requires motor activity. However, physical manipulation is slow and dangerous in comparison to “just looking and thinking.” Therefore, noticing patterns in past experience and selecting objects for interactions intelligently can dramatically reduce the amount of actual experiments needed to achieve a given level of accuracy.