Active scene understanding

Humans possess a remarkable ability to actively explore their environments in order to retrieve useful information. For example, we may flip over a book to see its title, push a large box to sense its weight, or dip our toes into a pool to gauge its temperature. In these examples, we use actions (flipping, pushing, dipping) to retrieve relevant information (title, weight, temperature) for planning future actions (reading the book, lifting the box, swimming).
Can our robots learn to do the same? In this project you will develop computer vision and machine learning algorithms that enable robots to actively interact with the environment and use the interactions to improve their understanding of the environments.