Patrick John Chia
Undergraduate Research and Innovation Scholar
Learning a 3D Representation of Objects in Scenes for Robotic Manipulation
- Artificial Intelligence and Machine Learning
Leslie P. Kaelbling
Humans rely extensively on sight to interact with the 3D world around us and we are able to reason about what we see and effectively manipulate objects in the 3D space. Current work on robotic prediction and planning for object manipulation operate on pixel level information which exist in the 2D space. We believe that this is not the ideal space when dealing with objects in the 3D space, especially for performing complex manipulation tasks. This serves as motivation to build a vision system that is able to construct a 3D representation of a 2D scene, which would enable robots to reason about and interact with the world in a similar way as humans.
“Through SuperUROP, I hope to gain experience and to be exposed to research in the field of machine learning. Having done several projects related to ML and machine vision, I want to find new ways that can help improve these vision-based systems. My project aims to improve a fundamental problem faced by many such vision systems and the notion of that is very exciting.”