
N. Hani, S. Engin, J.J. Chao, V. Isler
NeurIPS 2020
[pdf] [webpage] [code] [video]
CS Ph.D. student at UMN robotics, 3d vision, algorithms
How to generate novel views of objects using a small dataset as prior?
Our method learns a continous object representation in 3D that allows us rendering images at desired viewpoints
with the help of cyclic consistency losses.
How should a manipulator arm move to visually inspect and reconstruct an object?
We present a method that iteratively refines the reconstruction and uses it to plan for the next view for the camera.
Suppose a group of disconnected robots are deployed in a large open area. What strategy should a robot pursue to disseminate a piece of information to the rest of the robots (whose positions are unknown!) as quickly as possible?