Jiwoo Kim

I am an ECE Ph.D. student at Duke University, advised by Prof. Miroslav Pajic. Previously, I obtained my EEE B.S. at Yonsei University, Republic of Korea, where I was fortunate to be advised by Prof. Jongeun Choi. I am interested in Generative AI, Efficient AI, Representation Learning, and Reinforcement Learning.

My research focuses on developing efficient and multi-modal AI for task execution in dynamic environments. During my undergraduate studies, I explored the application of equivariance and representation theory to robotic manipulation. Currently, I am investigating generative models to develop efficient policies for real-world robotic policies.

Email CV Github Google Scholar

• [Apr. 2024] I will start my ECE Ph.D. at Duke, beginning Fall 2024.
• [Apr. 2024] Our paper "Diffusion-EDFs: Bi-equivariant Denoising Generative Modeling on SE(3) for Visual Robotic Manipulation" has been selected as a Highlight (11.9% of accepted papers) at CVPR 2024.
• [Jul. 2023] We won the Best Paper Award in Workshop on Symmetries in Robot Learning at RSS 2023.
• [Jun. 2023] Our paper "Robotic Manipulation Learning with Equivariant Descriptor Fields: Generative Modeling, Bi-equivariance, Steerability, and Locality" has been accepted to a RSS Workshop on Symmetries in Robot Learning 2023 (Oral).

We present Diffusion-EDFs, a novel approach that incorporates spatial roto-translation equivariance, i.e., SE(3)-equivariance to diffusion generative modeling.

We present a method that, during inference time, simultaneously generates only reachable goals and plans motions that avoid obstacles, all from a single visual input.

We introduce the recently proposed Equivariant Descriptor Fields (EDFs), focusing on the four key model properties: generative modeling, bi-equivariance, steerable representation, and locality.

template