Junwoo Chang
Papers
4
Total Citations
29
H-Index
3
About
Junwoo Chang is at the forefront of integrating geometric deep learning with generative models for robotics, with a primary focus on SE(3)-equivariant architectures and diffusion-based manipulation. His landmark work, "Diffusion-EDFs," introduces a bi-equivariant denoising generative framework that operates directly on the SE(3) manifold, enabling robots to learn complex manipulation tasks from stochastic human demonstrations while respecting the physical symmetries of the environment. This approach, which has garnered 18 citations since its 2024 publication, represents a significant leap in sample efficiency and generalization for visuomotor policies. Chang also contributed "Denoising Heat-inspired Diffusion with Insulators," a novel method that embeds collision avoidance directly into the diffusion process, eliminating the need for inference-time obstacle detection. Complementing these technical contributions, his tutorial survey on SE(3)-equivariant robot learning (2025) serves as an essential resource for researchers entering this rapidly evolving field. Through his work, Chang is establishing new standards for how robots can leverage geometric priors and generative modeling to achieve more robust, data-efficient, and physically-grounded manipulation capabilities.
Research Focus
Key Achievements
Top Papers
- 1
- 2SE(3)-equivariant Robot Learning and Control: A Tutorial Survey6 citations · 2025
- 3
- 4