Hyunseok An

Papers

Total Citations

H-Index

About

Hyunseok An is a leading researcher at the intersection of geometric deep learning and robotic manipulation, with a primary focus on SE(3)-equivariant models for control. His work addresses a fundamental challenge in robotics: enabling systems to generalize manipulation skills across diverse spatial configurations without requiring exhaustive retraining. An’s most impactful contribution is **Diffusion-EDFs**, a novel framework that integrates bi-equivariant denoising generative modeling on the SE(3) group. This approach, detailed in his 2024 paper (18 citations), allows robots to learn complex manipulation tasks from stochastic human demonstrations while maintaining rotational and translational equivariance—a critical property for real-world dexterity. His 2025 tutorial survey (6 citations) further cements his role as a thought leader, providing a comprehensive roadmap for SE(3)-equivariant learning and control. By bridging diffusion models with geometric priors, An’s work enables more sample-efficient, robust, and generalizable robotic policies. His research is particularly influential for students and engineers working on visuomotor policy learning, offering a principled path toward robots that can intuitively grasp, rotate, and assemble objects in unstructured environments.