Guanang Su

Papers

Total Citations

H-Index

About

Guanang Su is a roboticist whose research lies at the intersection of geometric deep learning and robotic manipulation, with a focus on achieving sample-efficient learning for real-world grasping. Su’s major contributions center on leveraging equivariant neural networks to model the inherent symmetries of physical tasks. In their seminal work, “Sample Efficient Grasp Learning Using Equivariant Models” (34 citations), Su demonstrated that the optimal grasp function is SE(2)-equivariant, enabling a convolutional neural network to learn robust planar grasps from significantly fewer examples than traditional methods. This insight was further extended in “On Robot Grasp Learning Using Equivariant Models” (9 citations), which tackles the challenge of adapting grasp policies to noisy, real-world hardware. Su also introduced SEIL: Simulation-augmented Equivariant Imitation Learning (6 citations), a framework that combines simulation data with equivariant augmentation to drastically reduce the need for expensive physical interactions. By formalizing the role of symmetry in manipulation, Su’s work paves the way for robots that learn faster and generalize better, making them more practical for deployment in unstructured environments.