Zhengnan Sun

Papers

Total Citations

H-Index

About

Zhengnan Sun is a rising roboticist whose research lies at the intersection of dexterous manipulation, multimodal perception, and representation learning. His work addresses one of robotics’ most formidable challenges: enabling multi-fingered hands to grasp and manipulate objects with human-like dexterity. In his highly cited paper “DexRepNet” (21 citations), Sun introduced a novel deep reinforcement learning framework that leverages geometric and spatial hand-object representations to tackle the high-dimensional action space and complex contact dynamics inherent in dexterous grasping. This work provides a scalable path for learning robust grasping policies without exhaustive manual engineering. More recently, Sun has pushed the boundaries of multimodal learning with “Masked Visual-Tactile Pre-training for Robot Manipulation” (7 citations), where he proposes a pretraining paradigm that fuses visual and tactile signals. By learning joint representations from these complementary modalities, his method enables robots to generalize manipulation skills across diverse tasks and environments—a critical step toward truly adaptive automation. Sun’s contributions are particularly notable for bridging the gap between data-driven pretraining and physical interaction, promising to make robots more capable in unstructured, real-world settings. His work continues to inspire new directions in embodied AI and robotic manipulation.