Jiajun Fei

State Key Laboratory of Tribology, Tsinghua University

Papers

Total Citations

H-Index

About

Jiajun Fei is a researcher advancing the frontiers of robotics and 3D computer vision, with a focus on making intelligent systems both more efficient and more practical. His work centers on two key areas: hierarchical reinforcement learning (HRL) for complex robotic control and efficient fine-tuning for 3D point cloud analysis. In his highly cited 2020 paper, "Data-Efficient Hierarchical Reinforcement Learning for Robotic Assembly Control Applications" (84 citations), Fei tackled a critical bottleneck in robotics—the high sample complexity of learning assembly tasks. By developing an HRL framework that decomposes complex tasks into manageable subpolicies, his method significantly reduced the number of real-world interactions needed, paving the way for more practical deployment of learning-based control in manufacturing. More recently, Fei has addressed the growing challenge of model efficiency in 3D perception. His 2024 work, "Fine-Tuning Point Cloud Transformers with Dynamic Aggregation," introduces a novel approach to reduce the storage and computational burden of full fine-tuning, a crucial step for deploying advanced 3D models in resource-constrained environments like autonomous vehicles. Through these contributions, Fei is helping to bridge the gap between state-of-the-art algorithms and real-world application, making both robotic control and 3D analysis more accessible and efficient.