Research Projects
-
SIT-LMPC: Safe Information-Theoretic Learning MPC for Iterative Tasks: This work introduces a GPU-parallelized learning MPC framework that combines MPPI, safe-set learning, and normalizing-flow value modeling to iteratively improve performance while enforcing safety constraints in stochastic nonlinear systems.
-
PoseINN: Realtime Visual Pose Regression via Invertible Neural Networks: A NeRF-guided, normalizing-flow–based invertible neural network maps images to pose distributions for efficient real-time visual localization with uncertainty estimation and deployment on embedded robots.
-
Local INN: Implicit Map Representation and Localization with Invertible Neural Networks: An invertible-network localization framework jointly encodes maps and performs fast LiDAR-based pose inference with uncertainty estimation, achieving particle-filter-level accuracy at lower latency.
Bio
Zirui Zang is a fith year PhD student. Previously, he obtained my MS in Robotics at University of Pennsylvania, and his BS in Optical Engineering at University of Rochester. His research interest is using genrative methods and distribution learning for Robotics.