Guofei Chen

Hi! I am Guofei Chen. I obtained my Master's in Robotics from Carnegie Mellon University, advised by Prof. Ji Zhang and Prof. Wenshan Wang. Before CMU, I received my Bachelor's in Automation from Zhejiang University, where I worked with Prof. Fei Gao. I am interested in navigation and manipulation systems that actively interact and evolve with the environment or human, which spans SLAM, planning, and machine learning.

Email  /  LinkedIn  /  Github

Research

My research focuses on autonomous navigation and manipulation systems for mobile robots, especially in complex unknown environments. Topics include SLAM, planning among movable obstacles, real-time visibility-graph methods, and full-stack platforms for data-driven navigation.

A 4D spatio-temporal mapping framework for language-guided navigation that integrates high-frequency geometric SLAM with asynchronous open-vocabulary perception. A consistency-driven mapping engine combines 3D-aware instance association/re-activation with a principled existence-and-label confidence update to maintain stable object identities and prune stale content under occlusions and scene changes, producing a queryable 4D scene-graph that interfaces naturally with VLMs.

An efficient navigation system that combines manipulating movable objects for shorter paths, even in complex unknown environments. Replans the global path in milliseconds.

Full-stack navigation systems on Unitree Go2 and Diablo, aiming to provide a low-cost and ready-to-use platform for AI navigation research.

A real-time 3D mapping and path planning algorithm for aerial robots that is up to 1000x faster than grid-map based strong benchmarks.

An inference framework using deep networks derived closely following optimal control principles for large-scale modeling of chemical processes.

Search-based planning algorithms on grid maps for solving the path planning among movable obstacles (PAMO) problem, with completeness and optimality guarantees.

Projects

A full-stack navigation platform using only the original onboard sensor (the L1 lidar by Unitree). Goes to a command position and avoids obstacles without hardware modification.

Open-sourced hardware and software stack for navigation on Diablo, a wheel-legged robot, using Livox Mid-360 LiDAR. Serves as a robust platform for language and other navigation tasks.

A self-contained, originally designed SQL database management system. Architect and code reviewer of the project: B+ tree indexes, page-based record manager, buffer manager, and a row-based storage engine.

Website template adapted from Jon Barron.