Research

I’m trying to explore and address fundamental problems in robotic state estimation using tools from optimization, differential geometry, probability theory, and information theory. Coming from an engineering background, I also strive to translate these ideas into practical systems — such as high-performance C++ libraries and field robotics experiments.

The dream seems a little bit ambitious, and I’m still learning along the way:)

Current research

My research focuses on advancing optimization techniques for robotic state estimation and navigation, particularly in SLAM back-ends. I am currently working on certifiable (globally optimal) algorithms, robust certifiable methods, and distributed optimization. I am interested in both foundational algorithmic problems and the development of novel applications enabled by these techniques — for example, designing velocity-related certifiable factors to extend certifiable algorithms to inertial navigation systems.

Certifiable Estimation

1, Simplifying Certifiable Estimation with Factor Graphs: Theory and System (Preprint coming soon!)

To lower the barrier to entry for certifiable methods, we propose a certifiable factor-graph optimization framework built entirely on GTSAM. It:

1. Makes explicit the theoretical connection between certifiable estimation via Burer-Monteiro factorization and factor-graph optimization.
2. Makes certifiable methods accessible to systems already using factor-graph inference, enabling new applications even for users familiar only with local search.
3. Facilitates rapid development of new certifiable algorithms with minimal implementation effort.

Certifiable Factors (Workshop version, new code base will be released soon!)

Workshop Version Poster:

2, Robust certifiable estimation.

Design robust algorithms for certifiable estimation based on Burer–Monteiro factorization, aiming to solve large-scale problems such as pose graph optimization with both performance guarantees and resilience to outlier corruption.

3, Incremental certifiable estimation.

Distributed Optimization

1, Distributed second-order bundle adjustment solver.(In preparation of a preprint)

2, Distributed certifiable estimation.

Based on our certifiable factors and distributed second-order optimization algorithm, trying to achieve faster convergence than the currently leading RBCD-based methods.