Personal Info
I am finding a New Job now!Up to now (2025.5), I have been employed by PICO (is a part of ByteDance Inc.) almost 3.5 years. With a decade-long career in robotics, autonomous systems and VR/AR, my core expertise centers on three principal domains:
-
(Lidar)-Visual-Inertial SLAM Systems
- UAV(Unmanned Aerial Vehicle): Camera + IMU + GPS.
- AV(Autonomous Vehicle): Lidar + Camera + IMU + GPS + Wheel Encoder.
- VR(Virtual Reality): Camera + IMU.
- Academic Achievement: 3 top-conference paper with first author.
-
UAV Localization + Perception + Control
- Localizer(Multi-Sensor Fusion)
- Main Sensors: IMU, GNSS, Camera, Compass, Barometer.
- Feature: Triple Redundancy Localization.
- Perception Obstacle Avoidance
- Dense depth mapping with stereo camera for obstacle avoidance.
- Obstacle Avoidance: Employ Dijkstra, A* or JPS to search a short path for bypassing the obstacle.
- Navigator and Controller
- Navigator: Employ Matlab + Simulink + Stateflow to design the Navigation strategy, and then generate the corresponding C++ code.
- Controller: Employ ADRC and modified PID algo to construct the controller of Flight Control.
- SIL + HIL Simulation
- Before testing with UAV devices, these above Localizer, Navigator and Controller are first simulated with Gazebo by SIL and HIL strategy.
- All-Weather Precision Take off/Landing Solution
- IR/QR marker solution.
-
Traditional Machine Learning
- Regression: Employ Regression method to conduct prediction.
- Recognition: Employ Fast-RCNN to conduct mark recognition.
- Data Enhancement: Employ warp, rotation, add noise, partial mask... to enhance train data of Fast-RCNN.
🔧 Technical Stack Tools：
- C++ (including NEON / SSE) / Python
- ROS / Docker / Gazebo
- Matlab (Simulink, MDB, HIL, Stateflow Code Generation)
- OpenCV / Ceres-Solver / Eigen / PCL / EKF
- Pytorch
Academic Volunteer (Reviewer):
- IEEE RA-L, NeurIPS'2025, ICRA'2025, ICPR'2024
Publication
-
SchurVINS: Schur Complement-Based Lightweight Visual Inertial Navigation System
First Author, 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR'2024) -
• Key Contributions:
- An equivalent residual model is proposed to deal with hyper high-dimension observations, which consists of gradient, Hessian and the corresponding observation covariance. This method is of great generality in EKF systems.
- A lightweight EKF-based landmark solver is proposed to estimate position of landmarks with high efficiency.
- A novel EKF-based VINS framework is developed to achieve ego-motion and landmark estimation simultaneously with high accuracy and efficiency.
• Arxiv： arxiv.org/pdf/2312.01616 -
Universal Online Temporal Calibration for Optimization-based Visual-Inertial Navigation Systems
First Author, 2025 IEEE International Conference on Robotics and Automation (ICRA'2025) -
• Key Contributions:
- A universal online temporal calibration strategy for optimization-based VINS algorithms is proposed.
- The proposed approach is integrated into the existing VINS framework to validate its feasibility.
- Comprehensive experiments are conducted to assess its impact on accuracy.
• Arxiv： arxiv.org/pdf/2312.01616 -
Stereo Visual Inertial Odometry with Online Baseline Calibration
First Author, 2020 IEEE International Conference on Robotics and Automation (ICRA'2020) -
• Key Contributions:
- A stereo VIO algorithm with online baseline estimation to overcome inaccurate stereo calibration.
- Evaluation on EuRoC to verify its effectiveness.
• Link： ieeexplore.ieee.org/document/9197581
• GitHub: github.com/bytedance/SchurVINS
• GitHub: github.com/bytedance/Ts_Online_Optimization