Jiabao Wang

Daejeon, South Korea
Email: jiabao.wang@kaist.ac.kr
Contact: +82 01046369387

Education

Harbin Institute of Technology, Weihai Weihai, China
B.E. in Robotics Engineering 2019-2023
Korea Advanced Institute of Science and Technology Daejeon, South Korea
Ph.D. candidate in Electrical Engineering 2023-Present

Publications & Manuscripts

J. Wang and D. E. Chang, "LoFSORT: Sample Online and Real-time Tracking in Low Frame Rate Scenarios," 2025 IEEE International Conference on Robotics and Automation (ICRA), 2025.

J. Wang and D. E. Chang, "Velocity-Obstacle Based MPC for Shape-Aware Mobile Robot Motion Planning," Under Review

J. Wang and D. E. Chang, "A Unified Euclidean Framework for Visual-Inertial Odometry via Stable Embedding of Unit Quaternions," Under Review

J. Wang and D. E. Chang, "Observability analysis for the VINS system with stable embedding," Inprocessing

Projects

Multi-object tracking at low frame rate scenario

2024-Present
  • Developed a OCSORT-based multi-pedestrian tracking pipeline for low-frame-rate scenarios (2–5 Hz) using motion prediction and ReID features.
  • Implemented pedestrian velocity estimation using Kalman filtering and interacting multiple model estimation as alternative estimators, and developed a robot-centric risk assessment module.
  • Extended the framework from static to dynamic robot scenarios by incorporating ego-motion and relative-motion estimation.

Lightweight Joint Pose Estimation and Re-Identification Network

2026-Present
  • Built a lightweight multi-task network using a shared CSPNeXt backbone to perform human pose estimation and person re-identification simultaneously, reducing redundant feature extraction.
  • Integrated a pose-aware ReID head that combines multi-scale appearance features with pose representations to improve identity feature extraction under occlusion and viewpoint changes.

Euclidean VIO Framework with Stable Embedding

2025-2026
  • Developed a Euclidean VIO formulation using stable embedding to represent and optimize unit-quaternion states directly in ambient space, avoiding conventional manifold perturbation models.
  • Derived the corresponding IMU propagation, visual measurement, and state-estimation models, and implemented the resulting methods within the VINS-Fusion framework.
  • Collected real-world visual-inertial datasets using an Intel RealSense D435i in handheld and UAV-mounted configurations for system evaluation.

Experiences

Internship, OpenMPD Lab of Tsinghua University2023.05-2023.11
Remote
  • Deployed PointPillars on NVIDIA Jetson Orin and accelerated inference using TensorRT; reproduced and deployed BEVFusion on the Jetson Orin platform.
  • Conducted research on vehicle-road-vehicle cooperative decision-making and planning for autonomous driving, as well as real-time simulation and testing systems for intelligent vehicles.
Internship, Wuhu Atech Automotive Co., Ltd.2022.10-2023.05
Wuhu, China
  • Developed automated tools for A2L file generation and maintenance for ECU calibration and measurement workflows.
  • Parsed ELF files to extract symbol and address information, enabling automatic synchronization of variable addresses in A2L files.

Technical Skills

  • Programming: Python, C/C++
  • ML: PyTorch, scikit-learn
  • Simulation platform: ROS1/2, Gazebo, MuJoCo
  • Optimization tools: cvxpy, ACADO, Ceres