EKF_SLAM_ROS2: Real-Time Multi-Sensor SLAM System

• Built a real-time EKF-based SLAM framework in ROS2, fusing odometry, IMU, LiDAR, and point cloud data for high-frequency state estimation and robust 2D mapping.
• Implemented core EKF components in C++/Python, including nonlinear prediction–update models, noise modeling, observation processing, and multi-frame coordinate transformations.
• Designed ROS2 node communication, topic and TF management, parameter configuration, and launch structure, forming a modular and reusable SLAM architecture.
• Performed Gazebo simulation and rviz2 visualization debugging; improved cross-platform portability through parameterized modular design, supporting rapid algorithm extension and real-robot deployment.

Autonomous Navigation System based on 3D LiDAR & Vision Fusion (ROS2)

• Built an autonomous navigation system under the ROS2 + NAV2 framework, fusing LIO-SAM lidar odometry with YOLOv8 object detection results.
• Implemented state estimation of moving obstacles via EKF for robust localization and environmental modeling in dynamic environments.
• Constructed task-level behavioral decision logic based on Behavior Trees, enabling the robot to autonomously complete target recognition, candidate task point search, and behavior sequence switching.
• Implemented global path planning (RRT / RRT*) and local obstacle avoidance control using NAV2; combined with obstacle motion trend prediction to execute online trajectory correction.
• Validated sensor noise modeling, parameter calibration, and behavioral strategies in Gazebo, then migrated the system to a physical mobile robot platform (18–25 Hz).

Ship Route Planning for ENC Vector Charts (JMSE)

• Addressed boundary errors and conversion losses in traditional rasterized charts by using ENC vector chart data to construct channel configuration spaces.
• Organized vector data based on Delaunay triangulation and embedded ship kinematics constraints (fixed speed, turning radius, heading change) in path search.
• Proposed VK-RRT* algorithm, integrating kinematics constraints and goal-biased sampling. Experiments showed ~23% reduction in path length compared to standard RRT*.
• Verified algorithm generality in island channel and port vicinity scenarios.

Multi-constraint MPC for Vehicle Obstacle Avoidance Trajectory Tracking

• Designed a multi-constraint MPC controller constraining sideslip angle, yaw dynamics, and tire forces, introducing a curvature-adaptive cost function.
• MATLAB/Simulink Simulation: Reduced vehicle sideslip by ~30% and improved path smoothness by ~20% compared to single-constraint methods.
• Demonstrated stability under different curvature, obstacle layouts, and road conditions, reflecting controller robustness.

Binocular Ranging & EKF-based Target Tracking & Shooting System

• Implemented a target keypoint detection pipeline based on YOLOv8-pose, extracting shooting part pixel coordinates and mapping them to 3D scene positions.
• Designed and implemented an Extended Kalman Filter (EKF) fusion module, fusing IMU, drive motor encoders, and camera observations for recursive estimation of target pose/velocity and short-term trajectory prediction.
• Modeled target motion and ballistic dynamics, constructing a real-time trajectory generation and launch angle/timing calculator.
• Verified automatic hitting performance at 6–8 m distances under continuous motion conditions (Experimental hit rate ~92%).
• Demo Video

STM32 HAL–Based Mid-Layer Robot Control Library

• Developed a reusable mid-layer control library based on the STM32 HAL framework, integrating motor drivers, IMU data processing, chassis kinematics, gimbal attitude control, and host-computer communication protocols into a unified interface.
• Supported multiple mobile platforms including Mecanum, Swerve, and two-wheel self-balancing robots modeled as inverted pendulums; built state-space formulations and implemented LQR-based attitude stabilization, providing unified kinematic and dynamic computation functions across platforms.
• Decoupled execution control and perception data flow, designing a modular task loop and device abstraction layer to improve maintainability and system extensibility.
• The control library was reused across several competition and research robots, significantly reducing hardware porting effort and control-strategy iteration cost.