The competition aims to develop an autonomous drone to pass through complex and dangerous outdoor obstacles, infiltrate an unknown two-story building, detect hidden targets, create precise 3D maps including target types and coordinates, and return to the starting point safely. I developed main framework including local/global path planners, 3D map logger with the detected target positions and a decision-making system.

　KHNP Virtual Robot Challenge aims to develop a autonomous navigation algorithm and a robust controller to complete diverse and challenging tasks including obstacle avoidance, exploration, and enduring the wind disturbances. I made Gazebo maps for the competition and coded scoring GUI. Additionally, I coded sample controllers for the quadrotor UAV.

　Low Power Computer Vision Challenges aims to develop light and fast computer vision solutions to be used in many fields including Robotics. In 2022, the goals of the competition was to track the non-uniform motion vehicle at constant distance away with a quadrotor UAV, while avoiding obstacles. We estimated the trajectory of the moving vehicle in the form of 5th order polynomial using the detected center point with YOLO network. Then, Adaptive weight Model Predictive Controller (AMPC) is designed to track the target effectively.

　The final goal of this research project is to develop an exploration system for monitoring unknown palm tree orchard environments while avoid‑ ing collisions and figuring out appearances of trees. I developed the whole system including pointcloud data processing, path planning, and controller. A state‑of‑the‑art open‑sourced SLAM algorithm is utilized.

　KHNP Virtual Robot Challenge aims to develop a autonomous navigation algorithm and a robust controller to complete diverse and challenging tasks including obstacle avoidance, manipulation for grasping a cube, climbing stairs, and enduring the disturbances. I made Gazebo maps for the competition and coded scoring GUI. Additionally, I coded sample controllers for the manipulator and the quadruped robot.

　In this project, we developed adaptive multi robot localization method. With the high fidelity networking, artificial intelligent cooperative control, and mobile ground control station, unmanned swarm system has been researched to operate cyber-physical systems.

　In this project, we developed indoor SLAM, navigation, and exploration method to operate an UAV exploring the underground power line tunnel safely. In consideration with the limited payload and computational resource of the UAV, the precomputed and lightweight local exploration planner was proposed. Additionally, to charge the battery of the UAV on the UGV with the docking station, relative pose estimation EKF and autonomous landing algorithm was developed.

　Research on Unmanned Aerial Vehicles has been actively conducted in recent years. In particular, the UAV to explore an unknown, GNSS-denied environment is required, but the self-localization method, such as Visual Inertial Odometry, is mandatory to operate it. Considering the payload and the operating time of the UAV, lightweight and low-power consuming cameras and IMU are preferred, and even Object Detection and 3D Mapping can be obtained using a RGB-D camera. In this work, we developed a 3D Mapping system including object positions in an unknown and GNSS-denied environment for the UAV with a RGB-D camera. The system is demonstrated in Gazebo simulator, and the quantitative and qualitative results are obtained.

　Korea Robot Aircraft Competition aims to promote the revival of the domestic aviation industry and respond to various demands, this competition will expand the base of unmanned aviation-related technologies through participation in high school and university (graduate) students, and contribute to discovering and fostering human resources in related industries. The competition has been held every year since 2002 for the purpose of raising awareness of unmanned aerial vehicles (drones) and training manpower through various missions using unmanned aerial vehicles (drones) developed by college students.

　AlphaPilot is the first large-scale open innovation challenge of its kind focused on advancing artificial intelligence (AI) and autonomy. Supported and hosted by DRL, Lockheed Martin and NVIDIA. For the 1 Million dollar prize only for the winner. The challenge consists of Preliminary test and Real competition. Firstly, DRL and Lockheed martin used FlightGoggles Simulator from MIT to narrow down 424 teams over 81 countries into 9 Qualifiers. Only 9 Qulifiers participated real RACEs using DRL made drone 'RACER AI' which is equipped with RTOS like kernel customized NVIDIA Jetson Xavier and few sensors. Our team won the 3rd prize.

　ROS based autonomous driving system for mobile robot (Turtlebot3) is developed for finishing the racing track with diverse missions. Using MATLAB, Lyapunov functional is proved to stabilize the error model of mobile robot. Simple HOG based Cascade Object Detector is trained using Computer Vision Toolbox of MatLab and then all systems are coded with Python to control the robot in real-time. Only 35 dollar Raspberry Pi computer was adopted and it was equipped with 1-D LiDAR and mono camera. Our team won the Mathworks Korea Special Prize.

　Measured data from Vehicle’s multi sensor system have asynchronized sampling rate, The final goal of this research project is to design multi-rate State Estimator that can assume exact state using asynchronized data. Model based prediction controller is designed to perform at real-time for improving control performance. In addition, ROS based mobile robot data processing system, LiDAR data based path planning, sign recognition algorithm are researched for implementing autonomous system.