Field Robotics in Environment and Geoscience
Multiple UAVs for Tributary Mapping
Korean tributary information is provided through KRF (Korea Reach File) based on GIS (Geographic Information System). For an efficient management of watershed and water quality, a precise tributary mapping is needed. Generally, it is very difficult to build a precise map of tributary due to various reasons such as limitations of access, difficulties of sensing, and so on. To overcome these limitations, we propose a tributary mapping system using autonomous unmanned aerial vehicle (UAV) and deep learning. Firstly, images of tributary will be captured using an on-board camera in real-time and then the captured images will be pre-processed to extract features such as the edge of tributary. Secondly, geometry information such as area, width, and location of tributary will be identified through a deep learning algorithm by using the extracted features as inputs. Finally, the trajectory of the UAV will be planned based on the identified geometry information of tributary to follow the tributary autonomously. Currently, the proposed autonomous tributary mapping system is under implementation in a physics-based hardware-in-the-loop simulator, Gazebo, and evaluation of robustness of mapping. To test a feasibility of the proposed tributary mapping system, it will be validated via tributaries near Jeonnam area. For a practical implementation of the proposed system, we will try to solve the following problems: 1) a hidden tributary mapping using an enhanced deep learning based estimation/recognition, 2) an obstacle avoidance of UAV using simultaneous localization and mapping algorithm based on on-board sensor fusion, and 3) a simultaneous mapping of multiple branch of tributary using multiple UAV system.
Telerobotics for Soil Sampling
Conventional soil sampling performed by humans is not possible in dangerous areas or inaccessible areas, but this problem can be overcome by using haptic teleoperation using unmanned ground vehicles or unmanned aerial vehicles with robotic arms. Research is ongoing to make the sampling work more comfortable by giving the user a haptic feedback about the soil during remote sampling. We established a soil sampling environment in virtual space and will measure how much haptic feedback contributes users' sampling task via various experiments.