We pursuit practical researches that can be applied in real world.
In this research, we propose a vision-based human and robot 3D pose estimation algorithm to prevent collision accident. We use two calibrated RGB cameras to estimate 3D position of each joints. In case of robot, we introduce a noble iterative method that estimates joint angles from 2D keypoints. For human 3D pose, we use two-camera-based stereo triangulation which converts 2D keypoint pairs into 3D points. With the estimated 3D joint positions of each object, we can measure Euclidean distances to predict whether they are in danger of collision.
Autonomous Control of an Unmanned Ship with a Single Waterjet
In this study, we develop a Joystick control algorithm for remote berthing and deberthing of an unmanned ship propelled by an waterjet.
We also develop a fully autonomous berthing and deberthing control algorithm of the ship that follows given way point sets and desired for speeds
In this study, an algorithm for autonomous driving of small agricultural tractors on various types of ridges was developed and the performance was verified through experiments. The algorithm was constructed using a deep learning network and depth images to generalize the performance. And the verification process of the algorithm has been completed in various ridge environments.
Development of Indoor Multi-Robot Operation and Control Technology
In this project, we develop core technology for indoor multi-robot operation and control. The technology will be verified using two indoor mobile robots as shown in the figure. The robot is equipped with sensors such as a camera, lidar, IMU, and RFID receiver. In this study, given the map data representing the indoor environment and the global route, a driving control algorithm will be developed that allows the robot to autonomously plan a route to avoid obstacles and follow it. Object recognition using a deep learning algorithm, an algorithm to estimate the robot's position by fusion of various sensor information, and an LQ-based path tracking control algorithm will be developed. In addition, we design and verify the docking control algorithm for the robors to perform the Pickup / Drop function of the box.
In this research, we propose a autonomous driving algorithm for a tractor with given paths. For this work, we designed the Position/Attitude Estimation Algorithm to find the current position and attitude of the tractor and the Path Tracking Control to control it to drive along the path. And as a result of the current field driving test, the maximum error of position estimation in the straight section was 3.06 cm, and the maximum error of position estimation in the corner section was confirmed to be 14.1 cm.
GPS based autonomous driving rice transplanter sometimes invades the area of planted seedlings when GPS signal is incorrect. To solve this problem, we suggest vision based autonomous vehicle which detects planted area and generates desired driving path
Agriculture requires a lot of manpower due to its nature. With the aging of the agricultural population, the need for intelligent agricultural machines that can replace rural manpower is emerging. Accordingly, research is being conducted to apply autonomous driving technology to agricultural machinery, and the current technology can only be carried out on relatively standardized farmland due to the limitation of obtaining GPS coordinates of farmland. In this study, we suggest an algorithm that recognizes the outskirts of farmland, autonomous driving, and obtaining GPS coordinates at regular distances.
In order to operate an unmanned autonomous ship, it is necessary to recognize other ship and use a collision avoidance algorithm. In this study, a semantic segmentation algorithm that receives RGB images and outputs a ship segmentation map is designed. Eight types of ships are trained as synthetic images and validation is performed with real images. To reduce the domain gap between real and synthetic data, domain randomization and normalization are used.