Yong-Deuk Shin

Spatial uncertainty model for visual features using a Kinect™ sensor.

This study proposes a mathematical uncertainty model for the spatial measurement of visual features using Kinect™ sensors. This model can provide qualitative and quantitative analysis for the utilization of Kinect™ sensors as 3D perception sensors. In order to achieve this objective, we derived the propagation relationship of the uncertainties between the disparity image space and the real Cartesian space with the mapping function between the two spaces. Using this propagation relationship, we obtained the mathematical model for the covariance matrix of the measurement error, which represents the uncertainty for spatial position of visual features from Kinect™ sensors. In order to derive the quantitative model of spatial uncertainty for visual features, we estimated the covariance matrix in the disparity image space using collected visual feature data. Further, we computed the spatial uncertainty information by applying the covariance matrix in the disparity image space and the calibrated sensor parameters to the proposed mathematical model. This spatial uncertainty model was verified by comparing the uncertainty ellipsoids for spatial covariance matrices and the distribution of scattered matching visual features. We expect that this spatial uncertainty model and its analyses will be useful in various Kinect™ sensor applications.

Integration of recognition and planning for robot hand grasping

A robot should be able to recognize and estimate the pose of an object in order to grasp it. In addition, the robot should be able to infer the most reasonable strategy for grasping the object, which varies according to the type and pose of the object. In this paper, we design a grasping strategy engine for this purpose and suggest a method for recognizing and estimating the pose of an object with no two-dimensional intensity image. We also introduce our grasping data acquisition system (GDAS) for learning the best grasping strategy. The grasping strategy is composed of the approaching vector, opposition vector, and grasping type. In this paper, we use the iterative closest point (ICP) [1] algorithm for recognizing and estimating the pose of an object, along with an artificial neural network for selecting the best grasping strategy.

Real-time point-cloud data transmission for teleoperation using H.264/AVC

This paper presents an algorithm that offers real-time point-cloud transmission for teleoperation using H.264/AVC. Because the 3D-visualized point-cloud data provides metric information regarding the remote workplace, the teleoperator can observe the remote workplace from several viewpoints. The size of point-cloud data is a limitation when transmitting in poor and unstable network environments in real-time. Therefore, to perform robust teleoperation, it is necessary to compress point-cloud data at the remote location and transmit it in real-time. The feasibility of the proposed algorithm is verified by applying it in indoor and outdoor mapping with a 3G network system. Real-time teleoperation is effectively realized by applying the suggested algorithm.

Interactive remote robot operation framework for rescue robot

Situation awareness and its ambiguities are some of the main problems for rescue robotics. In a disaster area, there are many obstacles and confined passages. The operator must perceive the environment and decide the robot motion based on limited sensor data. These make it difficult for the operator to perceive the situation. The operator should be aware of the relative size of the obstacles in comparison with the rescue robot and the direction of the robot to navigate the disaster area and avoid collisions with obstacles. In this paper, we propose a software framework to reduce these ambiguities. A 3D map constructed with a 3D laser scanner and a robot model in this map can reduce the ambiguities. We also estimate future robot paths using the control inputs. Then, we perform collision detection in the expected paths and their neighbors. The collision results are displayed in the reconstructed 3D map using iterative closest points(ICP) algorithm. In this paper, we define interactive remote robot operation(IRRO) by extracting the common properties for reducing the ambiguities and abstracting them. The IRRO is composed of two parts, a simulation and evaluation step for the future robot motion and an interactive user interface system.

Real-time Polygon Generation and Texture Mapping for Tele-operation using 3D Point Cloud Data

In this paper, real-time polygon generation algorithm of 3D point cloud data and texture mapping for tele-operation is proposed. In a tele-operation, it is essential to provide more highly realistic visual information to a tele-operator. By using 3D point cloud data, the tele-operator can observe the working environment from various view point with a reconstructed 3D environment. However, there are huge empty space in 3D point cloud data, since there is no environmental information among the points. This empty space is not suitable for an environmental information. Therefore, real-time polygon generation algorithm of 3D point cloud data and texture mapping is presented to provide more highly realistic visual information to the tele-operator. The 3D environment reconstructed from the 3D point cloud data with texture mapped polygons is the crucial part of the tele-operation.

6DOF pose estimation using 2D-3D sensor fusion

Object pose estimation is a fundamental problem for a robot when manipulating an object. In this paper, we propose a method for estimating the pose of an object using a 2D image and a 3D point cloud. The Speeded Up Robust Feature (SURF) descriptors between the model image and input image were used to match the keypoints. The pose of an object was estimated using the 3D points corresponding to these matches. To produce more accurate results, the outliers were removed from these matches using Random Sample Consensus (RANSAC) and the result was refined using the Iterative Closest Point (ICP) algorithm. The experimental result demonstrated the high efficiency of our method.

Development of multi-purpose universal gripper

As the industrial site is changing from the mass production of a few selected items to the small quantity batch production, various works are required in the factorys production line. The single degree of freedom(DOF) gripper that can grasp only the specific object has accounted for a large part in the industrial site until now. Recently, the underactuated-method gripper that can grasp adaptively depending on the shape of object is being released, but its grip target is limitative as well. Therefore, the gripper of fully-actuated method is required to grasp and operate various objects. Therefore, the multi-purpose universal gripper that can be used diversely in various production lines is being required. The existing fully-actuated method could not be easily applied to the industrial site because it was likely to break down and it was hard to maintain it due to the complex wiring. In order to solve this problem, the multi-purpose universal gripper(MPUG) of the fully-actuated method was manufactured which was maintained easily because the modular actuator replaceable by joint link was used and which could grasp and operate various objects. It was confirmed through the test that the manufactured MPUG could grasp various objects stably.

Framework of grasping planning for multi-fingered robot hands

In this paper, we present a framework of grasping planning for multi-fingered robot hands which is based on the planning scheme of human. Structure of the proposed grasping planner is composed of three sub planners: grasping type planner, opposition parameter planner and approach vector planner. This planner is based on the way of humans grasping plan, so it is suitable for learning of intelligences for grasping of human. Using this framework of grasping planning, we would like to utilize to study for robots imitating of humans grasping plan intelligence.