Visual servo control uses images that are obtained by a camera for robotic control. This study focuses on the problem of positioning a target object using a robotic manipulator with image-based visual servo (IBVS) control. To perform the positioning task, the image-based visual servoing requires visual features that can be extracted from the appearance of the target object. Therefore, a positioning error tends to increase especially for textureless objects, such as industrial parts. Since it is difficult to extract differences of the visual features between current and goal images. To solve these problems, this paper presents a novel visual servoing named “Active Visual Servoing.” Active Visual Servoing (AVS) projects patterned light onto the target object using a projector. The design of the projection pattern affects the positioning error. AVS uses an optimal pattern which is theoretically derived and maximizes differences between current and goal images. The experimental results show that the proposed active visual servoing method reduces the positioning error by more than 97% compared to conventional image-based visual servoing.

This study proposes a method of robust regrasping an object using a dual-arm robot with general-purpose hands, which is robust against the error of grasping. In this paper, one arm is assigned to hand over the object to the other arm that is named a receiver arm. The grasping error must be considered to increase the success rate of the regrasping since a hand-over arm first picks up the object with the general-purpose hand. In an online phase, the proposed method performs object positioning at an optimal pose at the time of regrasping using an image-based visual servoing (IBVS) approach to reduce the effect of the grasping error. In the planning phase, the proposed method computes the optimal pose for regrasping by maximizing the minimum singular values of the image Jacobian of IBVS to achieve a high positioning accuracy using a 3D model of the target object. To achieve the regrasping objects with various shapes robustly against image noises and changes in light environments, the image Jacobian of IBVS is computed by numerical differential using an actual data set. A large number of data sets corresponding to each candidate grasp are usually required for computing the image Jacobian. To reduce the number of data sets, we propose a conversion method of the image Jacobian requiring only one data set corresponding to one representative grasp. The experimental results show that the proposed method achieves regrasping of target objects with the general-purpose hands with high success rates and performs target object positioning with less than 0.7[mm] positioning error.

This paper proposes a deep learning-based fast grasp detection method with a small dataset for robotic bin-picking. We consider the problem of grasping stacked up mechanical parts on a planar workspace using a parallel gripper. In this paper, we use a deep neural network to solve the problem with a single depth image. To reduce the computation time, we propose an edge-based algorithm to generate potential grasps. Then, a convolutional neural network (CNN) is applied to evaluate the robustness of all potential grasps for bin-picking. Finally, the proposed method ranks them and the object is grasped by using the grasp with the highest score. In bin-picking experiments, we evaluate the proposed method with a 7-DOF manipulator using textureless mechanical parts with complex shapes. The success ratio of grasping is 97%, and the average computation time of CNN inference is less than 0.23[s] on a laptop PC without a GPU array. In addition, we also confirm that the proposed method can be applied to unseen objects which are not included in the training dataset.