In this paper, we develop a rotor-distributed aerial manipulator and propose a flight, perching, and end-effector position control. The robot that consists of a foot and rotor-distributed arm module can fly and perch on ceilings using the rotor thrust. For deformation during flight, we improve the linear quadratic intergral (LQI) control using gimbal control. For stable perching, we propose a quadratic programming (QP) based controller to calculate the desired contact wrench, considering the static friction and zero moment point (ZMP) conditions on the footplate. Furthermore, we propose an end-effector control based on the inverse kinematics (IK), considering rotor thrust and joint velocity/torque limitations for stable manipulation during perching. Finally, we verified the end-effector stability and conducted a drill manipulation. The experiments show that the end effector of the multilink aerial robot during perching becomes more stable than those during flight.

This paper addresses multi-work picking systems to shorten the cycle time of bulk picking task. Proposed systems equip robot hands that can pick multi-work and have an ability to find paths that can avoid collision of robots, hands and earlier picking works with surrounding environment and themselves. In collision avoidance of earlier picked works, we realized that by treating them as parts of robots during grasp planning and path planning for the second and the subsequent works. We demonstrate examples of multi-work picking and verify the validity of the proposed systems by comparing the average cycle time with conventional single work picking.

Imitation learning is a promising framework for realizing motor-skill learning by robots. Because it uses guide trajectories, a method of obtaining experts' demonstrations is required for the system. However, how such trajectories can be obtained has not been systematically discussed to date. Herein, we first organize expert demonstration-generation methods and then propose an imitation-learning support system that uses virtual reality (VR) equipment, wherein the experts' trajectories are relatively easy to generate while avoiding various problems arising from a lack of force feedback by appropriately switching subjective and objective viewpoints according to the robot and the target task.

In this paper, we propose a perching mechanism for a multi-rotor unmanned aerial vehicle to perch on the tip of a cylindrical object. The proposed perching system does not require any actuators and uses its own weight to open and close the gripper. In addition, the gripping force of the gripper is larger than its own weight, and the perching is stable. The system is developed such that it can perform stable landing on a flat surface as well.　In order to verify the grasping force, we performed a simulation using dynamics and compared the results with data from the actual aircraft. We have verified the usefulness of this system by actually flying and perching on cylindrical objects. As a result, it was confirmed that the system can generate a force greater than its own weight when the gripper is fully closed during perching, and that the gripper can open and close passively.

We deal with the problem of detecting anomalous objects at the pixel level for the purpose of picking anomalous objects up by the robot in the soil recycling process. A previous convex detection method does not distinguish anomalous objects and soil with stones, which makes it impossible for the robot to pick them up. In this paper, we focus on that we can get a lot of data of soil and stones although there are few data of anomalous objects. We get color images and depth maps of soil and stones by using a RGBD camera. Using the color images and depth maps, GAN is trained to estimate the depth map from the color image. GAN cannot estimate the depth map of anomalous objects because there are few training data of anomalous objects. Then, we can detect anomalous objects by comparing the real depth map and the depth map estimated from color image by using GAN. Experimental results show that our method can detect anomalous objects 2.4 times more accurately than the previous method can do. Moreover, we confirm that our method enables the robot to estimate grasping position of anomalous objects more accurately.

Light sheet microscopes are microscopes that can capture images with low noise by illuminating only the focal plane with sheet light. In the conventional 3D measurement of light sheet microscopes, the observation object or the objective lens is moved for measurement, but the slow movement of the stage makes it difficult to control at high speed. In this study, we proposed a high-speed light-sheet microscope that uses TAG lenses and galvanometer mirrors to focus and control the sheet light in milliseconds. From the experiments using a developed prototype system, we were able to measure 3D information consisting of 10 layers of images in 10 ms, and confirmed the high speed of 100 volume/s.