In this paper, we propose methods to acquire clear images from vibrating objects in the XY plane using a high-speed vision system. These methods consist of a hardware method using a galvanomirror and a software method that detects the amount of object movement. In order to verify the proposed methods, we have formulated these methods by experimenting with field oscillations caused by mirrors. A clear image could be obtained for vibrations with a higher frequency and larger amplitude than those generated in general equipment. The integrated system of these methods has achieved further improvement in specifications.

Since many of Japan's infrastructures were constructed during the high-growth period, the current infrastructures are aging and have become a serious social problem. Currently, periodic inspections are basically conducted visually by workers, but the introduction of robotic periodic inspections is being promoted. However, periodic inspections using robots have several problems. Therefore, this study proposes the development of an auto-steerable thrust mechanism with multi-rotor and its control method. We designed a model of the thrust mechanism and showed that the equilibrium point of the posture of the mechanism coincides with the direction of the target net force. It was also found that the equilibrium point deviates from the direction of the target net force when there is a difference in the thrust of each rotor. We developed a thrust mechanism that has a rotation axis at the center of the two rotors and can control the direction of each rotor. Using the developed thrust mechanism, the model was verified through two experiments. In the first experiment, it was shown that the equilibrium point of the mechanism's posture coincides with the direction of the target net force. It was also confirmed that the equilibrium point shifted when the difference in thrust between the left and right rotors occurred. In the second experiment, we confirmed from the step response of the developed mechanism that the response changes when the magnitude of the net force is changed. Although there were some errors from the model, it was shown that the trends of each element were consistent when the magnitude of the net force was varied.

This study focuses on the suboptimal spacecraft orbit design using a deep reinforcement learning (DRL) method. In the optimal orbit design issues, it is not easy to solve the nonlinear problems and we have many difficulties to solve them. However, reinforcement learning can be applied regardless of linear or nonlinear. Therefore, quasi-optimized solutions can be achieved more simply. In this paper, we show the effectiveness of the DRL algorithm in the orbit design in the case of super synchronous and spiral orbital transfer problems.

This paper discusses how to estimate the ground surface from a point cloud measured by airborne laser scanning. Airborne laser scanning is a measurement system which collects point clouds by laser emitted from hovering drones. Although airborne laser scanning is a useful surveying method, a measured point cloud includes a lot of unnecessary data which corresponds to trees. To remove unnecessary data, this paper proposes a method to estimate the ground surface from measured point cloud. In particular, this paper focuses on Non-Parametric Piecewise Affine (NPPWA) map to approximate ground surface, and employs hard and soft constraints; 1) measured point cloud must lie above the surface, and 2) the surface should be smooth. The effectiveness of the proposed method is demonstrated with real data.