The following three points have been investigated for improving the flight performance of a multiple rotor drone. The first one is `collective pitch control', which enables rotors to change the thrust much quicker than the rotor speed control method．Furthermore, a negative thrust can be generated by the rotors with a negative collective pitch. The negative thrust can cause a large control moment for controlling the attitude. Then, the drone has high maneuverability thanks to the quick thrust change due to the collective pitch control. The second one is `decreases of the moments of inertia', which can enhance the maneuverability of the drone. The variation of moments of inertia has also possibility to decrease a motion of the drone due to a disturbance moment when a feedback control is adopted. The third one is `a rotor with small aspect ratio rotor blades', which generates a larger thrust caused by the larger chord length and the larger thrust coefficient.

In a small room such as an operating room where there are multiple staff, some orders should be transmitted to a selected person. For achieving this without any wearable devices, we developed a voice transmission system using a parametric speaker. To use a parametric speaker in a small room, we had some experiments to investigate sound pressure characteristics in a small room and understandability of the transmitted voice with MOS. As a result, it was confirmed that the transmitted voice, which has possibility that reflect on the wall and reach to a not selected person, could be heard clearly.

Unmanned construction machine working in dangerous environments such as construction sites and disaster areas has been developed. However, it is still necessary to improve its work efficiency especially during bulldozing and excavating soil. This research aims to develop a method for predicting soil deformation using machine learning. The feasibility of the proposed method is verified in a scenario where a simple bulldozing blade excavates soil. In the experiment, soil deformation at a front part of the blade is captured by multiple stereoscopic cameras. The camera provides depth data that are then converted to height field data. This dataset is fed to machine learning using Recurrent Neural Network (RNN) because soil deformation is continuous phenomena depending on time variation. The learned model for predicting soil deformation is confirmed in varied intrusion depth of the bulldozing blade.

This paper reports social acceptance of a childcare support system with AI/robot through web-based survey and a field trial in Japan. The participants of both the web-based survey and the field trial showed positive attitudes toward the childcare support system with AI/robot. They also preferred a collaboration style between robots and people compared to fully autonomous/teleoperated style, although they preferred both a collaboration style and a fully autonomous style between AI and people.

Water-hydraulic system is environmentally friendly technology, and its application to robot manipulators is eagerly awaited. However, they are still expensive compared to oil-hydraulic systems because of the technical problems related to fast sliding motions between the solid materials contacting each other with the very small gap in the pumps or valves. To avoid these difficulties, we proposed an alternative solution, which lead to a pump-less hydraulic system called AHSB. The system utilizes an air-hydraulic booster to control the low-to-middle range water pressure by pneumatic servo valves. In this paper, we analyze the water-hydraulic dynamics and develop a robust pressure-based joint torque controller using Sliding Mode Control. The controller is implemented on a single-axis robot arm for the remote torque control application, which is crucially important for future tele-operated field robots. The experiments under quasi-static condition demonstrates the first successful torque control results on the water-driven robot with 10 kg payload, which is connected with 20 meter-long hydraulic hoses.

This paper presents a novel knowledge transfer method for heterogeneous robot systems. Leveraging a learned model of a robot, another robot improves its learning efficacy. A main problem we tackled is to overcome discrepancy of inputs/outputs in the two systems. We introduce a method to extend neural-network model inspired by Net2Net; and derive regularization term based on Kullback-Leibler divergence between the model parameter distributions to stabilize learning process. Simulation of transferring a learned 6 DoF manipulator model to a 7 DoF manipulator model demonstrated that our method can improve sample efficiency of reinforcement learning to optimize control law of the 7 DoF manipulator.

In the power distribution work using the indirect live-line method, it is necessary to frequently lift a tool weighing 15[kg] and hang it on the electric wire. To reducing the load on workers, this paper proposed a motor-less power assist device that can compensate the weight of the tool in the wide range of motion: vertical 600[mm], horizontal 600[mm], pitch 90[deg]. The device has weight of 16[kg] and depth of 200[mm], which is lighter and more compact than conventional ones, so that it can be mounted on the side of the bucket of aerial work platform. To verify the effectiveness of this device, we measured the surface myoelectric potential of the forearm flexor muscle group during the overhead wiring work and the time required for the work. Through the experiments, it was confirmed that the load on the muscles was reduced by 90% or more by using this device, and the working time was almost same as the case without assistance.