In the aging society, the assistive technology for walking in the rehabilitation and their daily life has been needed and developed. Especially the exoskeletons which assist the lower extremities movement still have difficulties to suit for their comfortable life. In this study, it is aimed to propose and develop a new walking assist mechanism which helps the hemiplegia walking without actuators and sensors. In this paper, the first prototype of the wearable mechanism has been proposed and evaluated in experiment and simulation. In the experiment, no effects of the attachment of the prototype on the walking motion and posture even the prototype accumulated over 4Nm for the walking assist. In the simulation, the features of the assist mechanism which has two assist terms consists of centripetal and eccentric contraction of the hip joint muscles were clarified. As future work, next prototype will be designed according to the walking features of hemiplegic people and evaluated by biomechanical simulation and experiment by subject users.

The localization technique is essential for a mobile robot to perform autonomous navigation safely. Visual odometry (VO) is one of the ego-motion estimation techniques by sequentially tracking the feature points in the image. VO is used for many robots because of its higher accuracy and usefulness in GPS-denied environments, such as indoor robots or planetary exploration rovers. It is required to carefully select the feature points for VO in order to reduce the processing time and improve the estimation accuracy. This paper proposes a new method, which selects feature points by taking advantage of separating rotation and translation estimation by Singular Value Decomposition. The proposed method can select suitable feature points in each. The effectiveness of the proposed method is validated by experiment using a wheeled testbed rover.

This paper describes the development of a cleaning mechanism using planetary gears to clean grease accumulated in kitchen ducts. We aim to develop a robot for cleaning grease in ventilation ducts, which can cause fires. Generally, ducts are cleaned by hand. However, small ducts that are inaccessible to humans have not been cleaned. In this study, we develop a cleaning mechanism using a planetary gear mechanism that can scrub the inner walls of square ducts. In a cleaning experiment, the proposed mechanism can scrub the inner wall of a small duct that was inaccessible to humans at a cleaning rate of more than 30% at once. This result shows the possibility of a proposed mechanism contributing to preventing duct fires by cleaning ducts that have not been cleaned.

In this paper, we briefly explained the curvature estimation of soft fingers with built-in IPMC sensors, and then proposed a method for estimating the size of a grasped object. We derived the equations necessary for estimating the size of an object. Finally, we evaluated the proposed method by estimating the size of three cylinders of different sizes.

This paper proposes the parallel adapter model (PAM) to improve the noise-robustness of automatic speech recognition (ASR) systems with a small amount of retraining. The PAM utilizes two small networks (called adapters) and a corresponding adapter initialization technique. The adapters are positioned in the models where mismatch tends to occur frequently. The experimental results demonstrate that the PAM improved the ASR accuracy with only seven retraining epochs, and the improvement reached 26.1 points with the best performance.

We developed a peristaltic mixing conveyor that imitates the intestines to enable continuous mixing and conveying in the foods and medicines manufacturing process. We are constructing a system to detect the degree of mixing of contents by attaching pressure and flow sensors to the device. To improve the generalizability of the system, mixtures with different mixing ratios of powder and liquid were input into a device. PCA was applied to data from sensors, and classification was performed using SVM, decision trees, and k-NN. As a result, Discrimination rates achieved 100% for all discriminators in the exhaust flow sensor data set.

We propose a state estimation-based fail-safe control framework for water-hydraulic robots driven by Air-Hydraulic Servo Booster (AHSB) to ensure its safe operation. A Kalman Filter estimates the most likely internal states such as water pressures and joint angles, by taking advantage of redundancy of the sensors installed to AHSB. Then, based on the covariance, the system detects and isolates failures such as sensor disconnection, friction fluctuation, and sudden external force. Experimental results demonstrate the effectiveness of the method, wherein, a robotic arm can continue the position tracking task even if a pressure sensor is completely broken.

When a quadrotor is used to inspect a tunnel-like environment such as a sewer culvert, it is known that the thrust of the propeller decreases near the wall due to the circulating flow. This study proposes a conditional distributor that distributes the estimated disturbance to each rotor mounted on the aircraft according to the relative positions of the quadrotor and walls or obstacles. Numerical simulations confirm that the proposed distributor can reduce the risk of collisions by moving the quadrotor away from obstacles while maintaining stable flight by suppressing the thrust reduction.