This paper proposes a mobile robot system which realizes mobility assistance with a manual wheelchair when a handcar chair user needs moving assistance. The mobility assistance is triggered by recognizing wheelchair user's demand by using visual sensor. We suggest a task consisting of three phases for realizing automatic connection of a wheelchair by a robot. The first phase is a proposal and verification of a method to estimate the actual position and orientation of a wheelchair from the depth information of the Kinect v2 sensor. The second phase is the implementation of the path planning of the robot using the estimated position and orientation of the wheelchair as the target value coordinates and its running control mode. The third phase is a mode in which the robot connects to the wheelchair hand unit from behind the wheelchair, and is integrated with the wheelchair as it is in the steering mode and autonomous driving mode. The effectiveness of automatic wheelchair connection system that integrates the tasks of each phase is verified by actual traveling experiments.

This paper tackles assembly sequence generation considering two trade-off objectives; one is insertion conditions and the other is degrees of constraint among assembled parts. The multi-objective genetic algorithm is used to balance the two objectives needed for planning of robotic assembly. Furthermore, the method of extracting parts relation matrices is extended for applying to 3D CAD models including deformable parts. Our simulation experiment using the proposed method shows a possibility of obtaining pareto optimal solutions of assembly sequences for a 3D CAD model with 33 parts including a deformable part.

Recognition of contact events is often required for contact-based tasks such as assembly and grinding. This paper proposes a recognition method based on the Mel Frequency Cepstrum Coefficient (MFCC) of force signals, which can be derived from an ordinary robot configuration with a force sensor and a manipulator. We demonstrated that the combination of MFCC and time delayed neural network is effective in the case of the contact event recognition. It was confirmed that the above method has the ability to recognize instantaneous responses that do not generate repetitive waveforms.

The recent growth of robotic manipulation has resulted in the realization of increasingly complicated tasks, and various kinds of learning-based approaches for planning or control have been proposed. However, learning-based approaches which can be applied to multiple environments are still an active topic of research. In this study, we aim to realize tasks in a wide range of environments by extending conventional learning-based approaches with parameters which describe various dynamics explicitly and implicitly. We applied our proposed method to two state-of-the-art learning-based approaches: deep reinforcement learning and deep model predictive control, and realized two types of non-prehensile manipulation tasks: a cart pole and object pushing, the dynamics of which are difficult to model.

In this paper, the authors described the modeling method of the instantaneous force generation mechanism inspired by the striking motion of Odontodactylus scyllarus and the evaluation of the mechanism. Animals use the contraction of living muscle as actuators to perform various movements. Some arthropods produce instantaneous movements by using the combined force of the exoskeleton spring and muscle force. Few reports deal with such an instantaneous force generation mechanism of the combined force in the numerical simulation. Our purpose of this study was to construct the instantaneous force generation mechanism by the combined force of elastic elements of the exoskeleton and pneumatic artificial muscles, inspired by the instantaneous force generation mechanism of Odontodactylus scyllarus. In the previous study, we have confirmed that the instantaneous force generation device and the possibility of generating instantaneous force by the physical model. In this paper, the authors further constructed a model of the proposed mechanism as a translational rotation system and performed a dynamic analysis by numerical simulation, confirming the effectiveness of the proposed model.

This paper proposed a developmental support system of toddlers' language skills by using a telepresence childcare robot. The childcare robot can attract toddlers' attention during interaction based on its actions. This ability is a significant advantage that differs from static devices (i.e., tablet). Our system and robot played language quizzes with toddlers to collect developmental information about toddlers' language skills. They made reports for a childcare worker to help their assessment in the visiting support program. Experimental results showed that our system has the potentials to evaluate toddlers' language development quantitatively and support childcare workers in the visiting program.

Interview survey is one of the options for investigations with light loads on participants to study how people fall compared with measurements by many sensors. In this paper, we aimed at predicting fall patterns from interview text data. We use k-means clustering method to confirm the validity of the labels attached to the interview data, and also confirmed the validity of the summaries of the interview data by interviewer researchers by focusing on the co-occurrence word analysis. After confirming the validity of the labels and summaries, we construct a naive Bayes model classifiers to classify the fall patterns. The average classification rate was 61.1% for 3 types of falls - falls by an unexpected external force, by losing balance or supports, and by other reasons.