Search

Speech-driven 3D facial animation is challenging due to the complex geometry of human faces and the limited availability of 3D audio-visual data. Prior works typically focus on learning phoneme-level features of short audio windows with limited context, occasionally resulting in inaccurate lip movements. To tackle this limitation, we propose a Transformer-based autoregressive model, FaceFormer, which encodes the long-term audio context and autoregressively predicts a sequence of animated 3D face meshes. To cope with the data scarcity issue, we integrate the self-supervised pre-trained speech representations. Also, we devise two biased attention mechanisms well suited to this specific task, including the biased cross-modal multi-head (MH) attention and the biased causal MH self-attention with a periodic positional encoding strategy. The former effectively aligns the audio-motion modalities, whereas the latter offers abilities to generalize to longer audio sequences. Extensive experiments and a perceptual user study show that our approach outperforms the existing state-of-the-arts. The code will be made available., Accepted to CVPR 2022

In recent years, various types of drones have been paid attention to and are expected to play an active role in many fields. However, when flying forward and backward, or right and left, the attitude of the conventionaldrones changes depending on the situation. In order to prevent the drones from changing its posture, several type of drones with tiltable rotors have been attracting attention in recent years. Among them, a tandem-type rotor with 2-DOF tiltable coaxial rotors has been already proposed. In this study, the servo motor to be used for tilting the pitching angle of the tiltable rotor is exchanged to a multi-turned type to exclude the rotational restriction of the tilt mechanism, so that it enhances the motion ability of the UAV. In addition, the size and weight of the machine is reduced so as to make it suitable for a simple experiment in a narrow space. Some experiments are given to demonstrate the effectiveness of the design and manufacture of the current aircraft on the ground motion.

The authors have developed CYber PHysical Overlay Network over Internet Communication (CYPHONIC) that supports accessibility and seamless mobility over the Internet. This paper introduces the detailed implementation of the simple prototype system that supports a basic communication mechanism of CYPHONIC. The developed prototype system consists of cloud service for authentification, end-node management, and a packet relay among end-nodes, and client service for end-nodes. The evaluation result confirmed that the cloud service and the end node operate normally.

Communication connectivity becomes more important according to the spreading of mobile terminals and the Internet of Things (IoT). The authors have designed a new overlay network protocol (CYPHONIC: CYber PHysical Overlay Network over Internet Communication) and have developed the prototype implementation for Linux based OS. This paper develops a shared C library of CYPHONIC to support multiplatform environments such as iOS, Android OS, Linux OS, etc. The evaluation showed that the redesigned implementation could improve the performance and realized the overlay network.

This paper presents an Antenna-Array-Integrated Module (AiM) with a low pass filter (LPF) in LTCC multilayered substrate with low loss, high attenuation and small size in 28 GHz band. A differential feeding method for each patch antenna is employed to improve symmetricity of directivity for beam forming. The insertion loss of the LPF is less than 0.5 dB and the spurious emission in 2nd harmonic band of Band-n257 was suppressed by over 20 dB thanks to putting the LPF in front of the antenna. The symmetricity of the directivity for beam forming was improved owing to the differential feeding.

VHI (Velvet Hand Illusion) is an illusion phenomenon to make someone feel a smooth sensation like velvet fabric and it is expected to be applied to haptic devices. To achieve this goal, it is very important to elucidate the mechanism of VHI. In this research, we investigate whether the brain activation related to VHI can be measured with NIRS. As a result, it is confirmed that the right middle frontal gyrus plays an important role in illusion occurrence which is same as the previous fMRI research.

To develop devices of neurorehabilitation for the stroke patients, we focus on the Kinesthetic Illusion (KI), which is a proprioceptive illusion phenomena. In KI, we feel the joint movement sensation without actual movement, when the vibrating stimulation is applied to a specific tendon. In KI research, it is very important to clarify the optimal condition to induce the stronger illusion, which has not yet been clarified. In this study, we try to clarify the optimal condition for inducing KI from the view point of vibration engineering. This result will contribute toward development of a new mechatronics system inducing KI. In order to the abovementioned investigation, we assume that there is an important relationship between a natural frequency of the specific tendon and the optimal frequency for KI. We focus on the tendon of the Flexor Carpi Radialis muscle (FCR) which is related to the wrist joint movement in the extended direction and conduct a series of psychophysical experiments. As a result, a correlation of R = 0.50 is found between the optimal frequency for KI and a natural frequency of the tendon. This result demonstrates that the joint moving sensation of limbs can be controlled through a basic vibrating stimulation to induce KI by means of applying the natural frequency of the specific tendon.

Smoothness is one of the important factors in controlling texture sensation in tactile VR. In previous research, the authors utilized Velvet Hand Illusion (VHI), which is a tactile illusion phenomena, to generate a smooth texture on a dot-matrix display. Since the dot-matrix display can generate arbitrary figures, we are attempting to evaluate our new hypothesis of the VHI mechanism using this display. In our hypothesis, a cognitive mechanism called Gestalt grouping [3] is related to VHI occurrence. If tactile Gestalt were elucidated, we could understand the tactile sensation mechanism deeply and the various devices would be developed in the VR research field. In this paper, we investigate the relationship between tactile Gestalt and VHI through psychophysical experiments. In these experiments, the appearance intensity of tactile Gestalt is controlled by means of density and length of the lines formed on the dot-matrix display. It is concluded that the law of closure stated in the principle of Pragnanz is an essential condition related to tactile Gestalt to generate VHI.

This paper presents a hierarchical framework for Deep Reinforcement Learning that acquires motor skills for a variety of push recovery and balancing behaviors, i.e., ankle, hip, foot tilting, and stepping strategies. The policy is trained in a physics simulator with realistic setting of robot model and low-level impedance control that are easy to transfer the learned skills to real robots. The advantage over traditional methods is the integration of high-level planner and feedback control all in one single coherent policy network, which is generic for learning versatile balancing and recovery motions against unknown perturbations at arbitrary locations (e.g., legs, torso). Furthermore, the proposed framework allows the policy to be learned quickly by many state-of-the-art learning algorithms. By comparing our learned results to studies of preprogrammed, special-purpose controllers in the literature, self-learned skills are comparable in terms of disturbance rejection but with additional advantages of producing a wide range of adaptive, versatile and robust behaviors., 2018 IEEE-RAS 18th International Conference on Humanoid Robots (Humanoids)

Fostering talented people with excellent OT security is an important issue that is working closely with the industry where cyberattacks are increasing. We examined educational effects and issues in KIPS (Kaspersky Interactive Protection Simulation) multiple exercise based on Gamification theory. Experimental results showed that multiple performances contributed to a simple educational effect and there was a possibility of positive skill transfer, but KOSEN needed educational content that could fill in gaps in multiple exercises, I have a big mission to continually update the practice-based curriculum.

To enhance the reality of touch feeling in VR, one effective presentation is the texture sensation of a virtual object. Although the texture sensation makes persons perceive a virtual object to be a real one, research on texture sensation has not always progressed due to the difficulty of developing a stimulus presentation device. To compensate for this hardware problem, we focus on the Velvet Hand Illusion (VHI), which is one of the tactile illusion phenomena, as a method of presenting texture sensation. In our previous study, VHI was generated by a tactile display using a refreshable braille display (dot-matrix display). Since the dot-matrix display can provide various dot patterns through combinations of stimulus-pin protrusions, it can be applied to clarify the essential stimulus conditions for VHI through a series of psychophysical experiments. We expect to obtain the essential conditions for VHI, which will then become the fundamentals of the VHI model. In the psychophysical experiments, two parallel lines formed by the dot-matrix display were reciprocated on the subject's palm. First, we investigated the relationship between the interval of pins forming lines and the VHI strength. Second, we investigated the relationship between the length of two parallel lines and the strength of VHI. As a result, VHI hardly occurred when the interval between pins forming lines was more than a two-point threshold. Consequently, it was clarified that the sensation of touching a line is an essential condition for VHI and that line length of 29 mm is the absolute threshold of VHI.

To produce human resources who are good at OT security is important task for KOSEN which is closely cooperation with the industry on which faces increasing cyber-attacks. We examined the educational effect and issues on multiple practicing of KIPS (Kaspersky Industrial Protection Simulation) which is based on gamification. Practicing experimental results showed that multiple playing simple contribute to the positive educational effect and we find the possibility of positive skill transfer, however we obtained a great future work which we need the educational contents that can fill the gap of multiple practicing effectively. KOSEN has the big mission that updates the practice-based curriculum constantly.

We develop a new rehabilitation device that combines the effects of two illusions: the kinesthetic illusion (KI) and the rubber hand illusion (RHI). We investigate whether KI can be reinforced with RHI where the former is caused by applying the oscillator's vibration to the flexor carpi radialis muscle (FCR) of a right hand. RHI is caused by showing moving images of a right hand overlapped on the subject's hand through a display pad. The illusion's vividness is induced in the right hand was represented by the left hand and measured by a rotary encoder, which can measure the joint angle of the flexion and the extension motion of the right wrist. We conducted experiments that quantitatively compared the KI strength that was reinforced by the visual stimulus with the pure KI strength, where the enhanced KI strength was about 1.4 to 1.5 times stronger than the unenhanced KI.

The skills of OT(Operational technology) security, has been practiced in the education of ICT security, and the spread of security knowledge, only a small number of specialist training is insufficient. In addition, in OT security that emphasizes the availability, to improve the skills of OT security experts is some non-IT officials, but exercises using the actual equipment is appropriate, it is practically difficult. Therefore, in order to improve OT security skills, it is necessary to practice countermeasures to know security incidents in the same situation as reality, and to learn cooperation by teamwork while improving individual skills. By comparing the results of comprehensive security training and the results of security game, it is revealed that gaming security teaching materials reflect actual security skills and are effective to measure skill level. We try to educate OT security for college students using KIPS (Kaspersky Industrial Protection Simulation). KIPS is a game based on Gamification theory, the game is a security training to learn the basics of cyber security in critical infrastructure in a short period while having fun. We do this exercise for about two hundreds of students and lecturers with two scenarios (water plant and power generation station). Originally, KIPS is suitable for non-ICT engineers, because KIPS needs no deep expertise of security. Therefore, engineering course students can measure security skills without great effort. The teaching materials of the security game are suitable for security education of OT. In a virtual but close to actual industrial/corporate information system, it became possible to utilize the knowledge learned by learners, and it could know their own skill level of security. In addition, individual teaching guidance (such as refinement of weak points) became easier by using teaching materials according to scenario-based contents like KIPS, and it was effective for improving security teaching materials.

This paper presents a hierarchical framework based on deep reinforcement learning that naturally acquires control policies that are capable of performing balancing behaviours such as ankle push-offs for humanoid robots, without explicit human design of controllers. Only the reward for training the neural network is specifically formulated based on the physical principles and quantities, and hence explainable. The successful emergence of human-comparable behaviours through the deep reinforcement learning demonstrates the feasibility of using an AI-based approach for humanoid motion control in a unified framework. Moreover, the balance strategies learned by reinforcement learning provides a larger range of disturbance rejection than that of the zero moment point based methods, suggesting a research direction of using learning-based controls to explore the optimal performance.

In this paper, we propose a novel way to produce voxel-based irregularity age map (IAM) for brain magnetic resonance image (MRI) to identify white matter hyperintensities (WMH) on scans with mild vascular pathology. Age map is a term used in computer graphic field that reveals age/progression of defected areas in images' texture. In this work, age map is used to reveal the age of irregularity of brain tissue (i.e., hyperintensities). Age map of WMH is useful because it shows not only the probability of voxels to be WMH but also the scale in the progression of voxels to become WMH. Our approach is fully automatic and unsupervised with little to none human interaction. We evaluated our approach using brain MRI data obtained from the Alzheimers Disease Neuroimaging Initiative (ADNI) database and visually compared the results with those obtained from the public toolbox Lesion Segmentation Toolbox (LST). We also evaluated our proposed approach on images from 10 different subjects using Dice similarity coefficient (DSC).

For MeV gamma-ray astronomy, We have developed an electron-tracking Compton camera (ETCC) as a next generation satellite telescope. Our detector consists of a Comptonscattering target and a scattered gamma-ray absorber. The target is a gaseous Time Projection Chamber (TPC), which uses $\mu $ PIC as the 2-dimensional readout detector, and the absorber is position sensitive scintillation arrays (PSAs). We developed a prototype ETCC and certificated its performance in a groundbased experiment. The prototype ETCC had a sharp PSF of 15° included 50%, 662 keV) at the ARM of 5° and the SPD of 100°. We plan to update the prototype ETCC and observe the electronpositron annihilation line of the galactic center in Australia in early 2018 (SMILE-II +). In the data acquisition (DAQ) system of the prototype ETCC, the dead time is estimated to be $\sim 20$ % at the trigger rate of $\sim 100$ Hz in a balloon observation, and the prototype ETCC can work. The SMILE-II + ETCC can also work at the balloon altitude. On the other hand, in the future observation with the ETCC, that has a larger effective area than the effective area of SMILE-II + ETCC, the DAQ rate is expected to be $\sim~1$ kHz. The dead time gets saturation due to sending data via VME bus and common start system in PSA. To overcome the two problem, we develop a new TPC readout board and a PSA readout unit. We create a test model ETCC which have the new TPC readout boards and the new PSA readout units. Although the dead time is $\sim 20$ % at 30 Hz with a small ETCC in the prototype DAQ system, the dead time is $\sim 20$ % at 2 kHz in the new system.

As an example of human-like dynamic manipulation, robotic juggling using simple end effectors has been studied. The purpose of this research is to achieve juggling with a multi-fingered hand and arm like those of a human. One serious problem is that the weight of a multi-fingered hand and arm is large, and it is necessary to optimize the movement of the arm in order to compensate for the weight of the hand. In this paper, we propose a method of optimizing the motion trajectory using a smoothing spline. Parameters of the smoothing spline are optimized offline using a genetic algorithm. Another problem is to catch a falling ball accurately by the multifingered hand, and we use a high-speed vision system to estimate the trajectory of the ball. The task to be performed online by the robot hand and arm is to use the optimized parameters to throw a ball up into the air and catch the falling ball with high-speed visual feedback, repeatedly. We show the effectiveness of the proposed method by experiments in which the hand and arm realized throw-up and catch motions seven consecutive times.

We developed a tactile mouse that presents convex dot-patterns on a human palm to obtain the design parameters of haptic devices for virtual reality (VR). In a tactile VR study, we focused on the following issue, “which tactile perception is superior, active or passive touch?”, because it is related to the design of haptic devices that can present detailed information in tactile VR. This study evaluates the differences in perception precision between active and passive touch to clarify the best presentation method in VR through a series of psychophysical experiments using a tactile mouse. Since the presentation precision of virtual figures depends on the sharpness of the edges generated by dot-patterns, our experiment subjects evaluate the slope of the oblique edges. In a psychophysical experiment, two oblique edges of dot-patterns are presented consecutively and subjects compare them to determine which is larger. We evaluate the perception precision using the difference threshold of the edge angles calculated by the simplified constant stimuli method. Experimental results show that the perception precision at low speed (45 and 90 mm/s) of edge movement is better than at high speed (130 and 170 mm/s) and also that there is no significant difference between active and passive touching.