| 熊本 光 | M, 1回目発表 | ロボティクス | 小笠原 司, 清川 清, 高松 淳, Gustavo Garcia |
|
title: Development of an underwater robot motion system equipped with an underwater suction manipulator
abstract: An underwater robot equipped with a manipulator performs maintenance work at a submarine plant, leading to more efficient marine resource development, lower costs, and improved safety. Focusing on the suction gripper as one of the manipulators, it can grasp objects of various shapes and sizes, and is widely used in industrial robots and service robots. However, since the suction gripper is pneumatically driven, it is difficult to apply underwater. In this study, we develop a hydraulically driven underwater suction gripper based on hydrodynamics. Furthermore, in order to realize to pick and place an object, we propose an interlocking system with an original gripper and underwater robot. language of the presentation: Japanese | |||
| 角川 勇貴 | M, 1回目発表 | 知能システム制御（ロボットラーニング） | 杉本 謙二☆, 池田 和司, 松原 崇充 |
|
title: Robot Control by Spiking Neural Network for Low Power Consumption
abstract: Reinforcement learning using DNN (Deep Neural Network) has been attracting attention with the success on ATARI games, and has been used in the robotics field. Using GPU to speed up DNN calculation consumes a lot of electric power, so using DNN for many robots can lead to power shortages. In recent years, the number of industrial robots has been increasing, so the problem of power shortage is expected to become more serious. However, the Neurochip is considered to solve the problem because of its low power consumption. It is difficult to calculate algorithms like conventional DNNs with Neurochip. Therefore, we introduce a method of applying SNN (Spiking Neural Network) to reinforcement learning which is one of the representative machine learning algorithms that can be calculated by Neurochip. However, SNN has been used for classification problems conventionally. In this study, it is necessary to solve regression problem in order to obtain the value function of reinforcement learning. Therefore, as a preliminary study of applying SNN to reinforcement learning, we present an extension of eRBP (event-driven Random Back Propagation), proposed as a classification method, to the regression method, and show the simulation results. language of the presentation: Japanese | |||
| 榎本 貴仁 | M, 1回目発表 | 知能システム制御 | 杉本 謙二, 安本 慶一, 小林 泰介 |
|
title:Shared Autonomy for Overlooked Visual Field
abstract:Recently, autonomous driving is required to resolve traffic accident, traffic congestion, and tailgating. Two kinds of the autonomous driving have to be developed depending on drivers’ demands: full autonomy and shared autonomy. This research focuses on the shared autonomy. The shared autonomy is to cooperate a driver and a system. When the driver is in danger, the system helps the driver avoid accidents. Machine learning is one of the promising approaches for this purpose, although it needs large dataset. To improve sample efficiency, previous work has employed autoencoder, which finds low-dimensional latent space in the input. As another way for sample-efficient learning, attention mechanisms are raised. If the system can find attentive visual fields of the driver, the system can assist the driver only for the overlooked visual fields while reducing the number of samples. To this end, this study first proposes the way to find the attentive visual fields of the driver using imitation learning with attribution map. The proposed method is applied to a wheel chair with three RGB-D cameras. In this system, the proposed method succeeds in imitating the driver’s policy and finding the attentive visual fields. language of the presentation:Japanese | |||
| SHI CHENGJIE | M, 1回目発表 | 数理情報学 | 池田 和司, 佐藤 嘉伸, 吉本 潤一郎, 久保 孝富, 福嶋 誠 |
| 大木 麻里衣 | M, 1回目発表 | 数理情報学 | 池田 和司, 松本 健一, 吉本 潤一郎, 久保 孝富, 福嶋 誠 |
|
title: Evaluation of Emotional Impact of Music Based on Sentiment Polarity of Lyrics and Heart Rate Variability Analysis
abstract: Music affects our emotion and causes physiological changes. Melody and lyric are important factors for the emotional impact. However, there are few studies that investigate the relationship between emotional impact and impression of words in lyrics. Also, we assume that singing and instruments have a role in amplification of impression and emotion. In this study, we assess the sentiment polarity of lyrics with natural language processing, and conducted an experiment where we collected questionnaire about emotional impact of music and heart rate. In this presentation, I will report preliminary results about the emotional impact of music. language of the presentation: Japanese | |||
| 澤田 悠冶 | M, 1回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
|
title: Similarity-based Coordinate Span Identification for Named Entity Recognition
abstract: In named entity recognition(NER) task, these models assume that named entity is contiguous. However, named entities of chemical or biochemical domain often emerged discontiguous in scientific papers. These models cannot correspond to named entities had complicated structure, thus to extract chemical named entities such as Metabolites or Species is an issue for recent NER model. To resolve it, we consider to employed coordinate structure parsing to NER models. Coordination parsing aims to identify the span of parallel phrases centred at coordination conjunction, thereby it seem to recognize the discontiguous entity had a coordinate structure. In this presentation, we propose a method based on the similarity between word sequences adjacent to coordination as the first step and mention the comparison result with the conventional parser. language of the presentation: Japanese | |||
| 岡 佑依 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 宮尾 知幸(MS), 須藤 克仁 |
|
title: Control Output Sequence Length using Prediction from Input
abstract:In neural machine translation(NMT) task, NMT model sometimes outputs too long sentences or too short sentences. This problem is called over translation or under translation. We propose control output sequence length using predicition from input and positional encoding. Length-ratio positional encoding(LRPE) can controll output length in summary task. LRPE needs length to output, which is predicted from input sequence in our proposed method. In this study, we analyze transformer model output length and compare transformer model output length and our prediction output length.At the same time, we try to translate using our proposed method. language of the presentation: Japanese | |||
| 国広 有衣子 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 作村 諭一(BS), Sakriani Sakti, 須藤 克仁 |
|
title: Tree-To-Speech for Appropriate Prosody of Synthesized Speech
abstract: Human speech includes prosody. The prosody means all speech information which disappears when the speech is written such as intonation, pause, etc.. It represents many features such as emotion, emphasis and also the structure of the sentence. Inappropriate prosody cannot describe the structure or the intent properly, therefore it may cause miscommunication. Recently, Tacotron is known as a speech synthesis system which predicts spectrogram directly from text. However, it may generate inappropriate prosody since it does not consider the structure of the sentence. We aim to synthesize speech with appropriate prosody by using parsed trees. In this presentation, we introduce some related works and show the future plan. language of the presentation: Japanese 発表題目: 構文木情報を用いた適切な韻律の音声合成 発表概要: 通常、人間の話す音声の韻律には、文の構造などの発話意図に関わる情報が含まれている。 ここでいう韻律とは、イントネーションや息継ぎなど、テキスト化した時に失われる音声情報全般を指す。 韻律の不自然な音声は、文意を適切に表すことができず、円滑なコミュニケーションを阻害する恐れがある。 一方、近年注目されている音声合成技術にTacotronがある。Tacotronは文字から音声のスペクトログラムを直接予測するが、文の構造は考慮されていないため、 不適切な韻律となる可能性がある。 そこで、構文木を用いて文の構造を考慮することにより、適切な韻律を持つ合成音声の実現を目指す。 本発表では、関連研究を紹介するとともに、今後の研究計画についても言及する。 | |||
| 岩田 和弘 | M, 1回目発表 | ネットワークシステム学 | 岡田 実, 笠原 正治, 東野 武史, Dong Duong Thang |
|
title:An investigation on near-field communications by the magnetic induction in under seawater
abstract:Recently, wireless communication methods with submersibles (manned or unmanned) are attracting great attentions. Acoustic communications are currently employed for many applications. However, this method is limited in the transmission rate, and simultaneous transmissions and reception with this method is impossible. Meanwhile, the wireless communications using radio waves suffer from severe attenuations when carried out under the seawater. To avoid this severe attenuation in the radio propagation, near-field communications (NFC) which is based on the magnetic induction has been proposed. In this work, we employ computer simulations to emulate an NFC communication link and measure its signal-to-noise ratio (SNR). We observed an SNR of 11 dB for a MI link at a distance of 50m, operating at the center frequency of 3.5 MHz over a bandwidth of 10kHz. This result show feasibility of the NFC link. language of the presentation: Japanese | |||
| 喜島 大揮 | M, 1回目発表 | 光メディアインタフェース | 向川 康博, 清川 清, 舩冨 卓哉, 田中 賢一郎, 久保 尋之 |
|
title: Distance and intensity correction from pulsed Time-of-Flight camera measurements in foggy scenes
abstract: Recently, Time-of-Flight (ToF) cameras have been used for various applications, such as autonomous driving. However, ToF camera measurements are distorted in foggy scenes because of light scattering, and this makes ToF cameras unreliable in autonomous driving. In this research, our goal is developing an algorithm which corrects distance and intensity from ToF camera measurements in foggy scenes simultaneously. As the first step of our research, we modeled an inpulse response in fog based on single scattering model for the purpose of modeling the observed light in foggy scene. language of the presentation: Japanese | |||
| 磯田 祥吾 | M, 1回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 中村 哲, 諏訪 博彦, 松田 裕貴 |
|
title: Sightseeing spot recommendations with reasons considering changes in the tourist destination context
abstract: In recent years, the demand of the tourism industry has increased. Accordingly , research on tourist spot recommendations optimized for individuals is actively conducted. However, Since tourist spots recommendations that are optimized for individuals are estimated in advance, there is a lack of flexibility in dynamic changes in the tourist destination context. This study takes into account dynamic changes in the tourist destination context such as congestion, weather conditions and unforeseen events, and supports decision making with consideration of trade-offs by presenting the next spot to visit with a reason. language of the presentation: Japanese | |||
| 後上 正樹 | M, 1回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 中村 哲, 荒川 豊（客員）, 松田 裕貴 |
|
title:
abstract: [ Private due to patent pending ] language of the presentation: Japanese | |||
| 杉山 健太 | M, 1回目発表 | インタラクティブメディア設計学 | 加藤 博一, 小笠原 司, 神原 誠之, Alexander Plopski, 藤本雄一郎 |
|
title: Using Virtual Simulation Towards Improving Propulsive Efficiency in Freestyle Swimming
abstract: Propulsive efficiency is important in long distance swimming race. Propulsive efficiency means how much proportion of the energy a swimmer put into water the swimmer can get from water. In freestyle swimming, there are several styles and parameters. Swimmers search for the best movement through repeated training. In this research, our goal is proposing the most efficient movement through virtual simulation to help swimmer training. We simulate freestyle swimming in a virtual pool with changing styles and parameters, then we compare each propulsive efficiency. language of the presentation: English | |||
| 人見 一宇 | M, 1回目発表 | インタラクティブメディア設計学 | 加藤 博一, 向川 康博, 神原 誠之, Alexander Plopski, 藤本雄一郎 |
|
title: Complimentary Pattern Projection Method with Projection Surface Color
abstract: Recently, there are many QR code everywhere which is detected by camera to access detail information. QR code has many advantage. For example, the camera detect the information easily and stably or people can understand where they should capture. However, sometimes QR code is an obstructive to see what people want to see. In order to avoid this situation, there is the complimentary pattern projection method. This method theory is that changing two colors serially above a certain frequency makes the mixed color which mixed these colors. The number of color combination is very high. In this research field, there are some related works and one reveals which color combinations are difficult to perceive. But there is few method with the projection surface color. In this study, I would like to reveal which color combinations are difficult to perceive with some projection surface colors. language of the presentation:Japanese | |||
| 横見 栄聡 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 加藤 博一, 酒田 信親, 磯山 直也 |
|
title: Gaze guidance method in VR environment using resolution control
abstract: Virtual Reality (VR) can see any places in VR environment by using Head Mounted Display (HMD). However, in such VR contents, it is a problem that the VR Experienced person does not always pay attention to what the content creator wants to show. On the other hand, it has been shown that gaze guidance can be performed by controlling the resolution in a two-dimensional image. By using this method, it is not necessary to give an arbitrary stimulus to the user, and it is expected that the natural user experience will not be hindered by gaze guidance. Therefore, we propose a method using resolution control as a gaze guidance method in VR environment. In addition, the effectiveness of the gaze guidance by the proposed method is verified by subject experiment. language of the presentation: Japanese 発表題目: 解像度制御を用いたVR環境での視線誘導手法 発表概要: Head Mounted Display（HMD）によるVirtual Reality（VR）の体験者はVR環境中の任意の場所を見ることが可能である．しかし，このようなVR環境で利用されるコンテンツでは，制作者が見てほしいものをVR体験者が注目するとは限らないことが問題となっている．一方，二次元画像で局所的な解像度制御によって視線誘導が可能であるのが示されていることから，VR環境においても，同様の手法で視線誘導を行える可能性があると考えた．この手法を用いるとユーザに恣意的な刺激を与えたり，明らかに目立つ標識を使用したりする必要がなく，自然なユーザエクスペリエンスを妨げずに視線誘導が実現されることが期待される．そこで，本研究ではVR環境で解像度制御を用いた視線誘導手法を提案し，被験者実験によって提案手法による視線誘導の有効性を検証する． | |||
| 喜多山 湧也 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 加藤 博一, 酒田 信親, 磯山 直也 |
|
title: Basic Investigation of Information Presentation Only to Peripheral Vision
abstract: In this research, we propose an information presentation method for the peripheral vision only, and consider the scenarios and specific effects that such information presentation is expected to be effective. Information presentation only to the peripheral vision can present information without obstructing the central vision, and various new applications are expected. For example, it may be possible to evoke a sense of security and empathy by displaying the face image of another person on the peripheral vision display. If we can find a scenario where the proposed method is effective through subject experiments, it will be a new tool in related research fields. Preliminary experiments were carried out using a prototype that assumed sympathy scenarios, and the effectiveness of this method was confirmed. language of the presentation: Japanese 発表題目: 周辺視野のみへの情報提示の基礎的検討 発表概要: 本研究では，周辺視野のみに対する情報提示手法を提案し，そのような情報提示が有効であると期待されるシナリオと具体的な効果について考察する．周辺視野のみに対する情報提示は，中心視野を阻害することなく情報を提示でき，様々な新しい応用が期待される．例えば，周辺視野ディスプレイに他者の顔映像を表示することで，安心感や共感を喚起することが考えられる．被験者実験を通じて提案手法が有効であるシナリオを見出だせれば，それに関連する研究分野の新たなツールにもなると考えられる．実際に試作機を用いて共感を与えるシナリオを想定した予備実験を行い，本手法の有効性を確認した． | |||