| WU POYEN | D, 中間発表 | ロボットラーニング | 松原 崇充, 和田 隆広, 鶴峯 義久 |
|
title: Reinforcement Learning of Tool-use Policies that Prolong Remaining Useful Life in Repetitive Tasks
abstract:Robots are increasingly used for repetitive tasks that show cyclic motion. These repetitive tasks require frequent tool replacements, which can be costly and time-consuming. Limiting or minimizing external torque through kinematic or mechanical constraints ensures the safety of the tool use policy avoiding frequent replacement due to greater damage. However, these methods do not actually guarantee an optimal tool lifespan. In tool use tasks, the overall life of the tool depends on the variation of the stresses, which are also influenced by the shape and material of the tool, making it difficult to develop an optimal tool use policy. To address this problem, we propose a Robotic tool use evaluation framework based on Reinforced Learning (RL) and Remaining Useful Life (RUL). Specifically, this framework predicts the Remaining Useful Life of the tool through Finite Element Analysis and Miner's Rule and incorporates it into the RL reward function, this combination approach forces policy learn to achieve a goal while maximizing the RUL. The effectiveness of the proposed method is verified through experiments using a pushing and dragging task with the tool in both simulations and a real robot, resulting in the method is effective in prolonging tool lifespan and working well for tools of different shapes compared to methods that do not consider lifespan. language of the presentation: English | |||
| 浦田 龍図 | M, 2回目発表 | ヒューマンロボティクス | 和田 隆広, 加藤 博一, 織田 泰彰, 劉 海龍 |
|
title: Investigation of feeling unnaturalness of observers and prediction method of eye movements for movable prosthetic eye
abstract: Posthetic eye cannot move. Regarding a unilateral prosthetic eye, it is challenging to achieve synchronized eye movement between the natural eye and prosthetic during everyday gaze shifts. The lack of motility in a prosthetic eye can trigger unnaturalness in observers who interact with prosthetic eye users. For people using prosthetic eye, this is a crucial concern in addition to the aesthetic nature and immobility of the prosthesis. To solve this problem, a movable prosthetic eye is required. In this study, we focus on a mechanical prosthetic eye system that moves by estimating the eye gaze based on the results of measured eye movements of the unaffected eye, and address two problems: “investigation of unnaturalness held by the observers” and “prediction of gaze direction.” In the unnaturalness verification, we investigate the acceptable range of time delay due to mechanical factors, such as dead time and first-order delay, which are thought to occur during actual prosthetic eye movements. Even if a difference in movement by time delay in between the left and right eye is discernible, there may be a range within which people do not feel unnaturalness. It is possible to move prosthetic eye without feeling unnaturalness of observers by moving it within those acceptable range. We conduct verification due to the limited understanding regarding the unnaturalness caused by delays. In prediction, we propose a method to predict the direction of eye gaze on the prosthetic eye. In actual prosthetic eye movement, mechanical delays such as dead time and first-order delay are occur. This delay makes observers potentially experience a sense of unnaturalness. Therefore, it is necessary to predict eye movements a little ahead of time to move the prosthetic eye. In this study, the eye gaze of the prosthetic eye is predicted based on the direction of gaze of the unaffected eye, which can be measured with a wearable device. Specifically, the angular velocity of the unaffected eye's gaze is used to estimate which object in space to focus on according to an eye movement model. Based on the results, the eye gaze of the prosthetic eye is predicted several frames (0.08 to 0.2 seconds) in the future. language of the presentation: Japanese 発表題目: 可動型義眼のための対面者が抱く不自然さの調査と眼球運動予測手法の構築 発表概要: 義眼は一般に動かすことができないため，片側が義眼である場合，両眼で同期した動作が行われず，対面者が不自然さを抱く．これは審美的なことに加え，義眼使用者の悩みの要因となっている．この悩みを解決するために，可動型義眼が求められている．そこで本研究では，健側の眼球運動を計測した結果から義眼側の視線方向を推定して動作させる機械的な義眼システムに焦点を当て，「対面者の抱く不自然さに関する調査」と「視線方向の予測」の2つの問題について取り組む． 不自然さに関する調査では，実際の義眼動作時に発生すると考えられるむだ時間や一次遅れなどの機械的な要因による時間遅れの許容範囲について調査する．義眼の動作の遅れが対面する人に気付かれたとしても，それが不自然さを抱かない程度であれば，その範囲内で義眼を動作させることで，対面者に不自然さを感じさせない動作が可能になる．しかし，義眼動作の遅れによる不自然さに関する文献は存在しないため，不自然さの調査を行う． 予測では，義眼側の視線方向を予測する手法を構築する．実際の義眼動作時には，むだ時間や一次遅れなどの機械的な遅れが発生することで，義眼の動作に遅れが生じて対面者が不自然さを抱くことになる．そこで，少し先の視線方向を予測して義眼を前もって動作させることでこの遅れを抑制し，対面者に不自然さを抱かせない動作を可能にするために予測が必要となる．本研究では，ウェアラブル端末で計測可能な健側眼球の視線方向をもとに義眼側の視線方向を予測する．具体的には，健側眼球の視線の角速度から，眼球運動モデルに従って，空間内のどの物体に焦点を移すのかを推定する．その結果から数フレーム（0.08～0.2秒）先までの義眼の視線方向を予測する． | |||
| 中村 俊介 | M, 2回目発表 | ヒューマンロボティクス | 和田 隆広, 松原 崇充, 織田 泰彰, 劉 海龍 |
|
title: Inter-connection: Generation of vehicle motion considering ride comfort abstract: The development of autonomous driving systems and the introduction of digital devices are progressing. Thereby, the possibility of developing motion sickness increases as the time spent working inside the vehicle increases. In recent years, active suspension seats have been developed and path planning methods for autonomous vehicles have been proposed to reduce motion sickness. However, few studies have controlled vehicle motion that is thought to have a major influence on the onset of motion sickness. Therefore, in this research, we aim to generate vehicle motions that reduces motion sickness. There are two main hypotheses for the onset of motion sickness: SVC (Subjective Vertical Conflict) and postural instability theory. Following these theories, we achieve vehicle motion to reduce motion sickness and muscle strain with a model-based design the SVC model and the postural control model. language of the presentation: Japanese 発表題目: Inter-connection: 乗り心地を考慮した車両運動の生成 発表概要: 自動運転システムの進展，車載デジタル機器の導入，車内でのモバイル機器の日常的な使用に伴い，車両内の作業時間が長くなるにつれて乗り物酔い（動揺病）発症の可能性が高まる． 近年，動揺病低減を目的としたアクティブサスペンションシートの開発や自動運転車両の経路生成手法の提案が行われている．しかし，動揺病発症に及ぼす影響が大きいと考えられる車両運動を制御した研究は少ない．そこで，本研究では，動揺病を低減する車両運動の生成を目指す．特に動揺病発症の有力仮説であるSVC(Subjective Vertical Conflict)や近年注目されているPostural instability theoryに基づき，SVCモデルや姿勢制御モデルを用いたモデルベースデザインによる動揺病や筋負担の低減を目的とした車両運動を生成した． | |||