| 市原 有生希 | M, 1回目発表 | 数理情報学 | 池田 和司, 川鍋 一晃（客員教授）, 杉本 徳和（客員准教授）, 田中 沙織 |
|
title: *** Towards Robust Reinforcement Learning Algorithms for Nominal Environment ***
abstract: *** One of the challenges in real-world applications of reinforcement learning is the mismatch between simulated and actual physical environments. In other words, when a policy learned in a simulated environment is applied to an actual environment, it will not produce intended actions due to several factors, such as noise, modeling error, and environmental conditions. To address this problem, the Robust Markov Decision Process (MDP) focuses on developing robust algorithms. In robust MDP, a set of environmental transition probabilities and reward functions are considered, and then, the worst-case transition probabilities and reward functions are adopted and used for learning. Recent studies have shown that the worst-case scenario of a given reward function can be considered by including the entropy and divergence of the policy. *** language of the presentation: *** Japanese (choose one) *** | |||
| 吉田 雄丸 | M, 1回目発表 | 数理情報学 | 池田 和司, 川鍋 一晃（客員教授）, 杉本 徳和（客員准教授）, 田中 沙織 |
|
title: *** EEG-based Emotion Recognition To Express Richer with Semantic Space Theory ***
abstract: *** This research plans to research EEG-based emotion recognition based on the Semantic Space Theory of emotions. This research focuses on how finely distinguishing emotions impacts stress resilience, and is developing novel methods to classify emotions more precisely using EEG data. In this experiment, brainwaves are measured while participants watch video stimuli, using this data to explore the relationship between emotional diversity and brain activity patterns. This approach has the potential to offer more detailed insights into emotion recognition compared to traditional methods. Furthermore, I delve into the connection between emotional granularity and an individual's psychological adaptability, aiming to gain a new understanding of emotion recognition. *** language of the presentation: *** Japanese *** | |||
| 市 尚都 | M, 2回目発表 | ロボットラーニング | 松原 崇充, 池田 和司, 花田 研太, 佐々木 光 |
|
title:Optimal Control Method for Explosive Motion with Soft Under-actuatede Manupilator
abstract:Unlike typical industrial robots, soft under-actuated manipulators can accomplish explosive motion(push-up motion and swing-up motion). However, optimal control of this motion requires stable long-term predictive model and identification of nonlinear dynamics. In this study, we focused on Deep Koopman Networks, a data-driven modeling based on Koopman operator theory, and proposed a learning framework that can solve the problem of soft pendulum models. Basic verification of the proposed framework confirmed that the structure explicitly incorporating affine systems is effective in optimal control. language of the presentation:Japanese | |||
| 矢野 嘉希 | M, 1回目発表 | ロボットラーニング | 松原 崇充, 和田 隆広, 鶴峯 義久, 佐々木 光 |
|
title: Robustness acquisition for applying Coach-player type Multi-Agent Reinforcement Learning to real-world environments
abstract: Various MARL (Multi-Agent Reinforcement Learning) frameworks have been studied as a method to operate multiple robots autonomously. In this study, we focus on the Coach-player framework which enhances task accomplishment efficiency by utilizing strategies that are the unified will of the group. Although the effectiveness of this framework has been demonstrated in an ideal simulation environment, there are some issues that need to be addressed when applying it to a real environment. In this study, we examine the introduction of adversarial agents for robustness in order to apply Coach-player to real-world environments. This will improve the robustness against various disturbances in the real environment and is expected to be applicable to the real environment. As a preliminary experiment, we examine a setting in which there is a delay in strategy distribution from the coach and confirmed the vulnerability of the Coach-player framework. language of the presentation: Japanese | |||
| 久下 柾 | M, 1回目発表 | ヒューマンロボティクス | 和田 隆広, 池田 和司, 織田 泰彰, 劉 海龍 |
|
title: Does eHMI make it easier for other road users to predict behavior of Level 3 autonomous vehicles during RtI ?
abstract: Currently, driver assistance and automated driving technologies for autonomous vehicles are advancing rapidly. When a car equipped with a Level 3 automated driving system (Level 3 automated vehicle) drives automatically, the driver is released from the driving task and can enjoy non-driving related activities (NDRA). On the other hand, during the process of taking over driving from the Level 3 automated driving system to the driver, the behavior of the autonomous vehicle becomes erratic and unstable, which may increase the risk that surrounding vehicles may be involved in accidents. We propose a method of presenting the operating status of the automated driving system and the driver's driving take-over status after request to intervene (RtI) is issued external human-machine interface (eHMI). We will verify the effectiveness of the proposed method by conducting driving simulator experiments on drivers of manually operated vehicles driving around a Level 3 automated vehicle. language of the presentation: Japanese 発表題目: Level 3⾃動運転⾞の運転引き継ぎ時における周囲の道路利⽤者に対する情報伝達とeHMIの活⽤ 発表概要: 現在，自動車の運転支援技術，自動運転技術が急速に進んでいる．レベル 3 自動運転システムの搭載車両（レベル 3 自動運転車）が自動運転で走行する際には，ドライバは運転タスクから解放されて，非運転タスク（Non-Driving Related Activities: NDRA）を楽しむことできる．一方で，レベル 3 自動運転システムからドライバに対して運転を引き継ぐ過程では，自動運転車の挙動が不規則かつ不安定になり，周囲の自動車が事故に巻き込まれるリスクが高まる恐れがある．この問題に対して，本研究ではレベル 3 自動運転車に外向けヒューマンマシンインタフェース (eHMI) を搭載することで，自動運転システムの作動状態や運転引継ぎ要請が発行された後のドライバの運転引継ぎ状況などを外部に提示する手法を提案する．レベル 3 自動運転車の周囲を走行する手動運転車のドライバに関するドライビングシミュレータ実験を行うことで，提案手法の有効性を検証する． | |||