自動的な知識の抽出に基づく動物行動学に基づくソーシャルロボットの行動モデルの獲得

基于自动知识抽取的行为学获取社交机器人行为模型

• 批准号: 19F19380
• 负责人: 新妻 実保子
• 金额: $ 1.34万
• 依托单位: Chuo University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for JSPS Fellows
• 财政年份: 2019
• 资助国家: 日本
• 起止时间: 2019-11-08 至 2022-03-31
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-19F19380/
• 关键词: reinforcement learning; Human-Robot Interaction; multi-agent systems; rule-base reduction; fuzzy rule interpolation; knowledge extraction; fuzzy control; antecedent redundancy; fuzzy rule clustering

In this period the possibility of using multiple agents in the Fuzzy Rule Interpolation-based Reinforcement Learning (FRI-RL) and running them distributed in parallel was investigated. As the FRI-RL knowledge extraction method is inherently sequential, some sub-results can be different in the parallel version, but still providing a sufficient solution. This way the knowledge extraction can be performed much faster, therefore using the method on problems with a high dimension count becomes practical. Also, a possible bridging interface between the behaviour simulation model (Strange Situation Test (SST) realized with an FRI-based fuzzy automaton) and real physical robots have been partly designed and implemented. Experimenting with real physical robots is underway. Furthermore a suitable indoor localization system was constructed and adapted to the needs of the planned Human-Robot Interaction (HRI) scenario. This system is able to easily calibrate the indoor localization system’s virtual coordinate system to the real-world physical coordinate system, which makes our planned HRI experiments possible with real humans and mobile robots. A customized robot behaviour engine and a motion control system was developed to support the proposed artificial Strange Situation Test experiments.

在此期间，研究了在基于模糊规则插值的强化学习（FRI-RL）中使用多个代理并并行运行它们的可能性。由于FRI-RL知识提取方法本质上是顺序的，一些子结果在并行版本中可能不同，但仍然提供了足够的解决方案。通过这种方式，可以更快地执行知识提取，因此在高维数的问题上使用该方法变得实用。此外，一个可能的桥接接口之间的行为仿真模型（奇怪的情况下测试（SST）实现了基于FRI的模糊自动机）和真实的物理机器人已部分设计和实施。对真实的物理机器人的实验正在进行中。此外，一个合适的室内定位系统的构建和适应计划的人机交互（HRI）场景的需要。该系统能够很容易地校准室内定位系统的虚拟坐标系的真实世界的物理坐标系，这使得我们计划的HRI实验可能与真实的人类和移动的机器人。一个定制的机器人行为引擎和运动控制系统的开发，以支持拟议的人工奇怪的情况下测试实验。

项目成果

Towards the automatic observation and evaluation of ethologically inspired Human-Robot Interaction

走向受行为学启发的人机交互的自动观察和评估

• 发表时间: 2021
• 作者: D. Vincze;M. Gacsi;S. Kovacs;P. Korondi;A. Miklosi and M. Niitsuma
• 通讯作者: A. Miklosi and M. Niitsuma

Demonstration of expert knowledge injection in Fuzzy Rule Interpolation based Q-learning

基于模糊规则插值的Q学习中专家知识注入的演示

• 发表时间: 2021
• 作者: Tompa;T.;Kovacs;S.;Vincze;D.;Niitsuma;M.
• 通讯作者: M.

Towards the automatic observation and coding of simple behaviours in ethological experiments

走向行为学实验中简单行为的自动观察和编码

• 发表时间: 2021
• 作者: Vincze;D.;Gacsi;M.;Kovacs;S.;Niitsuma;M.;Korondi;P.;Miklosi;A.
• 通讯作者: A.

Football Simulation Modeling with Fuzzy Rule Interpolation-based Fuzzy Automaton

基于模糊规则插值的模糊自动机足球模拟建模

• 发表时间: 2020
• 作者: Vincze;D.;Toth;A.;Niitsuma;M.
• 通讯作者: M.

Antecedent Redundancy Exploitation in Fuzzy Rule Interpolation-based Reinforcement Learning

• DOI: 10.1109/aim43001.2020.9158875
• 发表时间: 2020-07
• 期刊: 2020 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM)
• 作者: Dávid Vincze;Alex Tóth;M. Niitsuma