Memory Based Sensorimotor Learning by Using Data Compression Technique

使用数据压缩技术进行基于记忆的感觉运动学习

• 批准号: 10650392
• 负责人: MIYAZAKI Fumio
• 金额: $ 2.5万
• 依托单位: Osaka University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1998
• 资助国家: 日本
• 起止时间: 1998 至 1999
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-10650392/
• 关键词: perception; behavior; intelligent robots; mappings; locally weighted regression; k-dimensional trees; environmental change; LWR(Locally Weigted Regression); マッピング; 学習; 状態予測; K-D tree; 視覚サーボ系; オプティカルフロー

Remarkable features of biological systems are the robustness and adaptability. Human movement is highly resilient to environmental perturbations. This suggests the presence of a learned sensorimotor mapping that compares visual, proprioceptive and motor signals to make open-loop adjustments.The aim of this research project is to propose a new method of learning sensorimotor mappings that combines the KD-trees (k-dimensional trees) learning and LWR (locally weighted regression) learning. Key features of the proposed method are the ability to learn in "one shot" from an instance, to interpolate and blend across instances with sensitivity to local sampling density, to resist noise, and to learn a mapping piecewise (thus saving memory). The proposed method can be used to learn many kinds of sensorimotor mappings, such as stereoscopic hand-eye mapping for a robot that plays ping-pong and a sensorimotor mapping for navigating a mobile robot. The details of these examples are involved in this report.

生物系统的显著特征是鲁棒性和适应性。人类的运动对环境扰动具有很强的适应能力。这意味着存在一个学习的感觉运动映射，该映射比较视觉、本体感觉和运动信号以进行开环调节。本研究的目的是提出一种结合KD-树(k维树)学习和局部加权回归(LWR)学习的新的感觉运动映射学习方法。该方法的主要特点是能够从实例中“一次”学习，对局部采样密度敏感地跨实例进行内插和混合，抵抗噪声，以及分段学习映射(从而节省内存)。该方法可用于学习多种类型的感觉运动映射，如用于乒乓球机器人的立体手眼映射和用于移动机器人导航的感觉运动映射。本报告涉及这些例子的详细情况。

项目成果

T. Mitsuda, F. Miyazaki, et. al.: ""Visual Servoing Based on Coarse Optical Flow""Proc. of 14th IFAC World Congress, Beijing. B. 53-58 (1999)

T. Mitsuda、F. Miyazaki 等。

富士川和延,小荒健吾,宮崎文夫,丸典明: "キャリブレーションエラーのあるステレオカメラにおける逆透視変換不可能面とビジュアルサーボの安定性"日本ロボット学会誌. 16・7. 1026-1028 (1998)

Kazunobu Fujikawa、Kengo Koara、Fumio Miyazaki、Noriaki Maru：“无法转换为反向透视的表面和具有校准误差的立体相机中视觉伺服的稳定性”日本机器人学会杂志 16・7（1998 年）。 ）

K. F. Macdorman: "Partition Nets : An Efficient On-Line Learning Algorithm"Proc. Of 9th Int. Conf. On Advanced Robotics, Tokyo. 529-535 (1999)

K. F. Macdorman：“分区网络：一种高效的在线学习算法”Proc。

K. F. MacDorman: ""Partition Nets : An Efficient on -line learning Algorithm""Proc. of 9th Iat. conf. on Advanced Robotics. Tokyo. 529-535 (1999)

K. F. MacDorman：“分区网络：一种高效的在线学习算法”Proc。

N.Oshiro,A.Nishikawa,F.Miyazaki: "Using Foveated Vision for obust Object Tracking - 3D Horopter Analysis"IEEE Press "Robust Vision for Vision-Based Control of Motion". 7 (1999)

N.Oshiro、A.Nishikawa、F.Miyazaki：“使用注视点视觉进行复杂对象跟踪 - 3D 双眼视界分析”IEEE Press“基于视觉的运动控制的鲁棒视觉”。