Self-Organization of Modules and Their Hierarchy in Robot Learning Problems: A Dynamical Systems App
Self-Organization of Modules and Their Hierarchy in Robot Learning Problems: A Dynamical Systems App
复制标题
机器人学习问题中模块的自组织及其层次结构：动态系统应用程序
DOI：
--
复制
发表时间：
1997
期刊：
影响因子：
--
通讯作者：
S. Nolfi
中科院分区：
文献类型：
--
作者：
J. Tani;S. Nolfi
本文描述了如何在感觉运动系统的神经网络架构中，以模块化和层次化的方式自组织世界的内部表示。我们开发了一个在线学习方案（所谓的递归神经网络（RNN）专家的混合物），其中一组RNN模块在多个级别中自组织为专家，以考虑机器人经历的不同类别的感觉运动。该方案通过涉及导航学习问题的仿真实验进行了研究，其中配备了距离传感器的机器人在不同形状的房间内行走。结果表明，在较低层次网络中，与拐角处左右转弯、沿沿着走廊直行和遇到交叉口相对应的代表性建筑块或“概念”在其各自的模块中自组织。在高层网络中，通过将低层网络中得到的"概念”组合成序列，自组织出不同房间中旅行所对应的"概念”。机器人成功地学会了通过递归交互在多个层次上感知世界。在Sony CSL逗留期间
This paper describes how the internal representation of the world can be selforganized in modular and hierarchical ways in a neural network architecture for sensory-motor systems. We develop an on-line learning scheme { the so-called mixture of recurrent neural net (RNN) experts { in which a set of RNN modules becomes self-organized as experts in multiple levels in order to account for the di erent categories of sensory-motor ow which the robot experiences. The proposed scheme was examined through simulation experiments involving the navigation learning problem, in which a robot equipped with range sensors traveled around rooms of di erent shape. It was shown that representative building blocks or \concepts" corresponding to turning right and left at corners, going straight along corridors and encountering junctions are self-organized in their respective modules in the lower level network. In the higher level network, the \concepts" corresponding to traveling in di erent rooms are self-organized by combining the \concepts" obtained in the lower level into sequences. The robot succeeded in learning to perceive the world as articulated at multiple levels through its recursive interactions. During stay at Sony CSL