Software and Neural Architectures for Real-Time Adaptive Learning and Control

用于实时自适应学习和控制的软件和神经架构

• 批准号: 9412430
• 负责人: James Peterson
• 金额: $ 22.28万
• 依托单位: Clemson University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1994
• 资助国家: 美国
• 起止时间: 1994-08-15 至 1998-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9412430&HistoricalAwards=false
• 关键词: Software; Neural; Architectures; Real; Time

9412430 Peterson This project will develop the software and neurobiological connections that are important to making progress in the areas of real-time learning and control. The research will focus on methods using the Pontryagin optimality and Hamilton-Jacob-Bellman (HJB) equation in conjunction with on-line ( or real-time) measurements of state data to build local models of control law. These techniques share the spirit of the adaptive-critic methodologies for construction of models of the control law, but are tightly bound to the premise of real-time problems. The focus here is on pertinent refinements and extensions of these techniques in three major arenas: theoretical analysis, locally scoped artificial neural architecture and object oriented software architecture design. General software code using C++ language will be developed to support these research efforts. The graphical interface will be designed using X windows with the Motif widget set. The software environment will then be tested on problem domains of interest, e.g. reduced order model of Large Space Structures (LSS) and flexible robotic arms.

9412430彼得森该项目将开发对在实时学习和控制领域取得进展至关重要的软件和神经生物学连接。研究的重点将集中在使用庞特里亚金最优性和哈密尔顿-雅各布-贝尔曼(HJB)方程结合在线(或实时)状态数据测量来建立控制律的局部模型的方法。这些技术与构造控制律模型的自适应批评方法的精神相同，但与实时问题的前提紧密相关。这里的重点是这些技术在三个主要领域的相关改进和扩展：理论分析、局部范围的人工神经体系结构和面向对象的软件体系结构设计。将开发使用C++语言的通用软件代码来支持这些研究工作。图形界面将使用X窗口和Motif小部件集进行设计。然后，软件环境将在感兴趣的问题领域进行测试，例如大型空间结构的降阶模型和灵活的机械臂。