基于学习人类策略的动态稳定系统控制器切换方法研究

项目名称： 基于学习人类策略的动态稳定系统控制器切换方法研究

项目编号： No.61273335

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 欧勇盛

作者单位： 中国科学院深圳先进技术研究院

项目金额： 79万元

中文摘要： 现有的传统控制方法对控制器切换过程中可能遇到的问题研究非常有限。然而，许多系统，特别是动态稳定系统，相当多的失稳都是发生在控制器切换过程中。例如，双足仿人机器人抬脚上楼梯时可能会翻倒等。人类和动物能够轻松的完成复杂的动态的运动控制，这种能力对于当前的机器人而言却存在很大的困难。本项目试图利用基于学习人类策略的控制方法，探讨动态稳定系统的控制器切换问题。从控制理论的角度，将人类智能与控制结合起来，研究在基于学习的这类系统控制器切换的理论与方法。采用基于学习方法的优点是可以避免对复杂和精确的动力学模型的依赖。这一问题的难点和关键问题是如何对建立高精度的模型以及非线性系统收敛域的估计方法。该项目的成功实施，将对双足仿人机器人，四足仿生机器人，两轮直立式代步车，自主单轮机器人等具有动态稳定性的机械系统控制性能的改善产生积极影响。

中文关键词： 学习人类策略；动态稳定系统；控制器切换；机器人；智能控制

英文摘要： So far, research works on the problems involving controller switching are still very limited. However, for most moving systems, especially dynamically stable systems, instability occurs during the control target switching. For example, a humanoid robot may fall down to the ground when it lifts its leg to go upstairs. While human beings and animals can easily perform complex and challenging actions and control their bodies very well, such behaviors are far beyond the capabilities of modern robots. In this project, we attempt to use learning-by-human-demonstration approach to address the controller switching problem for dynamically stable systems. From the perspective of control theory, by combining human intelligence and control systems, we investigate the theory and techniques for learning-based control switching of such systems. One of the advantages of adopting a learning approach is that it can avoid or reduce the dependency on sophisticated and accurate dynamic models. The main challenges lie in the development of a mechanism to effectively build up highly precise leaning models, and an estimation method for the convergence regions in such closed-loop nonlinear systems. The successful completion of this project will have significant and positive impact on improving the control performance of many dynamically

英文关键词： Learning from demonstration；Dynamically stable system；Controller switching；Robotics；Intelligent control