Criticality of Adaptive Control Dynamics
Criticality of Adaptive Control Dynamics
复制标题
DOI：
10.1103/physrevlett.107.238103
复制
发表时间：
2011-12-02
影响因子：
8.6
通讯作者：
Pawelzik, Klaus
中科院分区：
文献类型：
--
作者：
Patzelt, Felix;Pawelzik, Klaus
我们证明，动力系统的镇定可以消除关于其结构的可观测信息。该机制在局部自适应控制中引入临界点作为吸引子。它还揭示，以前报道的简单控制器的临界是由适应引起的，而不是由其他控制器细节引起的。我们将这些结果应用于一个真实系统的例子：人类平衡行为。介绍了一种受现实约束的预测自适应闭环控制模型，该模型以前所未有的详细程度再现了实验观测结果。我们的结果表明，在Levy和Gauss域之间观察到的误差分布可能反映了消除随机局部趋势和罕见的大误差之间的近乎最佳的折衷。
We show, that stabilization of a dynamical system can annihilate observable information about its structure. This mechanism induces critical points as attractors in locally adaptive control. It also reveals, that previously reported criticality in simple controllers is caused by adaptation and not by other controller details. We apply these results to a real-system example: human balancing behavior. A model of predictive adaptive closed-loop control subject to some realistic constraints is introduced and shown to reproduce experimental observations in unprecedented detail. Our results suggests, that observed error distributions in between the Levy and Gaussian regimes may reflect a nearly optimal compromise between the elimination of random local trends and rare large errors.