机构地区: 上海交通大学机械与动力工程学院机器人研究所
出 处: 《上海交通大学学报》 2006年第7期1166-1169,共4页
摘 要: 提出了一种基于神经网络学习的机器人动力学建模方法.其特点是在网络结构中引入积分单元,构成含有积分回路的动态网络,使之能够很好地学习对象的动态特征.讨论了该模型在最优化等领域中的应用,及其泛化能力等问题.依据实际高尔夫挥杆机器人的结构参数以及其控制器特性进行了仿真实验.仿真结果表明,该模型算法简单实用,完全不需要对象的数学描述,泛化能力极强,具有良好的应用前景. A method of dynamics modeling based on ANN (artificial neural network) learning was proposed. Its characteristic is adding integral units into the ANN structure to build a recurrent network with integral feedback. The model can learn dynamic objects commendably. Generalization ability of the learning dynamics model was discussed, as well as its application in optimization. Simulated experiments were performed, which are based on parameters of a real golf-swing robot and its controller. The effectiveness of the method was confirmed by the results.
关 键 词: 机器人 人工神经网络 动力学模型 最优化 泛化能力
领 域: [自动化与计算机技术] [自动化与计算机技术]