中文摘要：

为提高类人机器人模仿学习的准确性及效率,建立了一种改进的粒子群算法优化超限学习机的模仿学习模型。采用非线性动态系统对示教时的相关数据进行建模;以动态自适应策略改进粒子群算法的惯性权重,并利用改进后的粒子群算法对超限学习机的网络参数进行寻优;利用该耦合学习模型对模仿学习动态系统的参数进行学习,并重现了模仿学习动作。实验结果表明,该耦合算法应用在类人机器人模仿学习方面具有很好的拟合精度、自适应性及泛化能力,重现模仿学习动作时的平均误差为0.0172。

英文摘要：

To improve accuracy and efficiency in learning from demonstrations by humanoid robots,an imitation learning model was established based on improved PSO to optimize extreme learning machine,to learn human motions on the robots herein.A set of motions which was performed by a human demonstrator were collected to model as a nonlinear autonomous dynamical system.PSO was improved with the dynamic adaptive inertia weight.Then the improved PSO was merged with ELM to optimize network parameters.Using a mathematical model of improved PSO-ELM to learn the parameters of the dynamic system and reproduce human motions.The experimental results prove the method has a better fitting precision,adaptability and generalization ability on imitation learning of humanoid robots.The average relative errors are as 0.0172 of human motion reproductions.