论文部分内容阅读
一个稳定且灵活机动的双足步行往往是仿人机器人能否顺利并快速完成任务的关键,为了让机器人使用机器学习方法获得这样的步态,在自适应协方差矩阵进化算法的基础上,提出了累积分层优化的步行学习方法.在步行学习过程中,将学习任务由简单到复杂地划分为几个关键层,各层的适应度函数依据该层的学习目标来设计,每一层的学习都在前一层的结果基础上完成,从而让每层的学习结果逐步累积,最终达到既定的学习目标.根据该优化方法,让机器人首先学习稳定步行,然后在稳定的基础上完成机动步行的学习,实验结果证明了此优化策略能够获得有效的双足步行.
A stable and flexible two-legged walking is often the key to the success of the humanoid robot. In order to make the robot use the machine learning method to get such a gait, based on the adaptive covariance matrix evolutionary algorithm, In the walk learning process, the learning task is divided into several key layers from simple to complex, and the fitness function of each layer is designed according to the learning objectives of the layer, and each layer of Learning is done based on the results of the previous level, so as to gradually accumulate the learning results of each level and finally reach the established learning goal.According to the optimization method, the robot first learns stable walking, and then completes the motorized walking on the basis of stability The experimental results show that this optimization strategy can obtain effective biped walking.