Abstract:
In this thesis, we proposed a signal-based omni-directional bipedal walking algorithm which will be used in humanoid robot soccer domain. We generate a central periodic signal which is used to synchronize other signals in the system. In order to model the system more clearly, we divide the main motion into four di erent components and each component is represented with a signal. While modeling the locomotion system, we tried keeping it as much parametric as possible. Hence, it is possible to change the characteristics of the motions. In addition to the implementation of the bipedal walking algorithm, an optimization algorithm, Evolutionary Strategies is used to nd the optimal parameters to the locomotion system. The aim of the system constitutes the tness function of the optimization algorithm: reach the destination point as quick as possible. As a result of this work, we could achieve a parametric omni-directional bipedal walking algorithm on Aldebaran Nao Humanoid robots and optimize the parameters in both simulation and real world environments with di erent population sizes. Although a signi cant improvement is achieved for the simulation environment, it works as a netuning process in real world experiment. In addition to the di erences for simulation and real world environment, we analyzed the e ects of population size on the training procedure. Training with bigger population sizes makes a deeper search and although its average tness value increases slowly, overall best tness value increases rapidly.
