Humanoid Online Locomotion Generation based on Markov Chain with Microsoft Robotics Studio

The aim of this work is to generate and synchronize biped walking online with Microsoft Robotics Studio simulation platform. By defining three action primitives as Swing, Lift and Move, pose nodes with different geometric configuration and dynamical stability profile would be created for Markov chain. Using ZMP displacement as cost function and Maximum Likelihood Estimation Algorithm, Markov chain cut off redundant similar pose nodes. After training, only ε Independent Pose Group will be used in online gait generation system. The full-scale humanoid robot Robo-Erectus Senior can achieve flexible and dynamically stable walking online using the proposed algorithm. Only successful gaits qualified by Microsoft Robotics Studio will be applied on Robo-Erectus for practical locomotion. Action primitives make it easy to stabilize walking and changing step length and velocity smoothly.