Evolutionary approach for developing fast and stable offline humanoid walk

Abstract

To make stable and fast walking in humanoid robot possible, lots of works has to be done including attitude estimation, dynamic stability controls, path planning and online position tracking. These processes need an exact mechanic modelling of the robot's body structure. In this paper, an evolutionary approach is used to model the human walk for a standard humanoid robot named NAO. This method makes use of inverse kinematics for trajectory generations. Additionally a novel method is proposed for mapping from gait design to genetic algorithm, in order to optimize generated trajectory parameters. The proposed method leads us to forward velocity of 64 cm/s, which is a superior result compared to the other common methods.

DeepWalk: Omnidirectional Bipedal Gait by Deep Reinforcement Learning

“…Learning approaches applied to walking controllers have been predominantly investigated for gait optimization [6]- [8]. These methods, however, depend on engineered components such as trajectory planners, central pattern generators and robot dynamics models.…”

Stability control for biped walking based on phase modification during double support period

“…For example, Wang et al [10] propose a novel scheme of the ankle and hip balance control strategies. Farazi et al [17] present an evolutionary approach to model the human walk which use of inverse kinematics for trajectory generations. But NAO is a robot whose weight is relative light compared with common humanoid robot, such as BIT's robot BHR-5.…”