A Neuro Fuzzy-Based Gait Trajectory Generator for a Biped Robot Using Kinect Data

Generating gait trajectories is an important step in biped motion control. So, in this paper, a neuro fuzzy (ANFIS)-based gait trajectory generator for a normal walking of a biped robot during the single support phase is developed. Since human sensorimotor controls are done in an optimal way following the principle of optimality, gait trajectories data of human subjects of different hip heights, and accordingly different steps, are captured by Kinect sensor and then used for training the developed ANFIS-based gait generator. In this context, for each subject, the hip and swing limb ankle trajectories are captured, filtered and averaged over steps. The averaged trajectories are approximated using least square fitting with polynomial functions whose coefficients are used as ANFIS outputs and hip heights as inputs. Additionally, constraint equations are obtained for those trajectories and compared with hypothetical constraints in literature to prove that the latter constraints are not consistent with the naturally obtained ones. Furthermore, the averaged trajectories are generalized in terms of the step length and maximum step elevation. The results prove the effectiveness of the approach to generate gait trajectories, which are optimal, for a biped robot.