Identification of Humanoid Biped Robot During Walking with Linear Regression Models

The purpose of the paper is to investigate walking dynamics of a biped humanoid robot with a multivariable linear time invariant structure. The assumption is that during low speed walking the angular velocity in a joint from the construction will vary proportionally to small deviations in the walking parameters. First a basic walking algorithm is implemented on a developed laboratory model of a humanoid robot with 17 degrees of freedom (DOF). Then a set of experimental data is recorded which is used for the identification procedure. After initial identification with a multivariable ARX a final state-space model from 24th order is estimated which is with 10 inputs and 6 outputs. The model is validated with an independent set of data and can be used for the purpose of closed-loop control.