Algorithms for Complex Bipedal Walking

Abstract

This thesis studies the problem of parameter estimation, model identi- fication and state estimation for underactuated bipedal walking robots. Two main results were developed. The first result is a novel identification method suited for this problem. The second result is the extension of existing algorithms for state estimation to the case of the hybrid model of an underactuated walking robot. The identification method takes advantage of the linear structure of the model with respect to estimated parameters. The resulting estimator is calculated iteratively and maximizes the likelihood of the data. The method was tested on both simulated and experimental data. Simulations were carried out for an underactuated walking robot with a distance meter to measure absolute orientation. Laboratory experiments were carried out on a leg of a laboratory walking robot model equipped with a threeaxis accelerometer and gyroscope to measure absolute orientation. The method performs favorably in comparison with other benchmark estimation algorithms and both the simulations and the laboratory experiments con rmed its high potential for the use in identi cation of underactuated robotic walkers. The state estimators were applied to estimate the absolute orientation of an underactuated walking robot in the presence of impacts which occur when the leg of the robot hits the ground. The proposed estimation scheme was tested on simulations of a 3-link robot and shows that proposed extensions yields improved estimation performance.