Omnidirectional humanoid fall avoidance strategies and decision volume -

Abstract

Our lives are becoming more and more dependent on different types of robots, such as cars, industrial robots, medical robots. Humanoids on the other hand, can easily be integrated into a human environment and perform some basic chores. For example, ASIMO robot developed by Honda can serve coffee on a tray and push a cart in an ordinary office. To accomplish a task in a dynamic environment such as an office, humanoids must be equipped with a fall avoidance protocol. This allows them to handle unexpected disturbances. The contributions of this thesis are twofold. First I introduce what I call a decision volume, which is used to determine the states from which the robot can recover. The decision volume is an extension of what is previously known in the literature as the decision surface---limited to perturbations and states in the sagittal plane, to handle disturbances emanating from random directions in the azimuth plane. The second contribution is the extension of the ankle and hip strategies for humanoid fall avoidance to handle disturbances in random directions. The robot is physically modeled as a 3D Linear Inverted Pendulum Model (LIPM) for the ankle strategy, and for the hip strategy a sphere is added. Using these models, the decision volumes are calculated analytically. After that to test their validity, both strategies are implemented using Virtual Model Control (VMC) in simulations and in the experiments on the real robot. The results show that the decision volume was successful in determining which strategy is applicable. For robot's states that are within the decision volume of the strategy the robot succeeds in keeping its balance, whereas for states that are outside the decision volume the robot falls down.