Balancing and grasping from visual feedback for an unstable wheeled humanoid

Abstract

Krang is a Wheeled Inverted Pendulum Humanoid, designed to accomplish strenuous tasks quicker, and with more strength, than the average human being. Weighing over 300lbs, Krang sits on a differential drive platform balancing on two wheels in an inverted pendulum configuration. The platform forms the first joint in a 17 degree-of-freedom upper body that possesses a waist, torso and two 7 degree-of-freedom arms. Through a whole body control scheme, this unique design allows Krang to manipulate its center of mass to locomote quickly on a plane, while the redundancy of joints enables second order tasks to be completed, such as carrying a tray of water or utilizing its weight torque to lift and move heavy objects. However, while Krang is very capable, it remains unaware of the environment in which it works.This research project aims to introduce localization and state estimation capabilities to Krang by giving it the ability to measure and analyze its surroundings. Currently, Krang must be positioned by humans before running experiments involving locomotion and end effector manipulation, making the robot blind to variations in its environment, and vulnerable to potentially poor state estimation of the first link. By attaching a vision system to the robots spine, this thesis project aims to introduce positional tracking and spatial mapping capabilities, which can act as a redundancy for stabilization of the robot, and give Krang a level of autonomy that requires less human oversight. In addition, a visual servoing formulation allows the robot to identify and pick up objects on its own.