Mechanical Design of a Simple Bipedal Robot

The thesis objective was to design a walking robot appropriate for testing a machine learning control system. As a robot that would learn how to walk, it was pre-maturely named Toddler. The passive dynamic walker, a class of simple robots driven by gravity and inertia, was selected as the model for the Toddler robot. Two preliminary passive walkers were explored: a three-dimensional, toy-like biped with one degree of freedom at the hip; and a two-dimensional, four-legged kneed walker with degrees of freedom at the hip and knees. Bench level and computer modeled analyses were conducted during the design and construction of these robots. The simpler, toy-like walker was selected as the unactuated model for actuated Toddler. Two degrees of freedom, controlling step size and frequency, were introduced at the ankle. Toddler was able to walk with a dynamically-stable gait either passively down a ramp, or actively on a fat or slightly uphill surface. With its simple design and robust performance, Toddler provides an excellent platform for testing machine learning control in walking robotics. Thesis Supervisor: H. Sebastian Seung Title: Professor, Department of Brain and Cognitive Sciences and Department of Physics