This paper analyzes algorithms and sensor fusion architectures used to mechanize a self-contained, pose-estimation for the feet of humans or humanoid robots. The approaches makes use of a network of low-cost, inertial measurement units (IMUs) affixed to the feet. By leveraging known equality and inequality constraints between the motion and location of the IMUs, drift due to inertial sensor output errors are reduced or eliminated. Two sensor fusion approaches are evaluated; a de-centralized estimator and centralized estimator. Experimental results demonstrating the performance of these fusion schemes are presented. Issues associated with tuning the de-centralized and centralized estimators are discussed in detail.