Utilizing Redundancy in Modular Robots to Achieve Greater Accuracy

Modular robot configurations typically feature many kinematically redundant loops. We believe that the information in these loops can be utilized in order to increase the accuracy of localization, in particular, at an end effector. We hope it will be possible to design an algorithm that can design configurations able to manipulate to a user specified level of accuracy. In order to do so, it would be necessary to predict the level of accuracy of a configuration from a priori information. In this work we provide experimental evidence that increased accuracy is easily achieved through redundancy. We then formulate a representation of accuracy as a distribution of location across space. We use Bayesian statistics to prototype three different models of the experimental system and test their ability to predict the increase in accuracy observed. We find that all three of the statistics prototyped were conservative estimators, leading us to the conclusion that our kinematic model of the system was too constrained.