Path integration is a common navigation method used by a wide variety of animals and various models with varying degrees of biological plausibility have been proposed. In this paper we investigated the implementation of a particular path integration model from theoretical biology based on a Modified Continuous Time Recurrent Neural Network on a mobile robot and evaluated how closely the resulting behaviour matched its biological counterpart. We tested the system in several experiments where the robot had to navigate between two locations without the use of landmarks. The results show the accuracy and also the limitations of the method.