The design of virtual reality simulators, and more specifically those dedicated to surgery training, implies to take into account numerous constraints so that simulators look realistic to trainees and train proper skills for surgical procedures. Among those constraints, the accuracy of the biophysical models remains a very hot topic since parameter estimation and experimental validation often rely on invasive protocols that are obviously not suited for living beings. In the context of Interventional Radiology the procedures involve the navigation of surgical catheter tools inside the vascular network where many contacts, sliding and friction phenomena occur. The simulation of these procedures require complex interaction models between the tools and the blood vessels for which there is no ground truth data available for parametrization and validation. This paper introduces an experimental testbed to address this issue: acquisition devices as well as a data-processing algorithms are used to record the motion of interventional radiology tools in a silicon phantom representing a vascular network. Accuracy and high acquisition rates are the key features of this testbed as it enables to capture dynamic friction of non-smooth dynamics and because it could provide extensive data to improve the accuracy of the mechanical model of the tools and the interaction model between the tools and the blood vessel.