Precise object-relative positioning for car-like robots

Abstract

In this paper, we present a method for increasing the object-relative positioning precision of a car-like robot. In an outdoor environment and on an electric vehicle retrofitted with commercially available actuators, we achieve sub-centimeter and sub-degree accuracy, longitudinally and in heading, respectively. Most noteworthy, this is higher precision than RTK-GPS - which is often assumed ground truth - can realistically measure. In order to get to this level of precision, the whole control chain, i.e., sensing, planning, high- and low-level control, needs to be tightly tuned. However, the biggest contribution to the final precision and hence the contribution of this paper, is the usage of object-relative coordinates for the final approach. The proposed algorithms were extensively tested using an autonomous docking maneuver at multiple sites. For the docking to be physically possible, errors of less than 1cm (longitudinally) and 1 degree (in heading) were necessary and reliably achieved.