Highly Accurate Positioning Method for Car-Like Robots Utilizing a Monocular Camera and QR Code Tracking

Abstract

Mobile robots often serve as a means for the transportation of goods. In most cases, this task requires high positioning accuracy in order to allow for a smooth transfer of goods from the transfer station to the robot and vice versa. This is especially difficult with car-like robots, as they lack a degree of freedom that e.g. omni-directional robots possess. In this paper, we propose a highly accurate positioning method for such car-like robots that consists of three consecutive modes. In the free navigation mode, the robot drives towards a predefined target position until it detects a transfer station. By using a single monocular camera and a trained convolutional neural network, our system detects the transfer station and approaches it automatically until it reaches a predefined distance, which marks the end of the vague positioning mode. The camera detects and tracks a QR code attached to the station, which we use to estimate the robots relative position to the code thus initiating the accurate positioning mode. In this mode, our system uses a third order polynomial path-planning approach that achieves an average positioning accuracy of 11mm longitudinally and 8mm laterally with an angular offset of 0.7° in front of the code. We show the applicability of this functionality through a set of experimental validation tests that mimic real-world use-cases.