Predicting away robot control latency

Alexander Gloye, Mark Simon, Anna Egorova,
Fabian Wiesel, Oliver Tenchio, Michael Schreiber,
Sven Behnke, and Raśl Rojas

This paper describes a method to reduce the effects of the system immanent delay when tracking and controlling fast moving robots using a fixed video camera as sensor. The robots are driven by a computer with access to the video signal. The paper explains how we cope with system latency by predicting the movement of our robots using linear models and neural networks. We use past positions and orientations of the robot for the prediction, as well as the most recent commands sent. The setting used for our experiments is the same used in the small-size league of the RoboCup competition. We have successfully field-tested our predictors at several RoboCup events with our FU-Fighters team. Our results show that path prediction can significantly improve speed and accuracy of robotic play.