A Robust Relative Positioning System for Multi-Robot Formations Leveraging an Extended GM-PHD Filter

We propose a multi-robot tracking method to provide state estimates that allow a group of robots to maintain a formation even when the communication fails. We extend a Gaussian Mixture Probability Hypothesis Density filter to incorporate, firstly, absolute poses exchanged by the robots, and secondly, the geometry of the desired formation. Sensory detections, information about the formation, and communicated data are all combined in the extended Gaussian Mixture Probability Hypothesis Density filter. Our method is capable of maintaining the state estimates even when long-duration occlusions occur, and improves awareness of the situation when the communication rate is slow or sporadic. The method is evaluated using a high-fidelity simulator in scenarios with a formation of up to five robots. Experiments confirm the ability of the filter to deal with occlusions and refinement of the state estimate even when poses are exchanged at a low frequency, resulting in drastic reduction of the chance of collisions compared to a tracking-free implementation.