Communication as a Spatial Sensor for Multi-robot Localization and Mapping

Abstract

With the maturity of single-robot systems and the advancement of autonomous mobile robots, there is a growing potential for deploying multi-robot systems in various applications. In this context, every machine can be considered as a robot, constituting a multi-robot system that enhances efficiency and capability. This thesis presents a novel approach to multi-robot coordination and perception by utilizing wireless communication signals as a spatial sensing modality. By extracting angle-of-arrival (AOA) information from wireless signals exchanged between robots, this research enables the estimate of relative positions, which is the key enabler for multi-robot coordination with off-the-shelf robotic platforms. The work begins by developing communication as a novel sensing modality for computing AOA multipath profiles using robots’ local mobility in 3D space. It characterizes the impact of the virtual antenna array geometry, formed by the robot’s trajectory, on the accuracy of the estimated AOA. The Cramér-Rao Bound is employed to analytically quantify the effect of various trajectories—such as 3D helical, 2D circular, and 2D linear motions—on AOA estimation performance. The concept of “informativeness” is introduced to compare different trajectory shapes, identifying 3D motion as more informative than 2D for AOA estimation. Furthermore, the influence of robot localization error on AOA accuracy is characterized by deriving mathematical bounds on the AOA error based on the statistics of the trajectory estimation error. The framework is generalized to handle the simultaneous mobility of both transmitting and receiving robots. Next, the AOA sensing is integrated into a multi-robot SLAM pipeline by identifying loop closure candidates in environments with perceptual aliasing. By leveraging geometric information from wireless signals, the algorithm efficiently and robustly identifies poses that are in proximity. The proposed method addresses the multipath phenomenon by identifying the direct signal path using geometric consistency maximization over the robots’ pose graphs. The algorithm is validated through various challenging environments with ambiguous geometric features. Building on these foundations, the thesis then explores active sensing strategies in a closed loop. Here, robots actively coordinate their motions based on AOA feedback to bring them into proximity, thereby enhancing pose graph optimization. Beyond SLAM, this work advances collaborative photorealistic 3D reconstruction by incorporating localization uncertainty derived from wireless sensing into a Neural Radiance Fields (NeRF) framework for active multi-robot 3D reconstruction. Extensive simulations and real-world experiments validate the approach, demonstrating significant improvements in multi-robot localization accuracy, loop closure detection efficiency, and detailed scene reconstructions. The developed techniques enable effective multi-robot coordination in perceptually degraded environments, advancing the state-of-the-art in multi-robot cooperation for complex real-world applications. The contributions of this thesis lay the groundwork for more efficient and robust multi-robot systems that can operate autonomously in diverse and challenging scenarios.