Kimera-Multi: Robust, Distributed, Dense Metric-Semantic SLAM for Multi-Robot Systems

Multi-robot simultaneous localization and mapping (SLAM) is a crucial capability to obtain timely situational awareness over large areas. Real-world applications demand multi-robot SLAM systems be robust perceptual aliasing operate under limited communication bandwidth; moreover, it desirable for these capture semantic information enable high-level decision-making spatial artificial intelligence. This article presents $ \mathsf{{Kimera-Multi}} $ , system that: 1) capable of identifying rejecting incorrect inter- intrarobot loop closures resulting from aliasing; 2) fully distributed only relies on local (peer-to-peer) achieve mapping; 3) builds globally consistent metric-semantic 3-D mesh model the environment in real time, where faces are annotated with labels. implemented by team robots equipped visual-inertial sensors. Each robot trajectory estimate using \mathsf{{Kimera}} . When available, initiate place recognition pose graph optimization protocol based graduated nonconvexity algorithm. The proposed allows improve their estimates leveraging inter-robot while being outliers. Finally, each uses its improved correct deformation techniques. We demonstrate photo-realistic simulations, benchmarking datasets, challenging outdoor datasets collected ground robots. Both simulated experiments involve long trajectories (e.g., up 800 m per robot). show that : outperforms state art terms robustness accuracy; achieves estimation errors comparable centralized distributed; parsimonious 4) produces accurate meshes; 5) modular can also used standard reconstruction (i.e., without labels) or reconstructing mesh).