Resource-Aware Collaborative Monte Carlo Localization with Distribution Compression

TL;DR

This paper tackles collaborative global localization under computational and communication constraints by introducing a distribution-compression-based approach (Compress++) to efficiently exchange beliefs between robots. It extends Monte Carlo Localization with reciprocal sampling and evaluates multiple baselines (DET, clustering, thinning) to analyze trade-offs in compression, bandwidth, and fusion costs. The authors provide a thorough complexity analysis and empirical validation on simulated and real-world setups, demonstrating improved convergence, reduced bandwidth, and real-time onboard performance. The open-source C++/ROS2 implementation and comprehensive evaluation offer a practical pathway toward scalable, resource-aware multi-robot localization.

Abstract

Global localization is essential in enabling robot autonomy, and collaborative localization is key for multi-robot systems. In this paper, we address the task of collaborative global localization under computational and communication constraints. We propose a method which reduces the amount of information exchanged and the computational cost. We also analyze, implement and open-source seminal approaches, which we believe to be a valuable contribution to the community. We exploit techniques for distribution compression in near-linear time, with error guarantees. We evaluate our approach and the implemented baselines on multiple challenging scenarios, simulated and real-world. Our approach can run online on an onboard computer. We release an open-source C++/ROS2 implementation of our approach, as well as the baselines

Figures (7)

• Figure 1: Top: two robots during one experimental run when robot $A$ detects robot $B$. Bottom: localization of robot $B$ in the same run using 3 methods; prediction, color-coded for time, is plotted against ground truth position (black). Two failures, using non-collaborative MCL and Prorok et al. prorok2012iros, and a successful convergence after 20s with our collaborative localization approach (Compress++).
• Figure 2: (a) An illustration of an experimental run up to the first detection event. (b) A run where robot $B$ is has no particles around its truth position at the time of detection (left), followed by reciprocal sampling (center) and successful localization (right).
• Figure 3: The robotic platform used in the evaluation.
• Figure 4: Environments have varying degree of geometric symmetry and feature richness. The area highlighted in red was reconstructed in our lab for real-world evaluation. The three maps are to scale; the LiDAR range, 12 m, is marked in blue.
• Figure 5: The success rate of all methods for each of the environments for robot $B$.