A framework for collaborative multi-robot mapping using spectral graph wavelets

Lukas Bernreiter, Shehryar Khattak, Lionel Ott, Roland Siegwart, Marco Hutter, César Cadena

Introduction

In multi-robot missions, such as search and rescue or subterranean exploration, robots must navigate unknown environments while maintaining accurate maps. Typically, robots send data to a centralized server that performs global optimization and loop closure. However, a significant challenge remains: providing real-time feedback to the robots. Without corrections from the server, individual robots continue to operate with drifted odometry, leading to inconsistent local paths and navigation failures.

This paper introduces a novel feedback mechanism that utilizes spectral graph wavelets to identify and correct odometry drift in real-time. By analyzing structural differences between a robot's local estimate and the server's optimized global map at multiple spatial scales, the system can pinpoint error origins and transmit sparse, high-impact constraints back to the robots, improving onboard localization accuracy by up to 90%.

Methods

The proposed framework treats a robot's trajectory as a pose graph where nodes represent poses and edges represent spatial relationships. The core of the approach lies in Spectral Graph Theory.

System architecture overview showing the interaction between multiple robots and a centralized mapping server.
  1. Graph Signal Processing: A signal is defined on the pose graph based on $SE(3)$ distances, capturing both position and orientation.
  2. Spectral Graph Wavelets: The graph signal is transformed into the spectral domain using the Graph Laplacian. Wavelets are applied at different scales: small scales detect local jitter, while large scales identify structural, long-term drift.
  3. Kron Reduction: To minimize bandwidth, the server uses "Kron reduction" to create a sparse abstraction of the global graph that preserves its essential structural properties.
  4. Constraint Generation: The robot compares its local graph wavelets to the server’s reduced "fingerprints." Differences guide the automatic generation of new constraints (adjacent, n-hop, or submap-level) for the robot's local optimizer (like iSAM2).

Experimental Evaluation

The framework was validated across diverse datasets and real-world deployments:

  • EuRoC Machine Hall: Standard visual-inertial odometry benchmarks using aerial drones.
  • Indoor/Outdoor ANYmal: Legged robot data featuring transitions designed to induce localization failure.
  • Hagerbach Underground Tunnel: A 1.2 km autonomous subterranean mission.
  • DARPA Subterranean (SubT) Challenge: Data from Team CERBERUS’s prize-winning run in complex environments.

Key Insights

  • Root Cause Identification: Multi-scale wavelets allow the system to identify where drift originated, enabling effective corrections with minimal constraints.
  • Modality Independence: By operating on abstract pose graphs rather than raw sensor data, the system supports heterogeneous robot teams (e.g., drones and legged robots).
  • The Role of $SE(3)$: Using full $SE(3)$ geometry for the graph signal is significantly more effective than simple Euclidean distance ($R^3$), as catching rotation errors early prevents massive downstream positional drift.
  • Bandwidth Efficiency: Kron reduction can reduce graph size by over 60% while maintaining high correction accuracy.

Results

The proposed method demonstrated significant improvements in localization accuracy. In the EuRoC datasets, it achieved the lowest collaborative error, often outperforming state-of-the-art centralized SLAM systems.

Quantitative results on the EuRoC Machine Hall dataset comparing the proposed method against state-of-the-art SLAM frameworks.

In the "Rooftop" experiment, where LiDAR-based localization failed, server feedback recovered the trajectory, reducing RMSE from 2.22m to 0.30m. During the DARPA SubT Final, the system corrected drift that would have caused a scoring failure for a detected artifact, bringing its reported position back within the required 5m threshold.

Conclusion

This framework provides a robust, bandwidth-efficient method for real-time collaborative mapping. Future work will focus on reducing the computational overhead of local graph optimization for long-range missions and exploring hierarchical graph structures to further optimize the trade-off between map reduction and accuracy.

Summary updated

Abstract

The exploration of large-scale unknown environments can benefit from the deployment of multiple robots for collaborative mapping. Each robot explores a section of the environment and communicates onboard pose estimates and maps to a central server to build an optimized global multi-robot map. Naturally, inconsistencies can arise between onboard and server estimates due to onboard odometry drift, failures, or degeneracies. The mapping server can correct and overcome such failure cases using computationally expensive operations such as inter-robot loop closure detection and multi-modal mapping. However, the individual robots do not benefit from the collaborative map if the mapping server provides no feedback. Although server updates from the multi-robot map can greatly alleviate the robotic mission strategically, most existing work lacks them, due to their associated computational and bandwidth-related costs. Motivated by this challenge, this paper proposes a novel collaborative mapping framework that enables global mapping consistency among robots and the mapping server. In particular, we propose graph spectral analysis, at different spatial scales, to detect structural differences between robot and server graphs, and to generate necessary constraints for the individual robot pose graphs. Our approach specifically finds the nodes that correspond to the drift's origin rather than the nodes where the error becomes too large. We thoroughly analyze and validate our proposed framework using several real-world multi-robot field deployments where we show improvements of the onboard system up to 90\% and can recover the onboard estimation from localization failures and even from the degeneracies within its estimation.
BibTeX
@article{bernreiter2022framework,
  author = {Lukas Bernreiter and Shehryar Khattak and Lionel Ott and Roland Siegwart and Marco Hutter and César Cadena},
  title = {{A framework for collaborative multi-robot mapping using spectral graph wavelets}},
  journal = {Int. J. Robotics Res.},
  year = {2022},
}
Figure