← Back ICRA 2026

A Distributed Gaussian Process Model for Multi-Robot Mapping

Seth Nabarro, Mark van der Wilk, Andrew J Davison

PDF

AI summary

Key figure (auto-extracted from paper)

DistGP enables accurate, scalable multi-robot mapping by replacing restrictive tree topologies with a loopy Gaussian process model trained via distributed belief propagation.

Distributed learning Gaussian processes multi-robot mapping belief propagation collaborative robotics

Problem

Centralized computation is infeasible for large-scale multi-robot mapping in dynamic or poorly connected environments, while existing distributed methods either suffer from accuracy limitations due to rigid connectivity constraints or lack scalability and local consistency.

Approach

The authors introduce DistGP, a sparse Gaussian process model that mirrors the multi-robot network structure and uses Gaussian belief propagation to synchronize local maps asynchronously without requiring a fixed communication topology.

Key results

Relaxes tree-structured connectivity to eliminate prediction discontinuities and improve accuracy
Achieves centralized batch-training performance through distributed, asynchronous message passing
Outperforms distributed neural networks in accuracy, robustness to sparse communication, and continual learning
Enables online, on-the-fly model construction with dynamic inter-robot connectivity

Why it matters

Provides a scalable and robust foundation for collaborative mapping in resource-constrained multi-robot applications like environmental monitoring and space exploration.

Abstract

We propose DistGP: a multi-robot learning method for collaborative learning of a global function using only local experience and computation. We utilise a sparse Gaussian process (GP) model with a factorisation that mirrors the multi-robot structure of the task, and admits distributed training via Gaussian belief propagation (GBP). Our loopy model outperforms Tree-Structured GPs [1] and can be trained online and in settings with dynamic connectivity. We show that such distributed, asynchronous training can reach the same performance as a centralised, batch-trained model, albeit with slower convergence. Last, we compare to DiNNO [2], a distributed neural network (NN) optimiser, and find DistGP achieves superior accuracy, is more robust to sparse communi- cation and is better able to learn continually.

Index terms

Multi-Robot Systems Distributed Robot Systems Mapping