Distributed Simultaneous Localisation and Auto-Calibration using Gaussian Belief Propagation (2401.15036v1)

Abstract: We present a novel scalable, fully distributed, and online method for simultaneous localisation and extrinsic calibration for multi-robot setups. Individual a priori unknown robot poses are probabilistically inferred as robots sense each other while simultaneously calibrating their sensors and markers extrinsic using Gaussian Belief Propagation. In the presented experiments, we show how our method not only yields accurate robot localisation and auto-calibration but also is able to perform under challenging circumstances such as highly noisy measurements, significant communication failures or limited communication range.

• The paper introduces a fully distributed framework that leverages GBP to perform simultaneous localisation and auto-calibration without sharing calibration parameters.
• It integrates factor graphs with an adaptive regularisation technique to achieve rapid convergence and robust performance against communication issues and sensor outliers.
• Experimental evaluations in both simulated and real-world settings validate the method’s high accuracy, scalability, and resilience in dynamic, multi-robot environments.

Introduction

In multi-robot systems, precise robot localisation and extrinsic calibration of sensors and markers are vital to ensure effective and safe robot interactions. Traditionally, research in localisation systems has primarily focused on single-robot setups, inadequate for multi-robot collaborative tasks where relative localisation determines the overall system's accuracy. Moreover, the prerequisite of accurate extrinsic calibration poses significant challenges and costs in practical, large-scale multi-robot settings.

Methodology

The paper introduces a scalable and fully distributed framework for simultaneous localisation and auto-calibration in multi-robot setups using Gaussian Belief Propagation (GBP). Each robot in the system can infer its pose and calibrate the extrinsic parameters of both sensors and markers through probabilistic inferences based on observations from other robots. Uniquely, the approach avoids the sharing of calibration parameters among robots, thus reducing communication overhead and enhancing scalability. Factor graphs represent the complex relationships between variables, and each robot incrementally builds and solves its portion of the graph. An adaptive regularisation technique within GBP is introduced to achieve convergence, addressing the non-linearity and non-convexity of the underpinning optimisation problems.

Evaluation and Results

The method is rigorously evaluated against other state-of-the-art distributed and centralised alternatives in simulated environments with varying robot count, communication constraints, and presence of outliers. The results demonstrate that the method achieves a rapid convergence rate and notable resilience to communication failures and measurement outliers. The accuracy of localisation and calibration remains comparably high, even under artificially induced sensor miscalibrations and limited communication ranges. Real-world experimental validation on the UTIAS MR.CLAM dataset further corroborates the system's practical efficacy.

Conclusion

This work presents a significant step toward self-reliant multi-robot systems capable of operating robustly in real-world situations. The proposed method stands out for its decentralised architecture that gracefully scales with the number of robots while maintaining high accuracy in localisation and calibration. This enables multi-robot collaborations even in challenging and dynamic environments with minimal manual intervention. Additionally, the system's robustness to communication and sensory perturbations positions it as an ideal candidate for further exploration and deployment in diverse multi-robot applications.