Poly-TagSLAM: Low-Cost Robot Localisation and Mapping in Polytunnels with Fiducial Markers

Agricultural robots operating in polytunnel environments face unique localisation and mapping challenges due to unreliable GNSS signals, repetitive structures, and vegetation motions. This paper presents Poly-TagSLAM, a robust and cost-effective mapping-then-localisation framework designed for agricultural robots operating in challenging polytunnel environments. In the mapping stage, a reliable LiDAR and webcams are combined to generate an accurate map of the pre-installed fiducial markers using the factor graph formulation and GTSAM toolbox. Subsequently, in the localisation stage, robots equipped with webcams, sparse-beam LiDARs or even just wheel odometry can reliably localise themselves within the Tag map. Extensive experimental evaluations demonstrate that our Poly-TagSLAM is capable of reaching reliable mapping and localisation accuracy even with affordable sensors, showcasing the practicality of Poly-TagSLAM in agricultural robotics applications.