Enhancing multi-robot map fusion in ROS 2 via dynamic trust factor weighting

Abstract Getting a team of autonomous robots to agree on a single, accurate map is harder in practice than single-robot Simultaneous Localization and Mapping (SLAM) literature might suggest. A map update that arrives at a fusion server two seconds after it was generated describes an environment state that the transmitting robot has already moved through. Treating that constraint with the same confidence as a freshly received one is the core problem of multi-robot SLAM. This paper presents the Enhanced Multi-Robot Map Fusion (EMRMF) framework, a Robot Operating System 2 (ROS 2) Humble architecture that assigns a scalar trust score θ ∈ [0,1] to every robot constraint before it enters the pose-graph optimizer. The score combines two independent signals: how geometrically consistent the observation is with the current map estimate of a robot, and how fresh the observation was when it arrived at the fusion server. This paper validated EMRMF using two evaluation environments. In Gazebo simulation, the ablation study shows that the full system achieves a per-step pose Root Mean Square Error (RMSE) of 0.149 m versus 0.359 m for baseline graph SLAM — a 58.5% reduction. On two physical robots in a 150 m 2 indoor laboratory, the cumulative session trajectory RMSE (accumulated drift over the 12-min mapping session, measured against a Vicon motion-capture ground truth) is 8.433 m for Robot 1 and 2.143 m for Robot 2 with EMRMF, compared with 8.686 m and 2.914 m for A-LOAM. Map alignment RMSE is 0.287 m. Under the harshest tested communication conditions (0.2 s delay, 30% packet loss), the Gazebo per-step pose RMSE degrades to 0.267 m, which remains below the 0.359 m ideal-condition baseline. A Wilcoxon signed-rank test across 30 paired trials returned W = 12, p < 0.001, confirming statistical significance.