Multi-Robot Cooperative 3D Frontier Exploration Using Stereo Camera Based Mapping in GPS-Denied Environments

Autonomous 3D exploration with lightweight uncrewed aerial vehicles is especially chal- lenging when relying solely on stereo vision in GPS-denied environments. This paper proposes a decentralized multi unmanned aerial vehicle (UAV) exploration framework that integrates Real-Time Appearance-Based Visual SLAM (RTAB-Map) with an entropy-filtered frontier detection pipeline operating directly on a volumetric OctoMap. Frontier voxels are refined using Shannon entropy and clustered with Density-Based Spatial Clustering of Applications with Noise (DBSCAN), while a ray-casting-based Next-Best-View utility function selects in- formative, low-cost goals without centralized coordination. Rapidly-Exploring Random Tree Star (RRT*) planner with OctoMap collision checks generates feasible paths, and a Nonlinear Model Predictive Controller (NMPC) ensures stable trajectory tracking using filtered visual odometry. The Extended Kalman Filter is used for the fusion of visual odometry and IMU data. This framework enables scalable vision-based cooperative exploration without GPS or external infrastructure.