L-VITeX is an AI heuristic based on FOMO- TinyML, specifically developed to efficiently prioritize regions of interest during terrain exploration conducted by lightweight UGVs and UAVs. It aims to facilitate the identification of target objects by leveraging a lightweight heuristic approach, thereby avoiding the substantial computational burden associated with conventional heavy object detection algorithms. This enables a more focused allocation of resources toward target objects, as opposed to a uniform distribution of effort. It operates with a minimal memory requirement of 200 KB of RAM while achieving a classification accuracy of 91.6%, making it particularly well-suited for deployment in secondary small explorers and micro-swarm systems for terrain exploration.
Recent advancements in Gaussian Splatting (GS) have demonstrated the potential to revolutionize 3D reconstruction and virtual testbeds for digital twins, significantly outperforming traditional approaches such as photogrammetry and Neural Radiance Fields (NeRFs). To explore its capabilities, I conducted an experiment using GS to reconstruct terrains from two aerial videos. The first video, a 10-second clip, captured the picturesque landscape surrounding Blüemlisalphütte in Switzerland, while the second featured a prominent location at my university known as "Low CG."
From these brief video clips, I generated preliminary Gaussian splats and utilized them to reconstruct the terrain within the Gazebo simulation environment. Additionally, I implemented Region of Interest (RoI) detection using a simulated drone integrated with Robot Operating System (ROS) in Gazebo. This experiment highlights the efficiency and versatility of GS in creating realistic and functional 3D models for applications in robotics and virtual environments.
3D Mapping using a Surveyor (SRV): This approach leverages advanced surveying techniques to generate accurate three-dimensional representations of environments. It serves as a foundational method for applications in geospatial analysis, virtual simulations, and autonomous navigation systems, ensuring precise environmental modeling and enhanced operational efficiency.