MapForest: A Modular Field Robotics System for Forest Mapping and Invasive Species Localization

A team of researchers has introduced MapForest, a novel modular robotics system designed to solve the critical environmental challenge of mapping forests and locating invasive tree species. The system's core innovation is its platform-agnostic sensing payload, which can be rapidly mounted on a UAV, bicycle, or backpack, making it adaptable for diverse terrains from dense forests to urban parks. To overcome the significant hurdle of degraded GPS signals under forest canopies, the team enhanced a standard LiDAR-inertial mapping system with sophisticated 'covariance-aware GNSS factors' and robust loss kernels. This allows MapForest to maintain high positional accuracy, demonstrated by a trajectory error of only 1.95 meters over a challenging 1.2-kilometer forest traversal.

The software pipeline integrates this precise mapping with computer vision. The researchers trained a custom object detection model to identify a specific invasive species, the Tree-of-Heaven (Ailanthus altissima), from RGB camera imagery captured by the moving platform. Detections are then fused with the reconstructed 3D map to generate georeferenced, GIS-ready outputs that directly inform conservation and eradication efforts. In evaluations across six sites, the Tree-of-Heaven detector achieved an F1 score of 0.653, a solid benchmark for real-world, cluttered environments. The team is supporting reproducible research by releasing their datasets and associated tooling, providing a valuable foundation for future work in ecological robotics and automated environmental monitoring.