This research will develop robust AI models for unstructured environments using Self-Supervised Learning (SSL) and Foundation Models.
The objective is to resolve the ””label bottleneck”” to enable autonomous UGV navigation, precision forest inventory, and crop yield prediction without extensive manual annotation.
The project addresses socio-economic challenges in primary production. By adapting foundation models to rural complexities, the research will support sustainable resource management. It aims to improve occupational safety by deploying autonomous systems in GPS-denied or hazardous areas, reducing human risk in remote zones.
The methodology utilizes SSL to pre-train on unlabeled datasets, developing specialized decoders atop shared backbones. I will implement transformer architectures for multi-modal fusion. Data sources include multi-resolution satellite time-series, UAV LiDAR, and proximal sensor streams from robotic platforms.
