Aurelio Noca

LINC is a DARPA-funded program that aims to develop machine learning-based introspection technologies allowing robotic systems to respond to unplanned perturbations at test time. The current phase of the program targets technologies to assist crane operators during ship-to-ship transfers of supplies. The project focuses on creating robust perception and control methods to handle external perturbations. These include challenges like wind and dynamic sea conditions that affect payload delivery. I develop visual perception pipelines integrating object tracking, semantic segmentation, and 3D modeling for robust pose estimation of payloads and targets.

RACER is a DARPA-funded program that aims to develop and demonstrate new technologies enabling Unmanned Ground Vehicles (UGVs) to navigate on unstructured off-road terrain at human-level speeds. My work involved developing real-time semantic perception pipelines, addressing challenges of deploying machine learning models in diverse and data-scarce settings, highlighted by our work "Few-shot Semantic Learning for Robust Multi-Biome 3D Semantic Mapping in Off-Road Environments." by Atha et al.. My contributions superseded these efforts with notable improvements in the semi-supervised setting, as detailed in our publication "COARSE: Collaborative Pseudo-Labeling with Coarse Real Labels for Off-Road Semantic Segmentation." by Noca et al.

EPFL Xplore is a student-led robotics association with the goal of designing and developing a Mars rover from scratch. The association was founded in 2020 and has since grown to over +100 members and competed in 4 editions of the European Rover Challenge. I was a communication software engineer for the 2021 edition and a computer vision engineer for the 2023 edition.