Luca Bartolomei

Luca Bartolomei

Robotics & Computer Vision Engineer

Publications

You can also find my articles on my Google Scholar profile.

Theses

Towards Robust Active Planning for Autonomous Aerial Navigation

Published in ETH Doctoral Examination, 2023

This thesis addresses the fundamental challenges of autonomous aerial navigation in GPS-denied and obstacle-dense environments across two core pillars: decentralized multi-robot coordination and semantic perception-aware active planning. By integrating high-level semantic scene understanding and deep reinforcement learning (RL) into path planning, the proposed frameworks improve localization reliability, mission execution speed, and robot safety in real-world scenarios.

Download Paper | Download Bibtex

Journal Articles

Fast multi-UAV decentralized exploration of forests

Published in IEEE Robotics and Automation Letters (RA-L), 2023

Addressing the challenges of high obstacle density and long-distance wireless communication limits in UAV fleets, the proposed multi-robot exploration framework achieves complete coverage of complex forest environments up to 30% faster than state-of-the-art baselines. Video available here.

Download Paper | Download Bibtex

Conference Papers

Voxfield: Non-Projective Signed Distance Fields for Online Planning and 3D Reconstruction

Published in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022

Outperforming state-of-the-art methods in map coverage, accuracy, and processing speed, Voxfield operates effectively onboard compute-constrained micro-aerial vehicles (MAVs), generating accurate ESDF maps to enable real-time trajectory planning and obstacle avoidance. Video available here.

Download Paper | Download Bibtex

Autonomous Emergency Landing for Multicopters using Deep Reinforcement Learning

Published in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022

The paper presents an autonomous emergency landing pipeline for multicopters that handles critical failures, such as mechanical malfunctions, GPS jamming, or sudden battery drops. By leveraging semantic and depth representations, our policy achieves direct zero-shot transfer to the real world without requiring any additional fine-tuning. Video available here.

Download Paper | Download Bibtex

Semantic-aware Active Perception for UAVs using Deep Reinforcement Learning

Published in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021

The paper presents a semantic-aware path-planning pipeline for UAVs that uses deep reinforcement learning to evaluate the perceptual informativeness of visible semantic classes. By training on semantic cues across non-photorealistic, randomly generated scenes, the deep policy network successfully generalizes to unseen environments while guiding the UAV away from perceptually degraded regions. Video available here.

Download Paper | Download Bibtex

Multi-robot Coordination with Agent-Server Architecture for Autonomous Navigation in Partially Unknown Environments

Published in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020

A server-agent system architecture for multi-robot autonomous navigation in partially unknown environments is presented, where computationally demanding tasks like global mapping and path planning are outsourced to a central server, while agents handle local pose estimation and obstacle avoidance onboard. Video available here.

Download Paper | Download Bibtex

Perception-aware Path Planning for UAVs using Semantic Segmentation

Published in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020

This paper presents a perception-aware path-planning pipeline for UAVs that leverages semantic segmentation to avoid texture-less or problematic areas, such as bodies of water, that degrade monocular VIO state estimation, actively steering the robot toward perceptually informative regions. Video available here.

Download Paper | Download Bibtex

Workshop Papers

Active Learning for UAV-based Semantic Mapping

Published in Workshop on Informative Path Planning and Active Sampling, Robotics: Science and Systems (RSS), 2019

Addressing the labor-intensive bottleneck of training data collection in vision-based UAV applications, our novelty-aware path planner is validated on real-world terrain mapping datasets, yielding substantially smaller, more informative training datasets compared to standard lawnmower pattern collection.

Download Paper | Download Bibtex