Luca Marzari
Papers
10
Total Citations
88
H-Index
5
About
Luca Marzari is a robotics and artificial intelligence researcher whose work sits at the intersection of deep reinforcement learning (DRL), safety verification, and autonomous robotic systems. His research addresses one of the most pressing challenges in modern robotics: ensuring that learning-based systems behave reliably and safely when deployed in real-world environments. Marzari's early contributions explored hierarchical task decomposition for robotic manipulation, demonstrating how DRL could be structured to improve sample efficiency for complex pick-and-place tasks (31 citations). He has since become a leading voice in safe DRL, pioneering frameworks that integrate formal verification techniques with reinforcement learning to identify and reduce policy violations in robotic navigation systems (18 citations). His work on curriculum learning further advanced safe mapless navigation by structuring training progressively to minimize unsafe behaviors (13 citations). A distinguishing theme across his portfolio is the development of online safety property collection and refinement methods, enabling agents to learn safer policies without excessive exposure to dangerous states. He has also extended these principles to medical robotics, applying constrained reinforcement learning to autonomous colonoscopy navigation. With over 80 cumulative citations and a growing body of work bridging neural network verification and practical robotics, Marzari represents an important emerging voice in trustworthy autonomous systems research.
Research Focus
Key Achievements
Top Papers
- 1
- 2Verifying Learning-Based Robotic Navigation Systems18 citations · 2023
- 3Curriculum learning for safe mapless navigation13 citations · 2022
- 4
- 5
- 6Verifying Learning-Based Robotic Navigation Systems4 citations · 2022
- 7
- 8
- 9Safe Deep Reinforcement Learning by Verifying Task-Level Properties2 citations · 2023
- 10Verifying Online Safety Properties for Safe Deep Reinforcement Learning1 citations · 2025