Sebastian Curi

École Polytechnique Fédérale de Lausanne, ETH Zurich

Papers

Total Citations

H-Index

About

Sebastian Curi’s research lies at the intersection of reinforcement learning (RL), robotics, and safe decision-making under uncertainty. His major contributions focus on developing algorithms that enable autonomous agents to learn and act reliably in high-stakes environments where exploration is costly or dangerous. In his highly cited work on **risk-averse offline reinforcement learning** (2021, 15 citations), Curi pioneered methods for training RL agents using only pre-collected, safe datasets—eliminating the need for risky online exploration. This approach is critical for applications like healthcare, autonomous driving, and industrial control, where failures during learning are unacceptable. His work on **gradient-based trajectory optimization with learned dynamics** (2023, 7 citations) further advances robotics by combining data-driven models with classical optimization, allowing robots to plan movements even when accurate physics models are unavailable. Curi’s research has been recognized for bridging theory and practice, earning him a reputation for tackling fundamental challenges in safe and sample-efficient learning. His papers are essential reading for researchers working on offline RL, risk-aware control, and real-world robot deployment, demonstrating how rigorous algorithmic design can unlock autonomy in safety-critical domains.