Pascal Poupart

Papers

Total Citations

252

H-Index

About

Pascal Poupart is a prominent machine learning and artificial intelligence researcher whose work spans sequential decision-making, probabilistic reasoning, and safe reinforcement learning. He is best known for his foundational contributions to Partially Observable Markov Decision Processes (POMDPs), particularly his landmark 2006 paper "Point-Based Value Iteration for Continuous POMDPs," which has accumulated 246 citations and remains a cornerstone reference in the field. This work was groundbreaking in extending POMDP optimization beyond discrete state spaces into continuous domains — a critical advancement for real-world applications such as robotics and autonomous navigation, where states, actions, and observations rarely fit neatly into discrete categories. More recently, Poupart has turned his attention to the intersection of inverse reinforcement learning (IRL) and safe AI, exploring how autonomous agents can infer not only reward functions but also underlying behavioral constraints from expert demonstrations. His 2022 papers on constraint inference and benchmarking reflect a growing commitment to making reinforcement learning agents reliable and deployable in physical systems where safety guarantees matter. Together, his body of work demonstrates a sustained effort to bridge theoretical rigor with practical applicability, making him a valuable figure for researchers working at the frontier of autonomous systems and probabilistic AI.