Boudewijn Bakker

University of Amsterdam, Intelligent Systems Research (United States), Dalle Molle Institute for Artificial Intelligence Research

Papers

Total Citations

239

H-Index

About

Boudewijn Bakker’s research lies at the intersection of mobile robotics, hierarchical spatial reasoning, and reinforcement learning, with a focus on enabling robots to navigate and learn efficiently in large, partially observable environments. His most influential work, “Hierarchical map building and planning based on graph partitioning” (76 citations), introduced a method to segment base-level maps into higher-level structures, dramatically improving path planning efficiency for large-scale navigation. Complementing this, his “Hierarchical dynamic programming for robot path planning” (56 citations) extended hierarchical planning to stochastic tasks using Markov decision processes, outperforming standard dynamic programming. Bakker also pioneered techniques to overcome key challenges in robot learning: his 2004 paper (56 citations) combined reinforcement learning with memory to handle partial observability and continuous domains, while “Quasi-online reinforcement learning for robots” (32 citations) enabled real-time policy training by building probabilistic models on the fly. Earlier work on unsupervised event extraction (19 citations) transformed noisy sensory streams into discrete events, making reinforcement learning tractable for mobile robots. Collectively, Bakker’s contributions have advanced autonomous navigation and learning, with over 200 total citations, establishing him as a key figure in hierarchical planning and reinforcement learning for robotics.