Arthur Guez

Papers

Total Citations

H-Index

About

Arthur Guez is a researcher whose work bridges the frontiers of robotics, reinforcement learning, and biologically-inspired AI. His early research tackled the foundational problem of simultaneous localization and mapping (SLAM) in robotics, specifically addressing the challenge of "multi-tasking SLAM" (2010, 12 citations), where a robot must perform localization and mapping while simultaneously pursuing other objectives—a critical step toward more versatile autonomous systems. Guez has also made significant contributions to improving learning efficiency by exploring how symmetries and invariances can augment machine learning in biologically-inspired domains (2019, 6 citations), offering a principled way to reduce problem dimensionality and accelerate training. Most recently, his work on "Beyond Tabula-Rasa" (2020, 2 citations) introduced a modular reinforcement learning approach for physically embedded 3D Sokoban, demonstrating how robots can achieve abstract goals by integrating visual, abstract, and physical reasoning—moving beyond tabula rasa deep RL. Though his citation counts are modest, Guez's research consistently tackles the hardest integration problems in embodied AI, making his work notable for its ambition in combining perception, control, and abstract reasoning in physically grounded systems.