Stephan Liwicki

Toshiba (Japan), Toshiba (United Kingdom)

Papers

Total Citations

H-Index

About

Stephan Liwicki is a leading researcher at the intersection of embodied AI, deep reinforcement learning, and visual navigation. His work focuses on bridging the gap between simulated training and real-world robotic performance, tackling fundamental challenges in sample efficiency and domain adaptation. Liwicki’s most influential contribution is **NavACL**, a novel method for automatic curriculum learning in visual navigation. By intelligently selecting training tasks based on geometric features, NavACL allows deep reinforcement learning agents to dramatically outperform state-of-the-art approaches in real-world environments—a breakthrough that has garnered over 40 citations since its 2021 publication. More recently, Liwicki introduced **ReCoRe** (Regularized Contrastive Representation Learning of World Models), a framework designed to overcome the poor sample efficiency and appearance-variation sensitivity that plague model-free RL methods in everyday tasks. This work addresses a critical bottleneck: while RL agents excel in gaming environments, they struggle with visual navigation under real-world conditions. Liwicki’s research is paving the way toward more robust, adaptable autonomous systems that can learn efficiently and generalize across unpredictable visual domains.