Hi, I'm Christian Gumbsch. I'm a postdoc at the University of Amsterdam working with Stratis Gavves. Previously, I did my PhD at the Max Planck Institute for Intelligent Systems and University of Tübingen jointly supervised by Georg Martius and Martin Butz, followed by a postdoc at the TU Dresden working with Katharina von Kriegstein.
Research interest
I work toward embodied artificial intelligence that helps us tackle general problems while remaining safe and ethical. Inspired by how humans and other animals adaptively learn and act, my research bridges cognitive science, model-based reinforcement learning, and foundation models for decision-making, and is organized along three directions:
1. Compositional world models. Agents that imagine possible futures by composing reusable, structured knowledge of how the world works. Representative work: GateL0RD (NeurIPS 2021), THICK world models (ICLR 2024).
2. Semantic reward models. Agents that verify whether imagined outcomes satisfy user-defined goals, using language-based foundation models. Representative work: SENSEI (ICML 2025), Demo2Reward (RLC 2026).
3. Cognitive-inspired decision-making. Agents that act through both hierarchical, temporally abstract planning and fast, reactive policies, drawing on how humans make decisions. Representative work: Hierarchical RL perspective (Nature Machine Intelligence, 2022), HierarchicalGateL0RD (ICDL 2022).
Showing selected papers. Toggle to view the full list. See Google Scholar for recent publications.