Hi, I'm Christian Gumbsch. I research how robots can use world models and foundation models to imagine the future, judge the outcomes of their actions, and act on that.I'm currently 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 the University of Tübingen, jointly supervised by Georg Martius and Martin Butz, followed by a postdoc at TU Dresden working with Katharina von Kriegstein.
Research interest
Inspired by how humans learn and make decisions, my research aims to build safe, generalist embodied agents that think before they act. To get there, I work on three building blocks:
1. Imagine: world models. Agents that predict possible outcomes of their actions using structured and hierarchical representations of how the world works, e.g., THICK, GateL0RD.
2. Verify: vision-language reward models. Agents that use vision-language models to judge whether outcomes, imagined or observed, meet a goal, e.g., SENSEI, Demo2Reward.
3. Act: planning and policies. Agents that act through both hierarchical, temporally abstract planning and reactive policies such as VLAs, e.g., L0RD CAPRI, Hierarchical RL perspective.
News
Showing selected papers. Toggle to view the full list.
