Robot Learning Paper Club - The World Model edition.

Robot learning in Zurich is scattered. Academia, big tech, and startups are all pushing the field forward, but rarely in the same room. If Zurich is going to lead in robotics globally, the people building it need to talk to each other. This is a space for exactly that. We mix 10-minute presentations from researchers in the local community presenting their own work and participants presenting recent papers they thought were interesting (5 minutes). Spots are limited. Contributing participants and people who have presented in the past have preferred access. — World model edition: This edition focuses on learned world models for robotics and how they connect to policy learning. A non-exhaustive list of what we think is interesting: Action-conditioned video world models and latent dynamics models (pixel space vs. latent space prediction) Joint world-action models that predict future observations and actions together World models as policy evaluators and simulators: closed-loop evaluation, planning, and RL inside the model Synthetic data generation and augmentation with world models, and how to measure whether their physics hold up for manipulation Want to present a paper but not sure which one? Feel free to reach out to us at dominique@dream-machines.eu and we can help you find one!