We are a small research group working at the frontier of reinforcement learning, world models, and robotics. Our work appears at venues like ICML, TMLR, RLC, ICLR, AAAI, etc.
Three threads run through everything we do — sample-efficient learning, world models we can trust, and robots that fail gracefully when reality drifts.
Sample-efficient RL, offline RL, model-based control, and uncertainty-aware policy learning for real robots.
Learned simulators, 3D / 4D Gaussian splatting representations, and predictive models that plan inside their own imagination.
Architectures and learning pipelines for robots that have to act safely under partial information and shifting dynamics.
We share our work the way we wish other groups would share with us.
Peer-reviewed papers at venues like ICML, TMLR, RLC, ICLR, AAAI, etc.
Code, datasets, and reproducible experiments accompany every paper we can release.
Long-form notes and writeups that explain the methods we use — shipped as plain web pages, not paywalled PDFs.
