Reproduce “Flow Equivariant World Models: Structured Memory for Dynamic Environments” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hansen Jin Lillemark, Benhao Huang, Fangneng Zhan, Yilun Du, Thomas Anderson Keller
FloWM uses spatial latent map memory to predict future views from partial observations. It updates the map for agent actions and uses motion-specific velocity channels to move stored features when objects leave view. In controlled 3D Dynamic Block World tests, its 210-frame mean squared error was 0.001539, versus 0.009614 without velocity channels and 0.021684 for DFoT. These results do not establish exact equivariance or open-world performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.