Reproduce “Plannable Approximations to MDP Homomorphisms: Equivariance under Actions” 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
Elise van der Pol, Thomas Kipf, Frans A. Oliehoek, Max Welling
The authors learn an abstract model that preserves action effects and rewards, then use value iteration to plan and transfer action values to image-based tasks. Their method combines latent transition and reward prediction with contrastive negative samples. In the double-object test-goal task, it reaches the goal in 8.38 ± 0.07 steps versus 38.25 ± 15.30 for a goal-conditioned baseline; results cover tested deterministic tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.