Reproduce “Diffusion-Steering-Based MPC with a JEPA World Action Model” 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
Shashank Hegde, Alexander Popov, Elie Aljalbout, Nikolai Smolyanskiy
The authors introduce LeWAM, a decoder-free world-action model that uses one latent representation and a shared transformer for prediction and action generation. Its planner searches the policy’s noise inputs rather than arbitrary actions, keeping candidates within the policy’s learned repertoire. Across six simulated tasks, action-space MPC achieved 3–8% closed-loop success, while noise-space planning improved LeWAM and RecWAM; real-world performance remains untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.