Reproduce “Learning Intrinsic Dynamics in Latent World Models” 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

Jiamin Shi, Yuanhong Zhang, Zhiyang Liu, Jiawei Sun, Jingwen Fu

Concat IDM adds inverse-dynamics supervision to a visual world model, predicting the action from consecutive latent states so the shared encoder retains action-dependent distinctions. On Push-T with 24 training-palette appearances and current and goal observations sharing an appearance, it reaches 77.2 ± 0.2% closed-loop success versus 9.1 ± 2.7% for Plain JEPA; this does not establish transfer to unseen appearances.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.