Reproduce “JEPA4R: Embodied JEPA World Models with 4D Point Planning” 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
Jiarui Dong, Zhiheng Liang, Yanjun Li, Zeyu Zhang, Heejeong Nam, Haoyu Chen, Hao Tang
JEPA4R gives a robot a single depth camera’s point cloud and predicts future scene states in latent space, where a policy selects actions. A separate decoder renders predicted latents as point clouds for inspection, not planning. On OGBench Cube-Double, its policy succeeds at 41.6%, versus 3.6% for pixel-based PLDM. The comparison does not isolate modality or establish a general planning advantage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.