Reproduce “JEPLO: Joint-Embedding Predictive Learning for LiDAR-Based Legged Locomotion” 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
Qihao Yuan, Yixuan Qiu, Ziyu Cao, Ming Cao, Kailai Li
JEPLO turns incomplete LiDAR scans into predictive terrain representations for locomotion without mapping. It predicts proprioceptive and LiDAR-depth embeddings rather than reconstructing observations, using encoders and recurrent memory. In simulation, on a 0.5-meter box under Cross occlusion, it achieved 50.2 ± 9.6%, versus WMP’s 30.8 ± 4.7%. This supports relative robustness under tested conditions, not immunity to severe occlusion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.