Reproduce “UniOMA: Unified Optimal-Transport Multi-Modal Structural Alignment for Robot Perception” 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

Xinrui Zu, Mark Hoogendoorn, Kevin Sebastian Luck, Shujian Yu

UniOMA augments contrastive learning with a shared, geometry-aware structural target to align relationships among sensor observations, not only paired samples. It builds a consensus from each modality’s input-space similarities and trains embeddings to match it alongside contrastive loss. On held-out VFD and VFP evaluations, it improves structural measures and cross-modal transfer over relational controls; on one 6-DoF manipulator, reaching-position error is lower.

Summary by alphaXiv.

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