Reproduce “Which Objects Stay? Learning Hierarchical Persistence Beliefs for Lifelong Open-Vocabulary Object Maps” 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

Thanh Nguyen Canh, Haolan Zhang, Nak Young Chong

HiPer estimates object persistence in lifelong maps by retrieving histories of visually and geometrically similar objects, then updating each object’s belief from observations and relocations. Across 120 held-out simulated THOR-Lifelong sequences, it achieved negative log-likelihood of 0.265, versus 0.341 for PF(λ), with lower scores in 116 sequences. Real 3RScan prediction favored HiPer, but sparse histories did not establish learning benefit.

Summary by alphaXiv.

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