Reproduce “Evaluating Long-Horizon Robot Planning under Uncertainty” 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
Harsh Muriki, Gayathri Aravindan, Yuchen Zheng, Sonia Chernova
The authors compare ReAct, VLM-TAMP, and Tru-POMDP on seven household tasks in a Habitat benchmark, varying initial memory and environment conditions across 1,350 episodes. ReAct achieved the highest task success (45.3%) and completion (71.7%), but its action-based SPL was 36.9%. This metric counts high-level skill attempts relative to a human reference, not runtime; these findings apply to the tested adaptations and setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.