Reproduce “SpatialMem: Building Quantitative Spatial Reasoning Agents with Hierarchical Memory” 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
Zinan Zheng, Nuo Chen, Jia Li
SpatialMem helps a vision-language agent answer quantitative spatial questions by retrieving and refining tool-use trajectories stored for indoor or outdoor scenes and three task types. In offline MetricBench evaluation, its average success rates were 61.8% indoors and 75.5% outdoors, versus 49.0% and 68.9% without memory. These results concern benchmark answers, not downstream robot performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.