Reproduce “Self-Evolving Just-In-Time Memory for Proactive Embodied Safety” 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
Bingrui Sima, Lizhong Wang, Xiaoya Lu, Kun He, Xiao Yang
The authors present a memory-based robot safety method that checks remembered states against imminent actions and offers mitigation when a hazard arises. Its working memory tracks objects and relations, while safety rules trigger procedural guidance. On 150 simulated IS-Bench episodes, Qwen3-VL-8B reaches 48.7% safe success versus 18.4% for Vanilla, without lower task success; this is benchmark evidence, not physical-world validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.