Reproduce “Serving Agentic Robot Fleets: Workload Characterization, KV-Cache Thrashing, and Admission Control” 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

Sida He

The authors measure how brain serving and action deadlines constrain a simulated robot fleet replay. They replay robot request patterns from 80 RPent episodes using a language-model proxy, while profiling the action policy separately. At 32 robots, limiting active episodes to 16 raises completed requests and trace-program throughput, but this does not demonstrate live fleet performance.

Summary by alphaXiv.

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