Reproduce “Embodied Playground: A Long-Horizon Agentic Robotics Benchmark for Physical, Strategic, and Spatial Reasoning” 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

Feng Chen, Pei Zhou, Lixuan He, Yunchao Yao, Li Sun, Shenghua Gao, Mingyu Ding, Kaifeng Zhang and 2 more

Embodied Playground evaluates whether foundation-model agents can complete robot-task episodes by observing permitted views, running control code, and revising actions as scenes change. Its protocol covers 100 tasks across five suites and independently checks goals, rule compliance, and physical validity. On EP-100, Pass@2 was 19.94% for GPT-6-Astra and 35.85% for Opus-5.5-extra, under differently documented observation access.

Summary by alphaXiv.

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