HKU-led team launches Embodied Playground, 410-task benchmark for model physical intelligence
YiMaTweets · x · 2026-10-22
A team from HKU, Tsinghua, Stanford, UNC and robotics company Sharpa released Embodied Playground, a long-horizon agentic benchmark testing foundation models on physical, strategic, and spatial reasoning, with paper and open-source code available.
The setup: a foundation-model agent reads a goal and scene, writes control code, and acts through a robot — with each action changing the state, forming a complete closed-loop evaluation.
- 410 simulated task environments, 100 curated tasks, 5 evaluation suites, 16 task types
- Physical: predicting outcomes of contact, mass, friction, and motion (launches, toppling, fracture)
- Strategic: maintaining legal intermediate states and revising plans as ammo and positions change
- Spatial: tracking poses and cube faces, aligning interlocking parts, grasp and placement within reachable workspace
The team says more models and tasks will be continuously added.
Related event: HKU-led Team Releases Embodied Playground Benchmark(2 posts)→
More from Embodied
- "Mad that this is a $10,000 car": 2nd-gen BYD Seagull stuns observers — alexmacgregor__ · 2026-10-22
- miHoYo scales embodied agents in virtual worlds: 120 configs map limits of continued pre-training — THUSI-Lab · 2026-10-22
- AFM scans compare Apple's AR coatings: iPhone Duo vs iPad nano-texture glass — jwt0625 · 2026-10-22
- Robotic bricklaying systems reportedly lay 1,000+ bricks per hour vs 300-500 a day for humans — 2C_ornot2C · 2026-10-22
- AgenticRoboRL lifts a 9B robot planner from 18% to 71% on LIBERO-PRO — UCF-Ai-LLMs · 2026-10-22
- Berkeley and NVIDIA's AW-0 couples code, video, and action streams for robot manipulation — Zhuoyang Liu · 2026-10-22