Research & Blogs

// RECENT PUBLISHING

01 ———Dataset

Ropedia Xperience 2: A Gold Standard for Scaling Physical AI Environments and Behaviors

Ropedia AI

Ropedia New Release · 2026

We introduce Ropedia Xperience 2, a unified physical AI dataset featuring 1,589 metric-scale 3D scans and 4,230 egocentric recordings across 74 real-world tasks. By converting real-world environments and human behaviors into reusable simulations and robot trajectories, it establishes a new gold standard for scalable physical AI data collection and robot learning.

02 ———Ropedia-A0

Ropedia-A0: Code-Only-as-Policy for Robot Control

Ropedia AI

Ropedia Super Week · 2026

We introduce Ropedia-A0, which explores Code-Only-as-Policy, a new robot control paradigm that uses a shared, reusable code library as the policy. It achieves a 70.24% success rate on RoboDojo, generalizes across embodiments, and transfers from simulation to the real world with 75% first-try success on a pick-and-place task.

03 ———DONIC

DONIC: Generalist Bimanual Manipulation Through Human Demonstrations

Ropedia AI

Ropedia Super Week · 2026

We introduce DONIC, a generalist bimanual manipulation framework that learns from human demonstrations to coordinate two-handed interactions and generalize to unseen trajectories and held-out objects without fine-tuning.

04 ———S-Agent 2.0

S-Agent 2.0: A Spatial Reasoning Agent That Outperforms GPT-6

Ropedia AI

Ropedia Super Week · 2026

We introduce S-Agent 2.0, a training-free spatial reasoning agent that equips existing coding agents with perception tools and persistent scene memory. It leads across all five evaluated spatial reasoning benchmarks, with a stronger backbone surpassing GPT-6 on spatial reasoning tasks.

05 ———HoliWorld

HoliWorld: Scaling 3D/4D Scene Understanding from First-Person Video

Ropedia AI

Ropedia Super Week · 2026

We introduce HoliWorld, a framework that transforms first-person videos into structured 3D/4D scene understanding. Across 444 videos and ~100 scenes, it improves Token F1 on EgoDynamic4D from 18.29% to 66.39%.

06 ———EgoXperience

EgoXperience: Learning World-Grounded 3D Representations from Human Experience

Ropedia AI

Ropedia Super Week · 2026

We introduce EgoXperience, a framework that reconstructs cameras, scenes, hands, and objects from first-person video. With 6 million frames of human experience, EgoX reduces world-space hand reconstruction error on EgoX-Bench by 65.6% compared with the strongest evaluated baseline.

07 ———World Model

Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience

Ropedia AI

Ropedia Super Week · 2026

We introduce Ropedia W0, a world model that predicts future first-person videos across human and robot embodiments, powered by 10K hours of human experience and 3K hours of robot experience.

08 ———Dataset

Xperience-10M: A Large-Scale 4D Human Experience Dataset for Physical AI

Ropedia AI

arXiv · 2026

We introduce Xperience-10M, a large-scale dataset of 10 million human experience episodes captured in-the-wild.

09 ———Technical Report

HOMIE: A Full-Stack Platform for Frictionless Human-Centric Xperience Capture

Ropedia AI

Technical Report · 2025

We present HOMIE, a hardware-software platform for scalable, lossless human-centric Xperience capture.

10 ———Technical Report

SpatialBench: A New Benchmark

Ropedia AI

Technical Report · 2025

Introducing SpatialBench — Ropedia’s benchmark framework for Spatial Intelligence evaluation.

// TALK TO US

Building world models,
robots, or embodied agents?

If you are building world models, robotics systems, or embodied agents — and care deeply about learning from real-world human experience data — we'd love to talk.

Contact Us