Physical AI Enters the Experience Engineering Era as Ropedia Builds Real‑World Data Infrastructure

Physical AI is moving away from relying solely on expensive robot tele‑operation data toward pre‑training on massive real‑world experience, a trend highlighted by recent advances such as Qwen‑VLA’s unified training of robot operation, navigation and trajectory prediction, ACE‑Ego‑0’s conversion of first‑person video into pseudo‑action trajectories, and τ₀‑WM’s joint learning on over 27,000 hours of heterogeneous interaction data.

The angel round was led by top dollar funds and industry players in intelligent manufacturing and large‑model sectors; the Xperience‑10M dataset ranked second on the platform and first in the embodied track, and has been adopted by models such as Qwen‑VLA and MolmoMotion.

Ropedia, a Singapore‑based Physical AI company, announced a new financing round of tens of millions of dollars aimed at the next‑generation multimodal perception hardware, a real‑world data production engine, global collection and delivery capabilities, and a verification system for robot foundation models and world models.

Ropedia’s vision is a Real‑World Experience Engine that sits between the physical world and foundation models. The company structures this engine into three layers:

Sense : hardware that captures diverse interaction scenarios, including the lightweight head‑mounted first‑person device HOMIE and additional sensors for hand, tactile and spatial motion.

Experience Engine : an automated data‑production pipeline that transforms raw sensor streams into model‑ready training data through time‑synchronisation, spatial alignment, camera localisation, 3‑D/4‑D reconstruction, motion recovery, segmentation, semantic annotation and quality control.

Deliver : datasets, model‑training and validation suites that evaluate how different data affect VLA, world models and spatial‑intelligence models, feeding the results back to guide the next round of collection.

The heaviest engineering effort lies in the second layer, especially the annotation stage. Collected data are first uploaded to the cloud for automated pre‑screening, then pass through automated quality checks and manual final review before being stored with full traceability.

Ropedia’s flagship dataset, Xperience‑10M , released in March 2026, contains roughly ten million real‑world interaction clips, 10,000 hours of first‑person video and audio, over ten synchronized modalities, and approaches one petabyte of raw data. The dataset’s impact is measured by three hard signals:

Who uses the data : Xperience‑10M has been incorporated into the training pipelines of Qwen‑VLA, MolmoMotion, ACE‑Ego‑0 and τ₀‑WM. On Hugging Face it has exceeded 2.7 million downloads, ranking second on the platform‑wide dataset leaderboard with 1,569 distinct accounts.

Performance gain in ablation studies : ACE‑Ego‑0 used 435.7 hours of Xperience‑10M video; when combined with robot data, task success rose to 72.8 % versus 68.3 % with robot data alone—a 4.5 percentage‑point improvement attributed directly to the real‑world data.

Customer repurchase and iteration : By mid‑2026 Ropedia had served over 20 global robot and foundation‑model companies. Clients typically purchase an initial high‑precision multimodal dataset, evaluate model improvements, and then expand the data volume for subsequent quarters, indicating that the data are reusable for 2‑3 quarters of model development.

Ropedia plans to launch a next‑generation multimodal system in August 2026, aiming to fuse first‑person vision, hand motion, full‑body movement and spatial coordinates into a unified representation that captures both visual and contact information, essential for tasks such as insertion, screwing, pressing and handling deformable objects.

The company describes its approach as an “experience flywheel”: data collection, automated processing, model training, and failure analysis form a closed loop where each iteration refines the next, turning raw sensor signals into structured experience that continuously fuels model capability growth.

In the authors’ view, internet data set the ceiling for large‑language models, while real‑world experience will define the ceiling for Physical AI. Ropedia’s strategy of scaling human‑world experience into model‑usable form positions it as a foundational player in the emerging Physical AI ecosystem.