Ropedia Introduces Xperience 2 As the Gold Standard to Scaling Physical AI Data Infrastructure

New dataset connects first-person human experience with 3D scene reconstruction, simulation, robot motion trajectories, and long-horizon task data

MOUNTAIN VIEW, CA, UNITED STATES, November 13, 2026 /EINPresswire.com/ — Ropedia, a Physical AI startup building data infrastructure for robotics and embodied AI, has introduced Xperience 2, a new dataset addressing a key robotics industry challenge: building the data infrastructure needed to develop, train, and evaluate increasingly capable robotic systems.

As robotics companies work towards systems capable of performing more complex tasks in real-world environments, they need more than larger volumes of training data. They need data connecting human actions with the physical environments in which they take place, alongside infrastructure that turns these experiences into reusable training and simulation resources. However, collecting data through physical robots remains resource-intensive.

Xperience 2 addresses this challenge by connecting first-person recordings of human activity with detailed 3D reconstructions of the environments in which those activities take place. The dataset transforms these experiences into reusable simulation environments, robot movements, and training trajectories without requiring physical robots at every stage of data production.

Traditional datasets often capture either human activities or physical environments, but rarely both together. Xperience 2 connects these dimensions by creating detailed 3D reconstructions of real-world environments and recording human activities within those same spaces using Ropedia’s HOMIE head-mounted device.

Data can be collected across diverse settings, from homes and restaurants to retail stores, offices, and outdoor environments. Ropedia’s processing pipeline aligns camera trajectories, hand and body movements, and other multimodal signals with the reconstructed 3D environment, capturing both what people do and where they do it. This provides a foundation for downstream simulation, training, and evaluation.

The current Xperience dataset comprises 1,812 reconstructed scenes from 596 collection locations, alongside 4,230 task recordings covering activities such as object manipulation, cleaning, and household tasks. These environments can be converted into reusable simulation scenes, allowing robots to practise across different layouts, object configurations, and starting conditions.

Building on these data assets, Xperience 2 supports four pathways for generating robot training experiences:
Real-to-Sim: Reconstructs individual objects from real-world scans, adding collision properties, joints, and physical parameters to make them interactive in simulation, with a demonstration producing 107 objects from a single home, 73 of which could be opened, pressed, or deformed.
Human-Robot Retargeting: Translates recorded human movements into robot joint trajectories, adapting demonstrations to different robotic configurations.
Virtual Teleoperation: Generates robot interaction demonstrations in simulated environments without physical robots or manual resets.
Long-Horizon Trajectory Generation: Produces extended sequences of robot actions, allowing systems to practise multi-step tasks across objects and rooms.

In one long-horizon trajectory generation demonstration, a simulated wheeled dual-arm robot completed eight consecutive subtasks in a reconstructed environment, including retrieving a cup, operating a microwave, and placing the cup in another room. All 30 executions across different starting configurations completed the tasks without intermediate resets. Each trajectory averaged approximately 2.2 minutes, with generation taking 10–16 minutes on a single simulation worker.

Once an environment and task are configured, additional trajectories can be generated using computing resources, without further human demonstrations or physical robot data collection. This allows existing data assets to support multiple training scenarios and simulation workflows.

Xperience 2 also provides infrastructure for more systematic robot development and evaluation. Researchers can test different models against consistent tasks and starting conditions, compare completion rates and failure patterns, and identify where additional training data is needed. They can also vary object positions, scene layouts, and robot configurations to examine performance under different conditions.

This creates a feedback loop between data production, robot execution, and evaluation. Insights from simulated failures can inform subsequent data collection and environment development, identifying which objects, actions, or scene conditions require further data. For robotics companies, this offers a more connected approach to developing, training, and testing their systems while making greater use of existing data assets.

“To scale Physical AI, we need infrastructure that makes every real-world experience more valuable by connecting it to the environments, actions, and outcomes that robots need to learn from. With Xperience 2, we are building towards a data infrastructure where human experience, 3D environments, simulation, and evaluation work together, allowing a single demonstration to support multiple training and development workflows,” said Zhaoxi Chen, Ropedia’s Co-Founder and CEO.

Ropedia continues to develop its Physical AI data infrastructure, connecting real-world human experience with the environments, training data, and evaluation capabilities needed to advance robotics development.

Aniket Gupta
A G Universal Branding APAC Pte. Ltd.
Visit us on social media:
LinkedIn

Media gallery