Gasgoo Munich- During the 2026 World Artificial Intelligence Conference (WAIC 2026), Noematrix showcased a complete technology ecosystem spanning real-world data, model training, and robotic applications under the theme "Enabling models to understand the world, bringing intelligence into reality."
Image Credit: Noematrix
A new generation of embodied intelligence pre-trained models took center stage at Noematrix's exhibition.
Unlike approaches that rely on robot teleoperation data for training, Noematrix's latest model used zero teleoperation data during both pre-training and post-training phases. This fully validates a viable path for pre-training embodied models using exclusively "in-the-wild" data collected without the robot's own body.
Specifically, the model adopts a world model architecture composed of a general embodied world model base, a specialized action prediction model, and associated post-training suites. Leveraging over 100,000 hours of home scenario data gathered across 47 cities nationwide in 2026, the model has completed pre-training. System evaluation and release preparations are currently underway, with a technical report slated for release between July and August.
At this year's WAIC, Noematrix also unveiled its self-developed RoboPocket body-less robot data collection system and DM3 data management platform. Together, they build a complete capability chain spanning real-world data collection, management, and data asset accumulation.
Image Credit: Noematrix
RoboPocket combines a wearable end-effector with a mobile collection app, allowing ordinary users to capture operational data without needing to interact with the robot itself. DM3 handles data visualization, retrieval, quality control, and asset management, providing continuous high-quality data support for model training.
Building on this real-world data foundation, Noematrix is advancing the development of a million-hour real-world operational data system, offering a richer data base for the continuous learning and iteration of its embodied large models.
However, even the most sophisticated models ultimately require validation in real-world scenarios.
At the conference, Noematrix demonstrated practical applications of embodied intelligence in two real-world settings: retail pharmacies and hotel laundry services. In the pharmacy scenario, Noematrix's solution covers more than 3,000 SKUs and requires only about 2.5 square meters for lightweight deployment—eliminating the need for extensive structural changes or operational overhauls. The system has already completed deployment and operational verification in chain pharmacies.
In the hotel laundry scenario, Noematrix utilized multi-robot collaboration to execute the entire washing, drying, and finishing workflow. A single-arm lifting mobile robot handled loading and unloading as well as equipment operation, while mobile chassis units managed autonomous navigation, item transport, and spatial coordination. Working under unified task planning, the robots demonstrated the practical application capabilities of embodied intelligence in live operational settings.
