Riemann Dynamics, in partnership with LimX Dynamics and Noitom, advances the construction of one million hours of embodied data, aiming for 2026 to build a closed loop of data, models, and deployment.
On August 6, Riemann Dynamics announced a strategic partnership with Lightwheel AI and Noitom Robotics. Centered on the Riemann-1.0 embodied world action model and the Matrix-Game 3.5 interactive world model, Riemann Dynamics will work with both companies on deep collaboration in high-quality embodied data collection, model training, large-scale evaluation, real-robot deployment, and feedback optimization, advancing the construction of one million hours of embodied intelligence data targeting 2026.
This partnership marks Riemann Dynamics' transition from breakthroughs in individual model capabilities toward a systematic closed-loop framework encompassing "data production, model training, real-world validation, and feedback iteration." Going forward, Riemann Dynamics will leverage the real-task performance of its Riemann series of embodied brain models to continuously identify data gaps, and accelerate the iterative evolution of general-purpose embodied intelligence models through multimodal data platforms, evaluation systems, and robot deployment feedback.
Building a One-Million-Hour Embodied Data Closed Loop Starting from Model Capabilities
For embodied intelligence models to truly understand and operate in the physical world, relying solely on visual information is far from sufficient. Robots need not only to "see" objects, but also to understand the changes in grip force when grasping, human joint movement states, continuous trajectories of actions in three-dimensional space, and the dynamic interaction relationships between humans, objects, and the environment. These high-precision multimodal data—covering human motion, force feedback, environmental changes, and interactive object states—form a critical foundation for building the next-generation universal brain for robots.
To this end, Riemann Dynamics plans to complete the collection and training of one million hours of embodied data by the end of 2026, continuously building embodied intelligence infrastructure spanning data, evaluation, model iteration, and real-robot deployment.
In the collaboration with Lightwheel AI, both parties will drive the adaptation and validation of the Riemann-1.0 embodied world action model and the Matrix-Game 3.5 interactive world model with Lightwheel AI's EgoSuite human data platform, RoboFinals evaluation platform, and RoboStack deployment feedback platform.
Image: Li Yangguang, co-founder of Riemann Dynamics, and Lian He, Vice President of Strategy and Ecosystem at Lightwheel AI, sign the strategic cooperation agreementBased on Riemann-1.0's capabilities demonstrated in mobile manipulation, object interaction, household services, and long-horizon tasks, Riemann Dynamics will precisely define the data requirements and capability gaps needed for model training. Lightwheel AI's EgoSuite platform will organize large-scale, high-quality human behavior video and multimodal data production around relevant tasks, feeding the data back into the model training pipeline. This means that each subsequent iteration of the Riemann series models will more directly target real capability weaknesses rather than relying solely on data scale accumulation.
Integrating Motion Pose and Pressure Tactile Data to Enhance Complex Manipulation Capabilities
In the collaboration with Noitom Robotics, both parties will focus on synergies in motion capture, human-robot interaction, multimodal behavior data collection, and data engineering capabilities, jointly building a high-quality, multimodal embodied data system.
Noitom Robotics has long-standing technical expertise in human motion capture, force feedback collection, and multimodal behavior data acquisition. This partnership will provide critical data support for improving Riemann-1.0's capabilities in long-tail tasks, complex manipulation, fine-grained force control, and cross-embodiment generalization. Going forward, the two parties will jointly collect training data encompassing human motion trajectories, joint movement states, contact force feedback, and interactive object states, with a focus on solving the challenge of high-precision temporal and spatial synchronization of multimodal data.
Image: Li Yangguang, co-founder of Riemann Dynamics, and Lian He, Vice President of Strategy and Ecosystem at Lightwheel AI, sign the strategic cooperation agreementRiemann Dynamics will combine model training results with real-robot validation outcomes to dynamically optimize data collection, processing workflows, and quality assessment systems, ensuring that data continuously serves model capability improvement while model performance in real-world tasks drives the evolution of the data system in return.
Parallel Advancement of Riemann-1.0 and Matrix-Game 3.5
Riemann-1.0 is the embodied world action model Riemann Dynamics has developed for Physical AI. By learning from large-scale human video data, the model builds an understanding of human behavior, environmental state changes, and action outcomes, and has achieved leading results across multiple international robotics benchmarks and real-robot tasks. In the RoboCasa-365 benchmark—widely recognized within the robotics community as an extremely difficult household challenge—Riemann-1.0 secured first place with an average success rate of 62.6%, surpassing the previous industry-leading performance by 8.4 percentage points.
Matrix-Game 3.5, on the other hand, is oriented toward interactive world models, dedicated to building a next-generation world model system with long-term memory, continuous interaction, and open-world simulation capabilities. Through collaborations with ecosystem partners such as Lightwheel AI and Noitom Robotics, Riemann Dynamics will further refine the closed loop spanning model evaluation, data construction, and real-world deployment, accelerating the application and validation of world models in embodied intelligence scenarios.
Li Yangguang, co-founder of Riemann Dynamics, stated: "For robot manipulation models to truly enter complex, open real-world environments, they need continuous access to high-quality, diverse, and verifiable data. Through our partnerships with Lightwheel AI and Noitom Robotics, we aim to let model capability requirements directly shape the data system, and continuously drive model iteration through real deployment feedback."
As the one-million-hour embodied data initiative advances, Riemann Dynamics will continue to make systematic investments across world models, embodiment design, motion control, evaluation systems, and real-world scenario data closed loops, driving robots from perception and understanding toward long-term, stable, and generalizable real-world interaction.
About Riemann Dynamics
Riemann Dynamics (full name: Beijing Riemann Dynamics Robotics Technology Co., Ltd.) focuses on embodied intelligence and Physical AI, dedicated to building the next-generation universal brain for robots in the real world. Based on the Matrix-Game world model technology framework, we build a unified intelligence architecture that integrates world understanding, future prediction, and action decision-making, enabling robots to continuously understand their environment, plan autonomously, and complete complex manipulations.
Centered on core scenarios such as household services and mobile manipulation, we have accumulated hundreds of thousands of hours of high-quality human and robot data, continuously advancing the integration of world models, embodiment design, motion control, and real-world scenario data closed loops—driving robots from perception and understanding toward long-term, stable, and generalizable real-world interaction.