Gasgoo Munich- Gasgoo Embodied Intelligence has learned that Lightwheel AI has recently struck a strategic partnership with Riemann Dynamics. The two plan to integrate Lightwheel AI's EgoSuite human data platform, RoboFinals evaluation platform, and RoboStack deployment feedback platform with Riemann Dynamics' Riemann-1.0 embodied world action model and Matrix-Game 3.5 interactive world model. Together, they aim to build a closed-loop ecosystem spanning real-world data, world modeling, capability evaluation, task deployment, and data reproduction.
Riemann Dynamics specializes in embodied intelligence and physical AI, working to build a next-generation general-purpose brain for robots designed for the real world.
Image Source: Lightwheel AI
Specifically, Lightwheel AI and Riemann Dynamics will deepen their collaboration across three key areas:
First, the companies will integrate Lightwheel AI's EgoSuite with Riemann Dynamics' Riemann-1.0. Riemann Dynamics will identify data gaps based on the model's performance in mobile manipulation, object interaction, home services, and long-horizon tasks. In response, Lightwheel AI will produce scaled, high-quality human behavior videos and multimodal data tailored to those specific tasks.
Simultaneously, they will align on task definition, scene description, behavioral trajectories, environmental states, time synchronization, and quality standards. This ensures data structuring captures environmental conditions, interaction objects, behavioral sequences, and state changes. By shifting data production from scale-based supply to a model driven by capability gaps, the partnership aims to accelerate model iteration and evolution.
Second, the pair will explore integrating Riemann Dynamics' Riemann-1.0 and Matrix-Game 3.5 into Lightwheel AI's RoboFinals large-scale evaluation platform. The focus will be on assessing Riemann-1.0's action prediction, long-horizon task execution, and strategic stability under disturbance conditions. For Matrix-Game 3.5, they will evaluate long-term scene consistency, interaction response, and world state evolution.
Lightwheel AI will construct a scalable evaluation environment covering diverse objects, environments, and task variations, linking results back to specific model versions and training datasets. These evaluations will pinpoint model capability boundaries and generate data requirements for the next round. By retesting updated models under identical conditions, the companies aim to establish a measurable, traceable cycle of continuous improvement.
Third, the companies will align Lightwheel AI's RoboStack with the task interfaces of Riemann Dynamics' Riemann-1.0 and Matrix-Game 3.5. They will adapt task definitions, action spaces, status feedback, and execution workflows, enabling action strategies generated by world models to map directly onto robot execution frameworks.
During this process, Lightwheel AI will provide platform support ranging from task modeling to deployment verification, while Riemann Dynamics will leverage its model's capabilities in spatial understanding, physical reasoning, and action planning. Through RoboStack, the two will explore pathways for migrating world models from simulation to real-world tasks, closing the loop between model prediction, action generation, physical execution, and feedback.
