OneRobotics Showcases at NVIDIA Startup Showcase Macau: OneModel 1.7 Drives Home Robots from Simulation to Reality

Recently, the 2026 NVIDIA Startup Showcase Macau was successfully held in Macau. The event focused on frontier areas including Physical AI, robotics, generative AI, open models, and AI applications, bringing together NVIDIA technical experts, startup representatives, and ecosystem partners to explore how AI technologies can be applied in real-world scenarios.

As a representative company in the field of embodied home robotics, OneRobotics (6600.HK) was invited to deliver a keynote presentation and showcase the application of its onero H1 home service robot in household scenarios. At the event, Huang Xiangchao, President of OneRobotics AI Innovation Research Institute, delivered a presentation titled “ From Isaac Lab to Your Living Room “ , systematically presenting OneRobotics’ latest practices in Physical AI, world action models, and home scenario deployment, as well as the company’s roadmap for embodied intelligence in the home.

During the keynote session, OneRobotics highlighted the core technical innovations of its recently released OneModel 1.7. The model adopts the RL-Latent World Action Model, or RL-LWAM, architecture, integrating world understanding, task planning, action execution, and real-world feedback learning into a unified closed loop. This creates a complete chain from instructions, observations, and skill inputs to world model reasoning, action policy generation, robot execution, reinforcement learning, and successful memory feedback.

Among these innovations, the Predictive Policy Latent mechanism is one of the key highlights of OneModel 1.7. Through implicit physical reasoning representations, this mechanism directly connects the world model’s understanding of the environment and tasks with action policy generation, reducing information loss between perception, understanding, and execution control. In simple terms, it enables the robot not only to predict “what will happen next,” but also to generate executable strategies for “what should be done next.”

Leveraging the full NVIDIA Isaac Sim and Isaac Lab simulation toolchain, OneRobotics has built an efficient and cost-effective system for model training and deployment. Before deploying robots on real hardware, companies can repeatedly train and validate robotic motion policies in simulation, significantly reducing trial-and-error costs in physical testing. With high-performance computing platforms for model training and fine-tuning, the system ultimately enables efficient on-device inference and precise control, forming a complete technical loop from simulation training and model iteration to edge deployment.

This technical pathway supports the continued evolution of OneRobotics’ “One Brain, Multiple Embodiments” product system. With OneModel as its core technical foundation, OneRobotics applies model capabilities across multiple robotic embodiments for home services, sports and health, and emotional companionship, including onero H1, Acemate, and Kata Friends, building an embodied intelligence ecosystem for household scenarios.

At the event, OneRobotics showcased an onero H1 home service demonstration focused on laundry scenarios. The demo covered clothing recognition, deformable object grasping, path planning, dual-arm coordination, and interaction with washing machine equipment, presenting the robot’s ability to continuously execute high-frequency household tasks in real home environments. For home service robots, clothing items are highly variable in shape, difficult to grasp, and uncertain in operation paths. This demonstration reflected onero H1’s ability to perform continuous tasks in unstructured home environments.

The demo further demonstrated the practical value of OneModel 1.7 in real household tasks. Through the model’s unified understanding of the environment, tasks, and actions, onero H1 can complete a task loop from perception to execution in specific household scenarios, helping move home service robots from single-action demonstrations toward continuous task execution.

In addition to the onero H1 home service demo, OneRobotics also shared its new explorations in home companion robots at the event, further demonstrating the scalability of its “One Brain, Multiple Embodiments” approach across different household interaction scenarios.

This appearance at the 2026 NVIDIA Startup Showcase Macau marks another important technical showcase for OneRobotics following the release of OneModel 1.7. From simulation labs to household scenarios, from world action models to on-device robotic execution, and from model capability validation to home task deployment, OneRobotics is continuing to advance embodied home robots from technological R&D toward real-world applications.

Looking ahead, OneRobotics will continue to use OneModel as its core foundation and “One Brain, Multiple Embodiments” as its technical pathway, focusing on real household scenarios to advance model iteration, data feedback loops, and product deployment, accelerating the entry of embodied robots into everyday life.