BeingBeyond and D-Robotics Enter Strategic Partnership

On June 11, 2026, BeingBeyond and D-Robotics officially announced a strategic partnership. By combining BeingBeyond's expertise in deploying latent world action model  with D-Robotics' high-performance on-device AI computing, the two companies have completed the world's first on-device deployment of a world action model. In testing, the system achieved inference speed of nearly 20 FPS, marking an important step toward the large-scale deployment of world models.

Over the past year, the industry has gradually shifted from VLA models toward world action models. Yet as model capabilities continue to improve in theory, a more practical question has come into focus: can world models break free from their dependence on data-center-class GPUs, move beyond the laboratory, and run directly on the edge chips embedded in robots? Only by overcoming this bottleneck can world models become viable for deployment at scale.

In April 2026, BeingBeyond released Being-H0.7, introducing a latent-space world model paradigm for embodied intelligence and scaling human-video pretraining to 200,000 hours. Around Being-H0.7, BeingBeyond built a complete post-training and inference infrastructure for on-device deployment, including several proprietary inference-optimization technologies such as Universal Async Chunking (UAC).

Building on Being-H0.7 and D-Robotics' S600 and S100 platforms, BeingBeyond and D-Robotics carried out deep chip-model co-optimization. Test results show that BeingBeyond's Being-H-Flash model product runs at close to 20 FPS on the S-series computing platform, with smooth and stable performance and extremely low response latency, reaching an industry-leading level of real-time interaction.

Real-time on-device operation means a shorter control loop, less dependence on network communication, and more stable system response. It also enables world models to participate directly in real-time decision-making for complex tasks such as dynamic catching, conveyor-line sorting, deformable-object manipulation, and liquid pouring, becoming an integral part of the robot's online control system.

In warehousing and logistics, robots can handle fast-changing parcels more reliably. On industrial production lines, they can adjust their actions in real time based on object trajectories and contact relationships. In retail, home, and service-robot scenarios, they can better handle deformable objects, liquids, tool use, and multi-stage tasks.

The successful adaptation further deepens the technical collaboration and joint scenario development between BeingBeyond and D-Robotics. It also lays a solid foundation for the two companies to jointly develop solutions for diverse application scenarios and accelerate the large-scale adoption of embodied intelligence across multiple industries.

About BeingBeyond

BeingBeyond focuses on developing and applying general-purpose embodied foundation models. It was among the first companies in China to propose a general-purpose model framework trained on large-scale human video data, and the first AI startup in China to introduce a natively latent-space world model. With a mission to bring humanoid robots out of the laboratory and into everyday life, BeingBeyond is committed to solving the core technical challenges of embodied AI and leading the transformation of humanoid robotics.