Press Enter to search · ESC to close

AI × Crypto

Huawei Cloud Launches Open-Source Embodied AI World Model, Denies Competing with Partners

Huawei Cloud has launched CloudRobo-WM1.2, an open-source embodied world model that predicts physical environments and generates robot actions. The company denies competing with partners, positioning itself as an enabler. The move could accelerate embodied AI adoption but faces technical and data challenges.

Huawei Cloud Unveils CloudRobo-WM1.2, an Open-Source Embodied World Model

Huawei Cloud has released its Ascend-native embodied world model, CloudRobo-WM1.2, and announced it will open-source the model. The launch, which took place at a CloudRobo event on November 16, marks a significant step in Huawei’s push into embodied intelligence—the fusion of AI with physical robots. The model is designed to understand physical environments, predict future changes, and directly generate robot actions. It ranked first in benchmarks such as RoboTwin 2.0 and LIBERO Pro 2.0.

The move raises questions about potential competition with the model companies that rely on Huawei’s computing platforms. Wu Xiaodong, president of Huawei Cloud’s Embodied Product Line and director of its Embodied Intelligence Lab, addressed this directly: “Will we become direct competitors? No.” He explained that Huawei Cloud’s investment in embodied models is not to compete for the model market but to share its training experience and engineering practices on Ascend Cloud, helping more enterprises train their own models.

Technical Approach and Challenges

Huawei Cloud’s model uses a joint modeling approach that combines perception, prediction, and action. Traditional robot models often require a conversion step between understanding the environment and executing actions. CloudRobo-WM1.2 aims to let the world model directly participate in action generation, enabling robots to not only predict what will happen next but also act accordingly. To address real-time response requirements, Huawei Cloud designed a dual-path action architecture: for complex tasks, the model can simultaneously predict future environments and actions for more thorough judgment; for simple, high-frequency operations, it can skip part of the prediction process and output actions directly.

This design addresses the computational cost challenge inherent in embodied AI. The more complete the world model’s prediction, the more computing resources are consumed, yet robots must respond promptly in real-world environments. Huawei Cloud offers four composable inference modes, allowing flexible adjustment of computation across different tasks. The company reports that computational demands vary by about 10 times between modes. Liu Liu, chief model expert at Huawei Cloud’s Embodied Intelligence Lab, revealed that the model’s inference latency is currently around 90 milliseconds, with a goal to reduce it to under 60 milliseconds.

Training data is another major hurdle. Unlike the massive text and image data available on the internet, real-world robot operation data is expensive to collect and difficult to transfer between different robot types. Huawei Cloud’s solution is to utilize data from diverse sources and learn motion patterns across different embodiments through a unified world model architecture. Liu Liu disclosed that CloudRobo-WM1.2 was pre-trained on about 8,000 hours of data and has now expanded to approximately 27,500 hours, incorporating 20 data sources and 33 robot embodiments. The model has developed 26 basic operation skills, covering tasks such as handling, assembly, material transfer, and flexible material manipulation.

Market Implications: AI and Robotics Ecosystem

For investors, Huawei Cloud’s move signals a deepening of the AI infrastructure race into physical robotics. By open-sourcing the model and providing computing power, data processing, and development tools, Huawei Cloud aims to lower the barrier for enterprises to develop embodied AI systems. This could accelerate adoption in industries such as manufacturing, logistics, and services. The company has already partnered with over 230 enterprises and universities, attracted more than 16,500 developers, and accumulated over 1,500 digital assets on its platform. It also announced support programs, including up to 1 million yuan in computing power for university projects and up to 2 million yuan for core partners.

While the immediate financial impact may be limited, the strategic implications are significant. Huawei’s push into embodied AI could bolster its cloud computing business and create a new ecosystem around its Ascend chips. For investors in AI and robotics, this underscores the growing convergence of cloud computing, AI models, and robotics—a trend that could reshape industries and create new investment opportunities. However, challenges remain: real-world factory environments demand high stability, continuous operation, and safety, far beyond lab conditions. Huawei Cloud must prove it can economically train embodied models suitable for industrial needs using Ascend computing power.

Key Takeaways for Investors

  • Open-source strategy: Huawei Cloud’s decision to open-source CloudRobo-WM1.2 and provide support programs could accelerate the embodied AI ecosystem, benefiting developers and enterprises while strengthening Huawei’s cloud platform.
  • No direct competition: Huawei Cloud explicitly denies competing with its partners, positioning itself as an enabler rather than a rival—this could ease concerns among model companies and foster collaboration.
  • Technical hurdles: Inference latency, data collection costs, and real-world reliability remain challenges. Success will depend on Huawei’s ability to deliver efficient, cost-effective solutions.
  • Ecosystem growth: With over 230 partners and 16,500 developers, Huawei Cloud is building a moat in embodied AI. Investors should watch for commercial deployments, such as the intelligent cleaning robot with Fulongma, slated for launch in December.
  • Long-term trend: The convergence of AI and robotics is a multi-year theme. Huawei’s entry validates the market’s potential but also intensifies competition in the AI infrastructure space.

View original

Share
Risk notice This site provides news and information on the crypto, blockchain and Web3 industry for reference only and does not constitute investment advice or any promise of returns. Virtual currency-related activities are illegal financial activities in mainland China; digital asset prices are highly volatile; use at your own risk. This site does not provide trading, token issuance or related referral services.

Related Reading

Latest News

TREE NEWS share card
Long-press image above → Save to Photos / Share
Pitch us Feedback