type
Post
status
Published
date
Nov 20, 2026
slug
summary
tags
category
技术分享
icon
password

CloudRobo Launch: Huawei Cloud Builds End-to-End Ecosystem for Embodied AI

The transition of robots from laboratory demonstrations to stable, daily industrial operations represents a significant engineering challenge. On November 16, Huawei Cloud announced the release of its Ascend-native embodied world model, CloudRobo-WM1.2. Two days later, at a high-level forum in Shanghai titled "Industry AI Dream Factory: Embodied AI," the company revealed a comprehensive infrastructure strategy designed to address the complex demands of embodied intelligence.
The core problem is clear: while models can be trained effectively in simulation, deploying them on physical robots requires robust computing power, precise simulation-to-reality transfer, and seamless adaptation to different hardware. Huawei Cloud's approach is to centralize the underlying computational, storage, and network layers, while providing a unified platform for development.

Scaling Compute for Agentic Workflows

Modern AI workloads are evolving. While standard question-and-answer tasks are relatively simple, Agent-based tasks require multi-step reasoning, tool usage, and iterative refinement. According to data presented by Huawei's chip management experts, Token consumption for Agent tasks is 100 to 1000 times higher than for ordinary queries.
For embodied AI, this demand is even more intense. Training a robot to perform a simple task, such as gripping a cup, requires massive data processing of video, joint positions, and motion records. Simulation results must be calculated before the training program can adjust the model, meaning that idle time for computing resources is costly.
Huawei Cloud addresses this by offering a flexible cloud-based compute model. Instead of companies buying fixed hardware clusters that sit idle during slow periods, they can scale Ascend compute resources on-demand. This is critical for embodied AI, where a single card batch can serve multiple companies sequentially. The infrastructure also includes specific optimizations for long-sequence parallel processing, balancing memory loads across different data modalities to prevent bottlenecks.

Bridging Simulation and Reality with CloudRobo

The CloudRobo platform serves as the bridge between data, training, and physical deployment. It unifies the pipeline through the R2C protocol and SDKs, supporting PB-scale multimodal data management. This allows developers to handle data from various collection devices without rebuilding the entire workflow for each new robot body.
The platform has already accumulated over 1 million navigation trajectories and 50,000+ operation trajectories, along with more than 50 data processing operators. To further assist development, CloudRobo integrates over 30 third-party models and provides specialized evaluation metrics.
A key feature is the support for URDF files, which allows teams to simulate robot structures in a virtual environment before physical testing. This reduces the risk of physical wear and safety hazards during the trial-and-error phase. Furthermore, the CloudRobo Operations Management Platform connects cloud models with on-site robots. While long-term planning and training happen in the cloud, real-time control and safety mechanisms remain on the edge, ensuring that anomalies are handled locally while data is logged for later analysis.

Industry Partnerships and Ecosystem Growth

To validate these capabilities, Huawei Cloud has collaborated with various industry partners and academic institutions. For instance, in a partnership with Fujian Xiangzhuang Intelligence, CloudRobo is being integrated into university training, allowing students to work with real robots. At Shanghai Jiao Tong University, 30 students used the platform to move from simulation to physical deployment, comparing different models and winning second place in the inaugural CloudRobo Embodied Intelligence Competition.
In industrial applications, partner Yijiahe is deploying robots for high-altitude live-line work. Their Z200Pro robot, developed in collaboration with Huawei Cloud, aims to reduce reliance on manual remote operation. Preliminary data suggests the new system can complete tasks in under 30 minutes, with an efficiency improvement of approximately 30%.
Huawei Cloud is also expanding its ecosystem through the "Industry AI Dream Factory." The platform provides specific financial support, with up to 2 million RMB for single partners in the embodied intelligence track. As of the forum, the initiative had engaged over 230 enterprises and universities, accumulating more than 1,500 technical assets.
By integrating compute infrastructure, development tools, and industry-specific solutions, Huawei Cloud is aiming to create a reusable framework. This allows new teams to avoid the repetitive engineering pitfalls of the past, leveraging existing adaptations and data pipelines to accelerate the deployment of reliable embodied AI solutions.
Source: QbitAI, https://www.qbitai.com/2026/11/511426.html
量子计算机如何打破经典逻辑悖论:谷歌最新实验揭示量子关联性的深层力量开源项目 Agnes Harness:通过双系统决策与自进化机制,将 Agent 执行成本降至 1/27
Loading...