AI News

Huawei Cloud Releases and Open-Sources the Ascend-Native Embodied World Model CloudRobo-WM1.2

Sina TechUpdated Nov 16 21:05

AI summary

Huawei Cloud released and open-sourced the Ascend-native embodied world model CloudRobo-WM1.2 at the "Industry AI Dream Factory," and the model ranks first on both the RoboTwin 2.0 and LiberoPro 2.0 evaluation benchmarks. The model adopts a shared world model backbone network and a native DualAct architecture, and proposes five innovations including the PPA multimodal world representation architecture, corresponding to the issues of multimodal physical understanding, cross-embodiment generalization, and real-time inference compute cost in embodied intelligence. Huawei Cloud has also opened its training framework and WAM Ascend compute best practices.

Why read this

AI score 62

The original text presents the model architecture innovations and rankings on two benchmarks, allowing readers to understand the technical approach of the Ascend-native embodied world model.

Topics

Tags

Body · AI translation

Related topics top start Related topics top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

Sina Tech News, November 16 afternoon: Huawei Cloud's "Industry AI Dream Factory" released the Ascend-native embodied world model CloudRobo-WM1.2 and announced that it is open source. The model uses a shared world model backbone network to achieve prediction-enhanced world representation, and is based on the world model's native DualAct architecture to achieve world-model-native action generation. It ranks first on both the RoboTwin 2.0 and LiberoPro 2.0 evaluation benchmarks.

According to the introduction, CloudRobo-WM1.2 proposes five innovations, including the PPA multimodal world representation architecture, the world model's native DualAct architecture, a multi-granularity supervised data adaptive pretraining framework, ontology-semantic-preserving cross-embodiment motion modeling, and four elastically composable inference modes—World/Expert Action and Joint/Action-only—that can be combined as needed. This set of technologies addresses the problems often encountered when embodied intelligence moves out of the laboratory: difficulty in multimodal physical understanding, difficulty in cross-embodiment generalization, and high compute costs for real-time inference.

In addition to the open-source model, Huawei Cloud is also opening up the training framework, WAM Ascend computing best practices, and more. Huawei Cloud said it hopes to open source and open up the best practices of embodied world models on Ascend Cloud to industry partners, model vendors, embodiment vendors, and end customers, helping them develop models and applications more efficiently on Ascend Cloud.

At the same time, Huawei Cloud emphasized that it will firmly invest in core technology research and development for embodied intelligence, continue to open up models, toolchains, and technical practices, and work with embodiment vendors, model vendors, and industry customers to jointly build solutions, striving to be the cloud that best understands embodied intelligence, is the most advanced, and is the most open, and to promote the large-scale implementation of embodied intelligence across thousands of industries. news_keyword_pub,stock, iframe below the main text iframe below the main text General statement General statement QR code at the end start QR code at the end start

QR code at the end start QR code at the end start Editor name and work code

Editor name and work code

Source

Sina Tech