Realtime AI News
Huawei Cloud releases Ascend-native embodied world model CloudRobo-WM1.2
Huawei Cloud used an embodied-AI forum to unveil CloudRobo-WM1.2, described as an Ascend-native embodied world model, alongside its one-stop CloudRobo development platform. The platform reports more than a million navigation trajectories, over 50,000 manipulation trajectories and more than 30 third-party models, aimed at sparing robotics teams from rebuilding infrastructure from scratch.
Huawei Cloud used an embodied-AI forum to present a stack that runs from Ascend compute up to real robots on the factory floor. At the centre of the announcement was CloudRobo-WM1.2, described as an Ascend-native embodied world model, alongside the one-stop CloudRobo development platform built around it.
The framing in the report is pragmatic. A robot that succeeds in a lab still has to repeat the same task daily in a plant, and swapping in a different robot can wipe out much of what was learned. Training a smarter model is not enough on its own; compute, simulation and real-machine adaptation all have to keep pace.
Compute came first. Qian Xinyuan, a product management expert on Huawei’s Ascend chips, cited roughly 1,400x growth in token consumption over two years, with agentic tasks consuming 100 to 1,000 times the tokens of ordinary question answering. Embodied workloads add long video and motion data, so memory bandwidth becomes the constraint; the planned 960DT is quoted at up to 9.6TB/s.
Huawei Cloud’s answer is to absorb hardware procurement, system adaptation and long-term operations into the cloud. Teams draw Ascend capacity when they need to train, and resize when they want more simulation runs, without buying a cluster sized for their peak week. The platform also uses long-sequence parallelism to balance memory load across modalities and prefetches the next batch of training data to cut idle time.
CloudRobo is the connective tissue. It unifies robot onboarding through the R2C protocol and SDK, and links data processing, training, simulation validation and real-machine deployment into a single chain. The platform reports more than one million navigation trajectories, over 50,000 manipulation trajectories and more than 50 data-processing operators.
Inside the same environment sit more than 30 third-party models, over 30 evaluation metrics and more than 20 embodied-development skills, with URDF import for describing robot structure. The practical payoff is diagnosis: when grasp success rates fall, engineers can adjust data and model separately inside the same evaluation harness and see which one is at fault.
Deployments span universities and industrial sites. Thirty students in Shanghai Jiao Tong University’s embodied-AI micro-major used CloudRobo to move from simulation to real hardware, compared models including Diffusion, ACT and Pi0.5, and took second place in the first CloudRobo embodied-AI competition; Huawei Cloud and Fujian Xiaozhuang Intelligent plan to apply the platform and Ascend capacity to university training. The report also says related robots have entered live high-voltage work sites, with more than 400 units delivered, over 80,000 operations a year, job times held under 30 minutes and roughly 30% efficiency gains.
The strategic logic is reuse: one pool of cards serving several companies in turn, one adaptation benefiting many teams, so compute does not have to be bought from scratch and engineering mistakes are not repeated. What remains to be seen is how much detail Huawei publishes on CloudRobo-WM1.2, and how cheaply the platform transfers across different robot bodies — the difference between a demo and shared infrastructure.
Why it matters
Cloud-hosted compute and adaptation lower the entry cost for embodied-AI teams and put Huawei Cloud inside the robot training loop. Cross-embodiment portability and published model evaluations will decide how far that position holds.
Nearby Updates
All11/20, 09:25
Korean exam test: GPT and Gemini score near-perfect, domestic K-EXAONE slips on inquiry section
On South Korea's national college entrance exam, GPT and Gemini turned in near-perfect answers while the country's top domestic model, K-EXAONE, lost ground on the inquiry section. The comparison puts the gap between frontier models and a sovereign Korean model onto a single, publicly legible scorecard.
11/20, 08:43
Anthropic's seven founders cashed out nearly $2B in record-breaking IPO
Anthropic's seven co-founders each took home about $280 million at the company's $83-per-share IPO price, for a combined payout of nearly $2 billion, TechCrunch reports. The offering was described as record-breaking, with the company raising close to $105 billion.
11/20, 08:06
UK regulators warn personal AI agent developers over uneven pre-launch engagement
Four UK regulators — the CMA, FCA, ICO and Ofcom — have sent a joint letter to ministers warning that personal AI agent developers are not engaging early or openly enough before launching products in the UK. The letter asks providers to address fraud, unexpected data sharing and misleading recommendations before launch, and says commercial partners such as financial services firms and telecoms operators retain their own obligations.
11/20, 10:34
Hainan releases its first filed tourism LLM as Yuduoduo agent lands in Rongshanliao
Hainan released its first filed large model for the culture and tourism sector, with an agent called Yuduoduo deployed in Rongshanliao on November 18, according to Guandian. The launch puts a locally filed model and agent into a concrete travel scenario rather than a generic chatbot.