Huawei Cloud launches Ascend-native embodied world model CloudRobo-WM1.2 for sim-to-real
量子位 · wechat · 2026-11-19
Huawei Cloud released Ascend-native embodied world model CloudRobo-WM1.2 on Nov 16 (open-sourcing Nov 30), alongside a full embodied-AI engineering stack.
Compute base: Huawei says token consumption grew 1400x in two years, with agent tasks consuming 100–1000x more tokens than plain Q&A. Ascend 960DT targets up to 9.6TB/s memory bandwidth, with long-sequence parallelism easing multi-GPU memory imbalance.
CloudRobo platform: R2C protocol/SDK unify robot data ingestion — 1M+ navigation trajectories, 50k+ manipulation trajectories, 50+ data-processing operators; 30+ third-party models, 30+ eval metrics, 20+ embodied skills; URDF import for sim pre-validation; cloud planning with edge multi-robot scheduling and on-robot real-time control.
Ecosystem: Shanghai Jiao Tong students ran sim-to-real pipelines with Diffusion, ACT and Pi0.5; Yijiahe's Z100 robots (400+ delivered, 80k+ operations/year) get a Z200Pro upgrade targeting <30min jobs and 30% efficiency gains. The zone has partnered with 230+ companies/universities in 5 months, amassing 1500+ technical assets, with up to ¥2M per embodied-AI partner.
Related event: Huawei Cloud Open-Sources CloudRobo-WM1.2, Tops Robot Benchmarks(2 posts)→
More from Embodied
- Inhabitant: training-free LLM agent completes 74/75 long-horizon household tasks — Yao Zhang · 2026-11-20
- Scoble lists 10 ways Tesla FSD fails at airport pickups ahead of wheel-less Cybercab sales — Scobleizer · 2026-11-20
- EPFL team unveils unified 3D generative model for structure-based drug design and docking — HannesStaerk · 2026-11-20
- GXO's 1,000-robot Apptronik deal is a letter of intent, hinging on 4 years of KPIs — YvesMulkers · 2026-11-20
- NVIDIA's SOMA Retargeter adapts human motion to five humanoid robots, from G1 to AgiBot — jonstephens85 · 2026-11-20
- AFFECT from KAIST makes world models dynamics-faithful via action effect matching — John Won · 2026-11-20