Sudo AI苏度科技

About

Sudo AI was founded in Shanghai in May 2025 by serial entrepreneur Han Zheng and Su Hao, Distinguished Haoqing Professor at Fudan University, with a dual-hub layout of 'Shanghai algorithms + Shenzhen manufacturing'. The company takes the pure simulation route (zero-shot Sim2Real), using no real-world collected data at all and training robots with an integrated design of a 3D world model and reinforcement learning: first-attempt grasp success of about 98%, near 100% within two attempts, and the Sudo R1 can grasp 100+ unseen objects nonstop for 60 minutes. In April 2026 it completed a $500M Pre-A round (also described elsewhere as Series A), with the valuation breaking $2B (~RMB 13.6B) and investors including CATL's Puquan Capital, Alibaba, Tencent, Ant Group, IDG Capital and Lanchi Partners. Commercially it co-develops with CATL around battery production, logistics delivery and other core manufacturing scenarios, and replicated a battery-module production line (four machines in coordination) live at WAIC 2026; capabilities have expanded to 10+ skills including bimanual manipulation, soft-object handling, precision assembly, mobile grasping and multi-robot collaboration, with customer acceptance criteria of 99%+; capabilities will be opened via SDK, API and virtual commissioning tools.

Technical Approach

Pure simulation training (zero-shot Sim2Real): uses no real-world collected data at all, achieving real-scenario deployment from simulation training alone. An integrated design of a 3D world model and reinforcement learning — the industry's only systematic implementation; 'simulation as the base, real machines as support'. In H2 2026 it will complete 'low-level skill models + upper-level developer tools', opening capabilities via SDK, API and virtual commissioning tools, with larger-scale deployment in 2027 driven jointly by channel partners, integrators and developers.

Commercialization（5）

• ◆Co-develops with CATL around battery production, logistics delivery and other scenarios, building the industry's first multi-station robot system, supporting cross-station generalization and rapid product switching; at WAIC 2026 replicated a battery-module production line — the first purely simulation-trained robot to complete complex tasks on a real production line
• ◆Delivery model: models can be initially deployed at zero-shot with high success rates, without collecting customer-sensitive data
• ◆Platform route: provides system interfaces and developer tools in a platform manner, uniformly packaged as standardized model capabilities and API interfaces
• ◆Customer focus: even with industrial shareholders and overseas industrial customers, it chooses partners very selectively, focused on 'extreme reliability' (99%+)
• ◆Goal: build a general robot brain, advancing embodied AI from 'walking intelligence' toward 'perception intelligence' and 'interaction intelligence'

Position & Strengths（5）

• ◆A $2B-valuation newcomer: valued at $2B (~RMB 13.6B) about 11 months after founding — a newly minted high-valuation embodied company of 2026, dispelling fundamental doubts about the pure-simulation (Sim2Real) route
• ◆The industry's only 'world model + reinforcement learning integrated design': the two are fused systematically in the underlying model
• ◆Pure-simulation zero-shot breakthrough: 98% first-attempt grasp success with 0 real-robot data, near 100% within two attempts, disrupting the traditional real-machine data collection path
• ◆Composite core team: serial entrepreneur (Han Zheng) + academic (Su Hao, one of ImageNet's builders) + investor (Chen Runze) + industry veteran (Zhang Xiaohang)
• ◆Top capital + industry binding: industrial capital from CATL, Alibaba, Tencent, Ant, Xiaomi and NIO + first-tier funds like Hillhouse, IDG and Yunfeng + long-term capital like China Life