Han Zheng / 韩正

Overview

Han Zheng is the 速度科技 / Sudu Technology speaker in the wiki’s embodied-AI branch. Across E244｜端到端vs上下分层：机器人路径之争，正在转向？ and 当具身智能走到十字路口｜对谈苏度、蚂蚁灵波、自变量、破壳：四种一线判断, he argues that robotics progress depends on reliable manipulation, useful simulation, and layered skill execution rather than humanlike appearance or polished demos.

Current Profile

Han’s position is that open-world manipulation needs Structured 3D Robot Data, Sim2Real, hardware/software co-design, and short skills that generalize. The later four-company debate adds a commercial layer: high success rate and low post-training cost are the threshold between research demos and deployable robots.

Key Characteristics

• Argues that scarce real robot data makes simulation and structured physical data necessary for open-world manipulation.
• Supports Layered Robot Architecture where upper-level models plan and lower-level skills execute reliably.
• Treats real-world data as useful but not sufficient by itself.
• Emphasizes business success requirements: high first-pass success, fast correction, object/environment generalization, and low adaptation cost.
• Is relatively optimistic that stable skills across major task classes can improve over the next one to two years.

Evidence

• Route evidence: E244｜端到端vs上下分层：机器人路径之争，正在转向？ attributes the structured 3D, Sim2Real, and layered manipulation route to Han.
• Fleet-data evidence: E244｜端到端vs上下分层：机器人路径之争，正在转向？ records his argument that robotics lacks a Tesla-like deployed fleet data flywheel.
• Data-tradeoff evidence: 当具身智能走到十字路口｜对谈苏度、蚂蚁灵波、自变量、破壳：四种一线判断 records him saying Sudu believes in simulation while also valuing real-world data.
• Commercial evidence: 当具身智能走到十字路口｜对谈苏度、蚂蚁灵波、自变量、破壳：四种一线判断 records his distinction between research and commercial robots through near-perfect success and generalization needs.

Qualifications

Han’s claims are operator judgments grounded in Sudu’s route and source-reported demos. The wiki should keep them alongside competing real-data and general-model views rather than treating simulation-led manipulation as a settled consensus.

What Changed

• Converted the page to the synthesis-v1 entity schema.
• Added the 2026 four-company debate as a second source for Han’s simulation, layering, and commercialization stance.

Relationships

• 速度科技 / Sudu Technology - company context.
• Sim2Real - core route Han emphasizes.
• Layered Robot Architecture - planning/execution architecture he supports.
• Embodied Robot Data Tradeoff - current debate where his simulation confidence is qualified by real-data arguments.
• Robot Deployment Success Economics - business standard added by the new source.