速度科技 / Sudu Technology

Overview

速度科技 is the robotics company represented by Han Zheng / 韩正 in 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. The wiki now tracks it as a full-stack manipulation and simulation route inside embodied AI.

Current Profile

Sudu’s route emphasizes Sim2Real, physically useful simulation, hardware/software co-design, and stable low-level manipulation skills. The new four-company debate sharpens the earlier E244 position: simulation is not a rejection of real data, but a way to scale pretraining for grasping, placing, assembly, and other short skills before they are composed under Layered Robot Architecture.

Key Characteristics

• Builds around low-level manipulation reliability rather than only high-level robot reasoning.
• Treats simulation as a scalable training and reinforcement-learning layer for basic physical skills.
• Still accepts real-world data as useful for grounding, validation, and physical transfer.
• Uses Layered Robot Architecture to connect upper-level understanding with lower-level execution.
• Defines commercialization through high success rate, generalized handling of objects and environments, and low deployment adaptation cost.

Evidence

• Full-stack evidence: E244｜端到端vs上下分层：机器人路径之争，正在转向？ describes Sudu as building robot body and robot brain together because hardware configuration and manipulation policy affect transfer.
• Simulation evidence: E244｜端到端vs上下分层：机器人路径之争，正在转向？ links Sudu to Structured 3D Robot Data, GPU-parallel simulation, hardware-specific noise modeling, and open-world manipulation tests.
• New debate 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 Han Zheng saying Sudu uses simulation while still treating real-world data as useful.
• 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 Han emphasizing near-perfect success, fast correction, object/environment generalization, and low post-training cost.

Qualifications

Sudu’s strongest evidence in the wiki remains source-reported demonstrations and operator claims. The new source records confidence in near-term stable skills, but does not independently benchmark Sudu against other robot companies or provide deployment-scale customer data.

What Changed

• Converted the page to the synthesis-v1 entity schema.
• Added the Shizilukou Crossing debate as evidence that Sudu’s simulation route is stage-specific, not simulation-only.
• Added commercialization requirements around high success rate and low deployment adaptation cost.

Relationships

• Han Zheng / 韩正 - Sudu speaker and route explainer.
• Sim2Real - core training and transfer method in Sudu’s route.
• Layered Robot Architecture - architecture Sudu uses to connect high-level planning with short skills.
• Embodied Robot Data Tradeoff - new data tradeoff sharpened by the four-company debate.
• Robot Deployment Success Economics - commercial test the new source applies to Sudu’s route.