刘松明 / Liu Songming

Overview

刘松明 (Liu Songming) is the founder of Liber AI, a physical-AI company combining world-action models, robot data, hardware integration, control, and deployment.

Current Profile

Born in 2000, Liu entered Tsinghua University through chemistry competition, moved from chemical engineering into computer science, and used AI-for-science work as a bridge between the fields. His robotics research included RDT-1B, whose pretraining results and open technical documentation helped create a research and industry network that later supported company formation.

Liu presents entrepreneurship rather than any single technical method as his long-term commitment. His operating view joins technical leadership with capital discipline: the lead founder must understand how compute, data, experiments, hardware cycles, and hiring compete for resources, while refusing to treat a large eventual market as proof of near-term product demand.

Key Characteristics

• Cross-disciplinary path from chemistry and chemical engineering into computer science, AI for science, and robotics.
• Early research emphasis on robot pretraining, diffusion-based continuous action modeling, scalable data, and reproducible technical disclosure.
• Founder-led technical resource allocation across compute, data, hardware, experiments, and product scenes.
• Selective full-stack thesis built around the expected expansion of general foundation models into robotics.
• Cautious commercialization posture that distinguishes technical progress from timing alignment among models, hardware, and market adoption.
• Organization-building style emphasizing taste, simplicity, debate, collaboration, and final founder accountability.

Evidence

• Education and research: 187: 对话LiberAI刘松铭：Astra来袭，00后物理AI创始人看到的胜率 traces Liu’s chemistry-competition admission, department transfer, AI-for-science entry, and RDT research.
• Company formation and strategy: 187: 对话LiberAI刘松铭：Astra来袭，00后物理AI创始人看到的胜率 records his decision threshold for starting a company, technical-founder argument, foundation-model forecast, and Liber AI’s architecture.
• Commercial and financial judgment: 187: 对话LiberAI刘松铭：Astra来袭，00后物理AI创始人看到的胜率 describes scenario criteria, the two-to-three-year target, fundraising reserves, scaling-efficiency comparisons, and three-to-five-year cycle preparation.
• Leadership and culture: 187: 对话LiberAI刘松铭：Astra来袭，00后物理AI创始人看到的胜率 attributes the company’s taste, restraint, simplicity, discussion, hiring, and final decision model to Liu.

Qualifications

The biographical and organizational account is self-reported through a founder interview. The episode title writes the name as “刘松铭,” while the body consistently writes “刘松明”; this page uses the body spelling and treats the title form as a likely editorial or transcription mismatch. Research, company-performance, financing, and hiring claims are not independently verified here.

What Changed

• Created Liu’s profile from the supplied founder interview.
• Connected his RDT research lineage to Liber AI’s data and pretraining strategy.
• Preserved the 刘松铭/刘松明 spelling conflict as explicit identity uncertainty.

Relationships

• Liber AI - company Liu founded and leads.
• RDT-1B - robot-pretraining research project that preceded the company.
• Tsinghua University / 清华大学 - institution where Liu moved from chemical engineering into computer science and robotics research.
• Physical Intelligence System Stack - system-level frame underlying his full-stack company thesis.
• World Action Models - current model route described for Liber AI.
• Kimi - prospective collaborator discussed in the interview.